‘91".

1
gr».

:5. in
1.552. .
is, .

 

s...
‘ J?
was .35
19“.
«F? a. u.
3.,

  

 

‘ .
n:
.= :15
.3: u. .2! a
15...}... . . Z...

. ,5
3...: ..
5.1.

 

5.13:1:
‘ I .v.\\uu1
v (v

 

 

.u. 11:! I...

EWEE. V . 7 . . . . . . I . .. a
l ‘ i. 1.. A . , . H . . , an. .
n r '

1.2. 2.

 

 
 

‘ luau“; u .rbﬁeﬁ:\«o\ .11. , magnum WY. .1331 oﬁ.‘
U . , . u . 1‘ ll Init: C ‘ 3 I

| 1 - ‘

I II! D

llBRARY
Michigan State
University

This is to certify that the

dissertation entitled

"The Stock Market Returns to Disaggregated
Intangible Investments"

presented by

Stephen B. Marasco

has been accepted towards fulﬁllment
of the requirements for

Ph . D . degree in Economics

 

wb‘gﬂgﬁh

Major professor

Date 0 2 a

MS U is an Afﬁrmative Action/Equal Opportunity Institution 0-12771

 

 

PLACE IN RETURN BOX to remove this checkout from your record.
To AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 c-JCIRC/DateDuepss-p. 15

THE STOCK MARKET RETURNS TO DISAGGREGATED INTANGIBLE
INVESTMENTS

By

Stephen B. Marasco

A DISSERTATION
Submitted to
Michigan State University
in partial fulﬁllment of the requirements
for the degree of
DOCTOR OF PHILOSOPHY
Department of Economics

2001

ABSTRACT

THE STOCK MARKET RETURNS TO DISAGGREGATED INTANGIBLE
INVESTMENTS

By

Stephen B. Marasco

Innovative activity, particularly research and development, have public good
characteristics that prevent ﬁrms from appropriating the full returns of their efforts, and
thus ﬁrms under-invest in innovative activity. This argument has been used as a basis for
implementing a number of government policies designed to stimulate investment in
intangible assets. However, one could argue that these policies have been limited in their
effectiveness because many aspects of the process of innovation remain poorly
understood.

This paper seeks to determine the relative proﬁtability to ﬁrms of tangible and
intangible assets, and the effect of ﬁrm size on the values of each. Intangible assets
consist of knowledge stock, represented by research and development expenditures, and
marketing stock, represented by advertising expenditures. The value of these intangible
assets are determined using a stock market valuation approach. In addition to aggregate
R&D, I examine the value of various types of R&D, including product, process, applied,
and basic research, and test for potential interactions between product and process R&D.

I ﬁnd that intangible assets are more highly valued than tangible assets, but the
value of different types of intangible assets varies widely. In general, the market places

more value on R&D types lying closer to the output end of the innovation process, such

as applied R&D and product R&D. Investments associated with the beginning of the
innovation process, such as basic research, are not highly valued by the market. In
addition, I ﬁnd that the value of product R&D and process R&D are complementary. In
other words, the effect of apprOpriability spillovers is stronger than any impact from
potential organizational design trade-offs. Finally, I ﬁnd support for an “appropriability
effect” in which the value of process R&D increases more rapidly with ﬁrm size than the

value of product R&D.

C0pyright by
STEPHEN B. MARASCO

2001

To my parents,
Peter J. Marasco
And
Helen Kimball,
And to my wife,

Arlene

ACKNOWLEDGMENTS

The completion of my dissertation would not have been possible without the
assistance of my dissertation committee. The Chair of my committee, Professor Kenneth
Boyer, provided me with invaluable advice, encouragement, and patience from the
beginning of the process to the end. His insights and attention to detail greatly improved
the quality of my dissertation. Professor Bruce Allen offered helpful comments
throughout many stages of the dissertation, and Professor Lisa George’s insights into the
empirical analysis were extremely helpful. The insightful and intuitive comments of
Professor Roger Calantone added depth to the ﬁnal narration of my dissertation.

Numerous people helped me at different stages of my Ph.D. program. My
gratitude goes to the professors in the Economics Department for their guidance and
instruction, and to my fellow Ph.D. students for their ﬁ'iendship and support. I would
particularly like to thank Industrial Research Institute/Center for Innovation Management
Studies for the use of their unique data. I also would like to thank my co-workers at the
House Fiscal Agency of the State of Michigan and Household International for their
support and encouragement, and for providing me the opportunity to use my formal
education in a professional setting, which in turn greatly added to the quality of my
dissertation and afforded me the resources to see my work through to its completion.

Finally, I would like to thank my family for their support. My deepest gratitude
goes to my wife, Arlene Castellanos, for her faith, encouragement, and patience. This

dissertation would not have been possible without her.

vi

TABLE OF CONTENTS

 

 

 

 

 

 

LIST OF TABLES IX
LIST OF FIGURES X
INTRODUCTION 1
1. REVIEW OF THE LITERATURE 8
A. PRODUCT, PROCESS, AND OTHER TYPES OF INNOVATION ................................................................. 8
B. PRODUCT AND PROCESS INNOVATION WITHIN AN ORGANIZATION .................................................. 16
C. RETURNS To INNOVATION ............................................................................................................... 23
I. Determinants .............................................................................................................................. 23
2. Measurement .............................................................................................................................. 29
II. THEORETICAL FRAMEWORK AND MODEL 39
A. INTERACTIONS BETWEEN PRODUCT AND PROCESS INNOVATION ..................................................... 40
B. THE MODEL ..................................................................................................................................... 44
C. SUMMATION OF HYPOTHESES/PROPOSITIONS .................................................................................. 66
III. EMPIRICAL FORMULATION AND RESULTS 67
A. DATA ............................................................................................................................................... 67
B. ESTIMATION AND RESULTS .............................................................................................................. 75
1. Empirical Framework ................................................................................................................ 75
2. Preliminary Analysis .................................................................................................................. 83
3. Appropriability Eﬁ‘ect ................................................................................................................. 91
4. Interaction Between Product and Process R&D ........................................................................ 98
5. Unobserved Heterogeneity ....................................................................................................... 106

vii

SUMMARY AND CONCLUSIONS 109

 

 

 

 

 

 

APPENDICES 1 1 5
APPENDIX A 116
APPENDIX B 118
APPENDIX C 122
APPENDIX D 136
APPENDIX E 138

 

REFERENCES 140

 

viii

LIST OF TABLES

Table 1. Summary Statistics: Before Independent Variable Based Exclusions ($MM).. 72

Table 2a. Summary Statistics — Sample Data aﬁer Exclusions (SMM) ........................... 73
Table 2b. Summary Statistics by Year-Sample Data (SMM) .......................................... 74
Table 3. Value of Aggregate Assets ................................................................................. 86
Table 4. Value of Disaggregated Intangible Assets ......................................................... 90
Table 5. Value of Disaggregated Assets Controlling for Firm Size ................................ 95
Table 5b. Stock Response at the Mean — Weighted Regression ...................................... 96
Table 6. Value of Aggregate Assets, Controlling for Firm Size ...................................... 97
Table 6b. Stock Response at the Mean — Weighted Regression ...................................... 98
Table 7a. Interaction Between Product R&D and Process R&D ................................... 101
Table 7b. Stock Response at the Mean — Weighted Regression .................................... 102

Table 7c. Interaction Between Product R&D and Process R&D, Controlling for Firm Size
................................................................................................................................. 104

Table 7d. Stock Response at the Mean - Weighted Regression .................................... 105

ix

LIST OF FIGURES

Figure 1. Relationship between revenue, cost, and R&D expenditures. .......................... 51
Figure 2. Relationship between R&D proﬁt and R&D expenditures. ............................. 52
Figure 3. Relationship between marginal revenue and marginal cost. ............................ 55

Figure 4. Relationship between marginal net revenue (proﬁt) and ex ante sales for each
type of R&D assuming all parameters except h equivalent in each dimension. ....... 58

Figure 5. Relationship between optimal (proﬁt maximizing) R&D expenditures and ex
ante sales. .................................................................................................................. 60

Figure 6. Optimal allocation strategy with respect to sales. ............................................ 61

INTRODUCTION

Technical innovation is a crucial input to economic growth and as such is a
primary target for public policy designed to enhance national, state, and/or local
innovation performance.‘ However, one could argue that these policies have been limited
in their effectiveness because many aspects of the process of innovation remain poorly
understood. Until recently, the majority of research on innovation has been focused on
macroeconomic or industry effects, arguably at the expense of research aimed at the ﬁrm
or business unit level. Accordingly, this study builds upon a relatively new and growing
body of literature that investigates innovation at or below the ﬁrm level. Speciﬁcally, it
examines the relative proﬁtability to individual ﬁrms of different types of research and
development expenditures and the value of alternative R&D resource allocation
strategies.

Economists long have recognized the important link between technological
change and economic growth, although the modern investigation of this topic can be
traced to Joseph Schumpeter (1942).2 Schmnpeter hypothesized that only ﬁrms in non-
perfectly competitive industries would be able to innovate and create the technological
change necessary for economic growth. In his view, technological change could only be

attained once ﬁrms grew large enough to raise prices above their competitive level and

 

' Examples include the Federal Research and Experimentation tax credit, the patent system, and numerous
state and local tax and investment credits. For a discussion of the inﬂuence of government policies on
innovation, see Horwitz (1979) and Hollomon (1979).

2 Early references to the importance of technical change include Adam Smith’s The Wealth of Nations
(1776) and David Ricardo’s 0n the Principles of Political Economy and Taxation (1817).

gain economic proﬁts — proﬁts that could then be used to invest in technical research and
development. The idea that ﬁrm size and technological change are related has become
known as the Schumpeter hypothesis. Although much effort has been devoted to
investigating the validity of Schumpeter’s hypothesis, today researchers are faced with
the hard truth that the relationship between market structure, ﬁrm size, and technological
change is not nearly as clear-cut as Schumpeter originally hypothesized. In addition to
ﬁrm size, other industry and ﬁrm characteristics play an important role in innovative
activity.3

Not surprisingly, the relationship between R&D and innovation has received a
considerable amount of attention in the literature. There is little doubt a strong link exists
between R&D and successful innovation, but thus far, empirical efforts to quantify this
link have been varied in both method and result. The reasons for this are several. First,
the true nature of the relationship between proﬁt and R&D, be it linear, quadratic, or
some other form, is not known with certainty. Second, long lags oﬁen exist between
R&D investment and the realization of the returns from R&D. The lag introduces
uncertainty in the appropriate speciﬁcation of the model, and disqualiﬁes many sources
of data that are not long enough to cover the appropriate time span. Third, a signiﬁcant

lag time between action and the observation of results limits the ﬁrm or policy maker’s

 

3 Examples include (1) the level of technological opportunities in the industry, (2) the appropriability
conditions in the industry, (3) sunk costs in the status quo, and (4) the stage of the product life cycle. At the
ﬁrm level, (5) the degree of interface between customer demand and the R&D department, (6)
organizational design, and (7) applied R&D expenditures. See Acs and Audretch (1987), Scherer and Ross
(1990), Hayes and Wheelwright (1984), and Utterback ( 1994).

use of the relationship as a tool for planning.

Another potential problem to measuring the returns to R&D lies in determining
the correct proxy for innovation. That is, the composition of R&D with respect to
product, process, or other types of R&D may by as important in determining the returns
to innovation as the aggregate amount of R&D. Unfortunately, severe data cOnstraints
have limited empirical efforts to decompose total R&D into reliable measures of product,
process, or other types of R&D.

From an empirical standpoint, the approach to measuring the returns to R&D
taken in this study avoids many of these issues. This method, known as stock market
valuation, relates R&D spending to the value placed on a ﬁrm’s assets by ﬁnancial
markets. It leverages the idea that publicly traded corporations are bundles of assets
(such as R&D spending, advertising, and tangible assets) by incorporating these assets
directly into the valuation equation. Therefore, it is particularly well suited to measuring
the value of intangible assets. This method also allows a great deal of ﬂexibility in terms
of data structure and hypothesis speciﬁcation. Multiple types of R&D may be regressed
against market value, thus enabling the relative value to the ﬁrm of each type of R&D to
be determined. The approach can also be used to determine if the returns from one type
of R&D are signiﬁcantly affected by an increase in returns from the other. Recent studies
using a stock market valuation approach include Hall (1993) and Doyle (1994).

Awareness that the composition of R&D is important to ﬁrm value is growing.
Recent studies have indicated that some degree of success in both product and process

innovation is necessary for a ﬁrm to be successful over the long term (Capon et. al.,

1992). However, product and process R&D may not be equally valuable for all ﬁrms or
completely independent of each other within ﬁrms. Cohen and Klepper (1996) suggest
that larger ﬁrms beneﬁt more ﬁ'om process innovation than from product innovation.
They also suggest that external factors, such as market conditions, can lead to positive
appropriability interactions between product and process R&D. Athey and Schmutzler
(1995) suggest that long-run organizational design factors such as ‘ﬂexibility’ may
contribute to ‘complementarities’ (i.e. positive extemalities) between product and process
innovation that enhance the overall innovativeness of the ﬁrm and thus its long-run
competitiveness.4

Yet, conventional wisdom suggests that ﬁrms have difﬁculty maintaining their
efﬁciency in both product R&D and process R&D simultaneously over an extended
period. This is sometimes referred to as a “trade-off” between product and process

innovation. A 1993 article in Harvard Business Review is typical of this viewpoint.

“.... either/or dichotomies dictated most managerial choices. A company

could pursue a strategy of providing large volumes of standardized goods

 

‘ The term ﬂexibility has been used in the economics literature to refer to a variety of concepts, although
some of these concepts are closely related. For example, Moroni (1992) describes what he terms
‘production ﬂexibility’, which he vaguely deﬁnes as the capacity of the ﬁrm to adjust to variations in
external conditions. This deﬁnition of ﬂexibility encompasses ‘strategic ﬂexibility’, such as "the ability to
Change production processes, production elements endowments and the qualities of outputs in relation to
Changes in environmental conditions" (p. 168). It also includes the concept of ‘operational ﬂexibility’,
Which "is related to the possibility of varying the quantities produced within a given mix, using a given
PTOductive structure.” (p. 168).

or services at low cost, or it could decide to make customized or highly
differentiated products in smaller volumes at a high cost. In other words,
companies had to choose between being efﬁcient mass producers and
being innovative specialty businesses. Quality and low cost and
customization and low cost were assumed to be trade—offs.” (Pine, et. al.,

p.111)

In this instance, U.S. automobile makers, in response to Japanese competition, adopted a
variety of new organizational designs in an effort to overcome this trade-off and improve
their competitiveness. To an extent, these designs, such as product teams, were
successful; but they also created other problems, “Cross-functional coordination has
improved, but at the cost of depth of knowledge within functions.” (Sobek et. al, p. 37).

Anecdotal evidence of this apparent trade-off between success in one type of
innovative effort and another abound. General Motors, IBM, Xerox, and Motorola, to
name a few, have had technologically superior products in the past. However, despite
their great size and obvious market power, each has had its ability to create or leverage
innovative and technologically competitive products questioned (Chesbrough and Teece,
1996)

In addition to the research noted above, the potential for infra-ﬁrm interactions is
reﬂected in the management science and organizational theory literature. This school of
thought puts more faith in the notion of trade-offs than complementarities because they

tend to view the ﬁrm as a complex, multi-function organization, rather than as a proﬁt

maximizing entity.5 In general, theories of the ﬁrm other than the standard neoclassical
economics viewpoint allow much more room for intra-ﬁrm interactions and sub-
optimizing behavior.

Unfortunately, few theoretical models have been developed that account for the
distinction between product innovation and process innovation. Of these, most tend to
focus on the relationship between ex-post complementarities associated with
appropriability and ﬁrm behavior. They generally ignore any factors that might impose
constraints on ﬁrm behavior.6 In chapter 2, a theory of R&D allocation is developed that
takes into account organizational constraints and market conditions that can form the
basis for interactions between product innovation and process innovation. The efﬁciency
of overall R&D may be affected positively, negatively, or not at all by these interactions.
In chapter 3, using a relatively new set of data, the contribution of R&D composition to
ﬁrm proﬁtability is investigated empirically. In particular, the relative value of product
R&D and process R&D is examined, along with an investigation of the potential for
interactions between product and process R&D.

In summary, the ability to measure the returns to different types of R&D will
provide new insights into how the allocation of innovative resources affect an

organization’s proﬁts and equity values, and how outside factors may inﬂuence these

 

5 Theories of the ﬁrm other than the standard neoclassical theory include contractual relationship theory,
principal-agent relationship theory, decision theory, etc. See Tirole (1990).

6 For example, Cohen and Klepper (1995) devise a model whereby ﬁrms choose to invest in product or
process innovation based solely on proﬁt incentives. They ﬁnd that these incentives are complementary,
but fail to consider the possibility of constraints in the ﬁrm’s maximization problem. Klepper (1996)
develops a model of product and process R&D based on this same incentive principle to explain many of
the stylized facts of the PLC. It works rather well, but is too general to be useful from a policy perspective.

choices. The results of these inquiries can lead to improved R&D allocations by
improving the information upon which the organization bases its R&D decisions. The
results also will be a step forward in the quest to improve public policy towards R&D by
providing policy makers with a better understanding of the determinants of ﬁrm
innovativeness. For instance, one might argue that the R&D tax credit, depending upon
how it is structured, beneﬁts one type of R&D more than another, resulting in a sub-
optimal allocation of R&D resources. Thus, a better understanding of the relative value
of different types of R&D and the different conditions favoring each type would lead to a
more efﬁcient structure for the R&D tax credit.

The study is organized as follows. The review of the literature is presented in
chapter I. Chapter 11 contains a theoretical framework and model for R&D allocation.
Data, empirical speciﬁcations, and empirical results are contained in Chapter III. A
summary of results and some opportunities for further research are presented in the

Conclusion.

1. REVIEW OF THE LITERATURE

Innovation has been deﬁned as “an idea, practice or a material artifact perceived
to be new by the relevant unit of adoption” (Zaltrnan et al. 1973, p. 10) and “... the search
for and the discovery, development, improvement, adoption and commercialization of
new processes, products, and organizational designs and procedures” (Jorde and Teece,
1992).7 This chapter presents a review of the literature on innovation, with a focus on
research pertaining to conditions, behaviors, and structures that inﬂuence ﬁrm-speciﬁc
innovation performance and value. I will discuss product, process, and other types of
innovation and the inﬂuences of ﬁrm organization on the performance of each.
Throughout this section I discuss the implications for innovation performance and
behavior at the ﬁrm level. Finally, I discuss the returns to innovation in terms of

determinants and measurement issues.

A. Product, Process, and Other Types of Innovation

Generally, product innovations affect the demand curve and process innovations
affect the supply curve. Lee and Stone (1992) ﬁnd that higher product R&D in an industry

is correlated with higher price levels in the industry, while higher process R&D leads to

 

7 Innovation and invention are often confused. Innovation is a much broader concept - one that
encompasses ﬁre entire process of developing a new product or process and bringing it to market. Invention is
critical but it is merely one facet of the entire innovation process. Scherer and Ross (1990) deﬁne invention
as the “act of insight by which a new and promising technical possibility is worked out...” (p. 616).

lower overall prices. More formally, product innovation has been deﬁned in terms of any
new product introduced by the organization (Knight 1967), or more broadly as any
emerging technology or combination of technologies (Utterback and Abernathy, 1975).
Tomatsky (1983) extends these to apply to any of the above that are exploited to produce
goods for consumption. Process innovation, on the other hand, has been deﬁned as “any
operations technology that is new to the adopting organization,” (Collins et. al., 1988, p. 1)
or “a change in the way products are made or deliver ” (T ushman and Nadler, 1986, p.
76). A broader deﬁnition put forth by Knight (1967, p. 482) deﬁnes process innovation as
the “introduction of new elements in the organization’s task, decision, and information
system or its physical production or service operations.”

The idea that a product innovation affects the demand curve implies that the intent
of a product innovation is to achieve two related goals: (I) retain or increase market share
(either by introducing a new product or improving an existing one) and (2) widen the price-
cost margin by increasing the price (value) of an existing product or create value by
introducing a new product. By extension, any change to a product that results in no new
market share nor the ability to raise prices, all things being equal, can not truly be
considered a commercial product innovation (at least not a successful one). In other words,
one must apply the following litmus test; if one holds market share constant, does the
innovation allow the ﬁrm to increase price? Alternatively, will demand for the product
increase while price is held constant? If the answer to either question is yes, then the

innovation can be considered a product innovation.

Again, the deﬁnition of a process innovation typically is dependent upon how it
affects supply. For instance, Kanrien and Schwartz (1982) deﬁne process innovation within
the context of productivity increases. They state that contributions to increases in
productivity over time arising from technical change can be associated with shifts in the
production function. Therefore, process innovations are speciﬁcally designed to increase
the price-cost margin of a product by lowering the costs associated with production, via
improved manufacturing techniques and/or equipment (of course, lower costs also may
contribute to increased market share).

However, the above deﬁnitions of product innovation and process innovation are
inadequate in many respects. Most innovations display characteristics of both. Therefore,
all innovations can be viewed as existing on a continuum with a pure product innovation at
one extreme and a pure process innovation at the other, with most innovations lying
somewhere in between. For instance, a given innovation could be viewed as a product
innovation by the ﬁrm that created it, but a process innovation by a downstream ﬁrm that
uses it in its manufacturing process. Consumers of the innovation (the downstream ﬁrm in
this example) can decrease their costs and increase their productivity by using the new or
improved good. Producers of the innovation increase their demand and obtain higher
proﬁts. In this case, an innovation displays characteristics of both a process innovation and
a product irmovation depending on who beneﬁts from it. In general, most product
innovations, with exceptions such as consumer goods, can be considered someone else’s
process innovation.

Another illustration of the difﬁculty of classifying innovations as product or process

10

is the case of a product improvement not only increasing the demand for the product, but
also simultaneously becoming easier to manufacture than the earlier version. This would
both increase the value of the product and decrease its production costs, thereby exhibiting
characteristics of both a product innovation and a process innovation as viewed by a single
ﬁrm. Conversely, a process innovation may simultaneously reduce the manufacturing costs
of a product, improve its quality, and ultimately increase demand. In either case, unlike the
previous example, the same innovation is affecting a demand and supply curve in the same
market. Therefore, most innovations affect a demand curve and a supply curve either
within the same market or in two or more markets.

Bhoovaraghavan et. a1. (1996) attempt to address these deﬁnitional issues by
classifying product and process innovation in terms of consumer choice rather than in terms
of supply side/resource allocation. In their view, an innovation is classiﬁed as product or
process based on the perception of the consumer towards the innovation in relation to an
already established product(s) that satisﬁes the consurner’s “core want”. The classiﬁcation
of innovation depends upon the additional wants that are satisﬁed by the new product that
were not satisﬁed by the already existing product(s). “Process innovations are always
perceived by the consumer as products that cater to or supplement only existing core wants.
Product innovations are perceived by the consumer to be products that either cater to a new
core want or supplement existing core wants.” (p. 237). Unfortunately, these deﬁnitions
allow much room for interpretation and still place any innovation on the continuum
somewhere between a pure product innovation and a pure process innovation.

Despite the issues with deﬁnitions, there are several reasons why drawing a

11

distinction between process and product innovation can be useful. First, studying how
ﬁrms allocate innovation-related resources provides insight into what industry deems
important. Second, product and process irmovations, from a macroecononric perspective,
affect the economy in different ways. For example, Japan’s economic emergence has been
attributed in large part to its focus on process innovation, in contrast to the United State's
focus on product innovation (Mansﬁeld, 1988). Third, product innovations are the basis for
the creation of new markets.8 In contrast, process innovations are usually created with the
intent to increase productivity, lower costs, and/or redirect existing demand to different
producers.

The distinction also helps shed light on strategic interactions among ﬁrms. Product
innovations are associated with innovative spillovers. Once a product is sold, the
innovation is much more difﬁcult to keep from being imitated, even with patent protection.
Conversely, process innovations are easier to keep within the ﬁrm (Cohen and Klepper,
1998).

Product and process are not the only possible innovation distinctions. Another
useful distinction can be made between input and output innovation investments. Input
investments include research and development (R&D), physical capital, and any
investments undertaken before the innovation is ready for sale. Output investments are
those that are intended to facilitate the sale of the innovation once it is at or near its ﬁnal
form. These include advertising, patents, liquidity, legal fees, distribution networks, sales

forces, etc. Although much attention is paid to R&D, some estimate as much as half or

 

8 New technological trajectories are an important example of this. See CED (1980).

12

more of the total investment needed for a successful (proﬁtable) innovation usually comes
after the completion of the R&D eﬁ‘ort (CED, 1980). In other words, an input investment
will determine the ex-ante capability to generate an innovation whereas an output
investment will increase appropriability, i.e., enhance the ex-post capability to proﬁt from
an innovation once it has been created.

Bernstein (1986) incorporates the distinction between input and output investments
by decomposing the production function of the ﬁrm into three primary inputs: labor,
physical capital, and formal R&D. He analyzes the relationship between these inputs in the
innovation process and ﬁnds two main effects. First, he ﬁnds a “substitution” effect among
the inputs such that physical capital and R&D are complements, whereas labor is a
substitute for the other two. In other words, ceteris paribus, an increase in the expenditure
on physical capital also increases the expenditure on R&D, and vice versa, but reduces that
of labor. However, increases in inputs presumably lead to increases in output as well, and
thus demand increases for all inputs, including labor. He identiﬁes this as the “output”
effect. Essentially, there are complementarities (positive interactions) between input
investments and output investments.

Innovation-related investment also can be classiﬁed as tangible (physical) or
intangible. Intangible investment is usually ﬁrm speciﬁc, such as organizational design,
management, ﬁrm culture, knowledge, reputation, and informal R&D. A growing literature
indicates that intangible capital may play a disproportionately large role in high technology
and start-up ﬁrms, and is a growing proportion of corporate R&D (Crawford, 1991). This

could be attributed to the fact that start-up ﬁrms in general do not have the ﬁnancial

l3

resources of larger, more established ﬁrms, and are forced to develop other strategies to
compete, such as a heavier reliance on superior technology or superior management. Still
another reason that ﬁrms may be becoming more intangible could be that technology itself
is becoming less physical capital intensive and more information oriented (computers,
software, intemet, etc.).9

Innovative activity is categorized oﬁen as basic research or applied research. Basic
research can be deﬁned as “original investigations that advance scientiﬁc knowledge but do
not have speciﬁc commercial objectives, although such investigations may be in ﬁelds of
present or potential interest to the company” (IRI/CIMS survey documentation, 1999).
Applied research can be deﬁned as “Investigations directed to the discovery of new
scientiﬁc knowledge having speciﬁc commercial objectives with respect to products,
processes or services” (IRI/CIMS survey documentation, 1999). An innovative ﬁrm
undertakes both types of research to remain competitive.

Finally, innovative activity, particularly R&D, often is classiﬁed as internal or
external. Internal innovation is research and development performed in-house, whereas
external innovation is outsourced. External R&D can take the form of research joint
ventures, purchased R&D, purchased technology, patent purchasing and licensing,
limited partnerships, and certain mergers and acquisitions. Crawford (1991) suggests that

technology/knowledge gathered outside the ﬁrm is a growing proportion of total R&D.

 

9 R&D in general commonly is referred to as intangible, although here I am using the term to differentiate
R&D itself.

14

Mowery (1994) also provides evidence that external R&D is growing in importance.10
Given the apparent growing importance of external R&D, treatments of external
R&D issues have not been prominent in the literature. Exceptions are Audretsch et. a1.
(1996) who analyze a data set ﬁ'om a survey of Dutch manufacturing ﬁrms. The ﬁrms
were asked to report whether or not they had conducted external R&D during a speciﬁed
year (1983). Unfortunately, the amount of external R&D and the type, product or
process, was not speciﬁed. A probit model was used to compare the frequency of
external R&D to a variety of independent variables, such as ﬁrm size, capital intensity,
skilled labor, amount of total R&D, etc. They ﬁnd that capital intensity (positively) and
skilled labor (negatively) were signiﬁcantly correlated with external R&D. They explain
their ﬁndings by theorizing that a high level of ‘asset speciﬁcity’, in this case labor assets
(researchers and scientists), tends to increase the value of internal R&D but not that of
external R&D. On the other hand, “a highly capital intensive ﬁrm will tend to produce a
relatively standardized product, which can only be copied with great difﬁculty by another
ﬁrm. Thus, ceteris paribus, external R&D is expected to be more prevalent in ﬁrms
which are capital intensive.” (p. 521). This explanation is consistent with Bernstein’s

( 1986) results that showed a complementary relationship between physical capital

 

'° The changing nature of technology itself may contribute to the ability of ﬁrms to ﬁnd compatible
external R&D. Information technology (computer design, faxes, intemet, and e-mail), decreasing barriers
to transportation (lighter materials, faster transport, etc.), and other technologies are giving ﬁrms access to
a larger pool of potential research partners.

15

accumulation and R&D investment, a signiﬁcant portion of which, in light of the results

reported by Audretsch et. al., may consist of external R&D.“

B. Product and Process Innovation within an Organization

Do factors other than proﬁt maximization, such as organizational design
constraints, inﬂuence the returns to R&D? In general, the standard neoclassical
viewpoint of the ﬁrm does not allow for intra-ﬁrm interactions and sub-optimizing
behavior. Other, less restrictive, viewpoints allow a variety of factors to affect the R&D
allocation decision and to depart from an exclusive focus on proﬁt maximization. As
Athey and Stern (1998) point out, the objective function faced by the agent responsible
for R&D decision making is not necessarily limited to the overall economic proﬁts of the
ﬁrm.

If one views the ﬁrm as a bureaucratic organization, with all of an organization’s
complex interactions, then there are several theoretical reasons to believe extemalities
may exist between product and process R&D. Athey and Schmutzler (1995) suggest that
long-run organizational design factors such as “ﬂexibility” (in either the product or the
process dimension) may contribute to complementarities between dimensions that

enhance the innovativeness of the ﬁrm and thus its long-run competitiveness.” They

 

” Gompers and Lerner (1998) suggest that corporate investments in venture-type endeavors can be nearly
as successful as private venture capital projects provided they are strategically aligned with the parent
company. They ﬁnd these strategic investments stable over time, which is consistent with the theory that
ﬁrms use external R&D to address structural shortcomings in their internal R&D program.

'2 See footnote 4.

l6

develop a two-period decision model. In the ﬁrst period a ﬁrm chooses its ability to
generate innovations, or in their language, chooses product and process ﬂexibility and
physical research capabilities. In the second period the ﬁrm takes these as given and
maximizes proﬁt by choosing a degree of innovative intensity in both dimensions. Their
results suggest that there are complementarities in the innovation process that. manifest
themselves in three particular ways: complementarities between dimensions in the short-
run resource allocation decisions, complementarities between dimensions in the long-run
capability decisions, and complementarities between the long-run and the short-rim
decisions.

Although Athey and Schmutzler support the idea of a positive interaction between
product and process innovation, management theory offers some support to the idea that
trade-offs, not complementarities, characterize the relationship between product and
process innovation. This opposing position is based on the belief that the complexity of
the ﬁrm necessitates compromises among multiple organizational fimctions, typically
characterized as engineering, marketing, and manufacturing (Clark and Wheelwright,
1993, p. 161). Likewise, Volderba (1998) and Teece et. a1. (1997) suggest the ﬁrm is
composed of a spectrum of ‘capabilities’. Chesbrough and Teece (1996) stress the
importance of balancing the beneﬁts of centralization against the potential rewards from
risk-taking, which over-centralization can undermine.

The idea of the ﬁrm as a bureaucratic organization also is common in principle—
agent and contractual relationship lines of research. In this spirit, allocation decisions

between product and process innovation can be viewed as a function of organizational

l7

design. Organizational design is a term that encompasses the administrative and
management structure of the ﬁrm (how decisions are made), the manufacturing system
(e. g. “continuous process improvement” strategies and “teams”), the extent and structure
of the distribution network, incentive packages, etc. Organizational design determines
how physical capital is used and is an intangible asset (whereas research facilities,
equipment, and personnel are tangible in nature).

Hayes et. al., (1988) point out that ﬂexibility in the manufacturing process is
important to the long-term survival of the ﬁrm, but that within the ﬁrm there may be
trade-offs between product ﬂexibility and process ﬂexibility. Utterback (1979), echoes
this idea, “... conditions necessary for rapid (product) innovation are much different from
those required for high levels of output and efﬁciency in production” (p. 40). Reasons for
this trade-off are numerous and include both tangible and intangible factors. For
example, a tangible factor could involve the ﬁrrn’s manufacturing equipment. The nature
of the equipment may impose constraints on the ability to retool in response to a product
innovation that has different manufacturing requirements than the previous product.
Intangible factors include organizational and administrative design limitations, such as
the failure to learn across fimctions and projects and/or the existence of cultural rigidities
within the fum.l3 For example, Hayes et. a1. (1988) describe how the failure to align
development “windows” properly between product and process innovations can

undermine innovation (pp. 298-99), and they emphasize that strategic interactions

 

’3 See Hayes et.al (p. 338) for discussions relating to organizational ﬁrnctions. See Ginn (1995, pp. 357-
410) for a discussion relating organizational culture to technical creativity.

18

between the two dimensions are of primary importance to the success of any project.”

Many of these factors could be characterized as side-affects stemming ﬁ'om problems
related to ﬁrm size, such as excessive bureaucracy, communication breakdowns, and
labor issues.”

In the economics literature, the idea of a trade-off is reﬂected in research that
details empirical differences in the intensity and composition of R&D among ﬁrms. Acs
and Audretsch (1987, 1991) provide evidence that not only does R&D intensity vary
among ﬁrms in different industries, but among ﬁrms in the same industry according to
ﬁrm size (and other factors). They ﬁnd that many of the most important innovations are
produced by relatively small, new ﬁrms, whereas large ﬁrms tend to focus on incremental
innovations. This implies one or both of the following: (1) ﬁrm size provides incentives
for larger ﬁrms to concentrate on incremental product innovation, or (2) ﬁrm size inhibits
the ability to create drastic product innovations. Doyle (1994) provides evidence that the
value of overall R&D varies with ﬁrm size in high technology industries. Of course, this
suggests the question: does the efﬁciency/value of product R&D and process R&D vary
with the size of the ﬁrm, and if so, how?

Whether the interaction, if it exists, is positive, as Athey and Schmutzler suggest,
or negative, as much of the management literature implies, could have important

consequences for ﬁrm evolution. For example, the trade-off hypothesis provides a

 

" For further evidence, see Gruber, 1981. There is also evidence that venture-type corporate programs are
more successful and stable if they ﬁt the overall strategy of the corporation (Gompers and Lerner, I998).
The fact that non-strategic ventures are less successful is consistent with the idea that organizational factors
are important to R&D.

19

plausible explanation for Cohen and Klepper’s (1996) ﬁnding that process innovation
increases with ﬁrm size. Organizational constraints, be they tangible or intangible,
prevent the ﬁrm from performing product R&D and process R&D with equal efﬁciency
as the ﬁrm grows in size.

If extemalities between product and process R&D exist, they will be inﬂuenced
by observable internal and external ﬁrm and industry characteristics. These
characteristics are closely linked to organizational factors within the ﬁrm and external

appropriability incentives.

“Forces both inside and outside the industrial ﬁrm inﬂuence the
(innovation) process. Outside forces include users’ needs, changing prices
of inputs, competitive stresses, and government stimuli and regulations.
The inside forces include the ﬁrrn’s resource allocations; the product and
process technologies themselves; the people, organization, and
communication patterns involved in producing innovations, and the

technical resources and strengths of the ﬁrm” (Utterback, 1979).

In addition, a trade-off would likely be a result of a combination of both
organizational design constraints and tangible factors. For example, organizational
design constraints may imply that a fertile product-nmovating environment requires “a

stimulating, supportive leadership and a positive, encouraging organizational climate that

 

‘5 For example, see Hayes and Wheelwright (1984), Utterback (1994).

20

allows employees to take risks” (Holt and Seetzen (1975), p. 226), whereas a fertile
process-innovating environment requires a structured organization with a focus on cost
containment and repetition. These goals may very well be mutually exclusive. Creating
an environment conducive to one inhibits the ability to create an environment that fosters
the other. Furthermore, the frrm could be faced with tangible constraints as well. For
example, if the ﬁrm competes successfully in a market with a relatively fast innovation
pace (3 short technological-life-cycle), it will be proﬁcient, by deﬁnition, at product
innovation. However, it will also have less time to develop production efﬁciencies
(process irmovate) because the optimal production process may change over the course of
the PLC. This will impose additional costs on manufacturing because of the constant
need to retool and alter production processes to meet the demands of the new product. In
this way, a product innovation may create an extemality on the process side by
necessitating different production techniques and/or forcing the ﬁrm to retool.

Pine et. a1. (1993) discuss innovation within the context of mass customization
and the problems inherent in establishing an organizational design supportive of both
dimensions. They relate a particularly illuminating story about the difﬁculties Toyota
Motor Company encountered as it attempted, beginning in the late 1980’s, to transform
itself ﬁ'om a ﬁrm that focused on ‘continuous process improvement’ to one which was
able to make “... varied and often individually customized products at the low cost of
standardized, mass-produced goods” (p 108). Mass customization, in which a ﬁrm
strives to add additional value to each good produced yet manufacture them at low cost,

could be considered a product AND process-improving system. It attempts to increase

21

demand for the product using product enhancements while simultaneously keeping costs
to a minimum through process enhancements. As Toyota came to realize, mass
customization is an entirely different world than continuous process improvement
(whereby a ﬁrm endeavors to manufacture goods at low costs - although with the co-
requisite of preserving quality), which could be considered a one—dimensional process
innovating system. The authors illustrate this point by noting that after some early
success, Toyota was forced to retreat from its goal of becoming a mass-customizer. Part
of the problem was the economic recession in the early 1990’s, but that wasn’t the whole
story. Management was forced to acknowledge that “continuous improvement and mass
customization require very different organizational designs, values, management roles
and systems, learning methods, and ways of relating to customers” (p.109). Whereas
process improvement requires a relentless pursuit of quality through a vision of “being
the best” and managers constantly striving to tighten the links between processes, mass
customization requires a “... dynamic network of relatively autonomous operating units”
(p. 110). In attempting to evolve into a ﬁrm proﬁcient in innovation in two dimensions,
Toyota discovered that success in one dimension may not be compatible with success in

the other.

22

C. Returns to Innovation

1. Determinants

The following discussion is concerned exclusively with returns ﬁom private ﬁrms
and does not consider the deterrninates or measures of value derived from innovation in
the public domain, such as Universities or publicly funded research institutes. With that
said, before a ﬁrm can undertakes an R&D project it “... must consider the cost of doing
research, the probability of succeeding, and the likely degree of competition...”
(Quinnbach, 1993).

The cost of doing research is heavily inﬂuenced by the ﬁrrn’s level of
commitment and its ability to manage costs, which in turn rests upon the efﬁciency of the
ﬁrm’s organization - a consideration I already have discussed. The other two
considerations, the probability of succeeding and the degree of competition, depend on
three types of uncertainty: (1) technological uncertainty, (2) strategic uncertainty, and (3)
market uncertainty. Technological uncertainty refers to the possibility that the ﬁrm may
not possess the necessary technical expertise to develop an opportunity into a viable
commercial innovation. Strategic uncertainty arises from the possibility that the ﬁrm
may not be the ﬁrst to introduce the innovation to the market. Market uncertainty results
from a potential lack of demand from buyers.”

An alternative but related way of looking at these considerations, sans costs, is

within the conditional probability framework developed by Mansﬁeld et. a1. (1977).

23

They use three success probabilities to characterize the success rate of innovations: (1)
the probability that a technical goal will be achieved, (2) the probability that the resulting
innovation, conditional on technical success, will be commercialized, and (3) the
probability, conditional on successful commercialization, that the innovation will yield a
proﬁtable return. On average, 27% of the projects they studied ultimately were
proﬁtable.

Mansﬁeld’s ﬁrst success probability corresponds closely to technological
uncertainty. His third success probability can be seen to include strategic uncertainty and
perhaps market uncertainty. Success probability 2 is less concrete. It could refer to
market uncertainty in that the ﬁrm decides simply to shelve the product because there is
no anticipation of demand. It also could refer to ﬁnancial constraints, to a ﬁrm’s inability
to adequately take advantage of its resources, or to all of the above. In any case, R&D
costs and the rate of success of any of these probabilities most certainly will vary among
ﬁrms and therefore will depend upon two considerations. First, the characteristics of the
ﬁrm itself (its assets and organizational design), and second, the characteristics of the
industry or industries in which it competes (market conditions).

I have already discussed the impact of organizational constraints on innovation, so
I will turn my attention to the importance of market conditions, or “outside forces” in
Utterback's terminology, to determine the proﬁtability of innovation. Market conditions

provide the external incentives that drive the ﬁrm to innovate, and include technological

 

'6 Adapted from Encaoua et. a1. (1996).

24

push, market pull, and appropriability.17 Technological push refers to the idea that there
are exogenous technological innovation opportunities that ﬁrms can develop and
commercialize proﬁtably. Market (or demand) pull refers to the ﬁrrn’s perception of the
potential demand for the innovation.18 Appropriability is the extent to which a ﬁrm can
capture the new economic rents associated with a product or process innovation, and
depends inversely on the extent to which the innovation is a public good. -

In regard to the ﬁrst two conditions, the essential point is that the ﬁrm’s ability to
develop technical opportunities is squandered if it cannot identify which of those
opportunities will have a market demand. The ability to both recognize and develop
technical opportunities allows a ﬁrm to deﬁne a set of possible and “demanded”
innovations (in the case of a process innovation, the demander can be thought of as the
ﬁrm itself) which will lead to proﬁtability. Calantone et. al. (1996) point out that
although a ﬁrm’s R&D department responds primarily to technological push factors,
without an effective interface with the demand-pull source of innovative ideas,
technologically superior products can be developed that ﬁll no need in the marketplace.
Thus, an organizational failure can be the cause of a lower success probability 2.

Proﬁt from innovation ultimately is determined by appropriability conditions in
the industry. Therefore, of those innovations that pass the ﬁrst two criteria, the ﬁrm must
reasonably be sure that the returns are proprietary and large enough to earn a proﬁt.

Demand for the product must translate into demand for the product produced by the ﬁrm.

 

‘7 Scherer and Ross (1990).
'8 Holt and Seetzen ( 1975, p. 225) refer to this as “need pull”.

25

If other ﬁrms can quickly and cheaply offer a competitive product and compete the
economic rents to zero, then the product likely will not pay for its development costs.
The level of appropriability in large part is a function of the concentration in the industry,
the essential idea behind Schumpeter’s Hypothesis. Appropriability also is inﬂuenced
greatly by patent laws and regulation. See Scherer and Ross (1990) for a discussion of
appropriability and patent protection.

Quirmbach (1993) analyzes appropriability under a variety of industry conditions;
he considers alternative market structures (Bertrand competition, Coumot competition,
and collusion), patent protection or no patent protection, and projects with a high
probability of success vs. projects with a low probability of success. He restricts his
analysis to product innovation, primarily to cases where patent protection is unavailable
and irrritation is not immediate. He points out that a higher technical success probability,
assuming it is industry-wide, could adversely affect the industry’s level of appropriability
and reduce incentives to perform R&D. This effect tends to increase as collusion
decreases or the new technology exhibits increasing returns to scale.

His results also suggest that the anticipation of post innovation collusion
encourages R&D investment in the short run, but realized collusion later may discourage
continued R&D investment if patent protection is available. Under his assumptions,
patents also are more likely to increase appropriability in industries where new
technologies exhibit increasing returns to production. Generally, appropriability is lower
in Bertrand industries if the industry is not perfectly collusive (in fact, expected proﬁts

are zero if there is more than one ﬁrm). Appropriability is generally higher in Coumot

26

industries with the level varying depending on the project parameters.

Although Quirmbach derives a wide range of results under a variety of conditions,
he does not consider process innovation. This is an important omission because the level
of appropriability can vary between product innovation and process innovation. There is
little doubt that revenue incentives to innovate are strong in both dimensions. In addition,
it is likely that these incentives are complementary. Recall that studies by Capon et. a1.
(1992) and Landau and Rosenberg (1992) indicate that those ﬁrms who are able to
perform well in both areas of innovation are usually the most competitive and proﬁtable.
Taking this idea a step further, Cohen and Klepper (1996) develop a model of product
and process R&D based on such complementary incentives. They derive a positive
relationship between ex ante output and process R&D intensity, as measured by the ratio
of process R&D to the sum of process and product R&D. The returns to a product
innovation are enhanced if the ﬁrm subsequently decreases its production costs via a

process innovation. They state their basic relationship as follows:

In =fq"! /(fq”1 + grhq + 10”?) [1.1]
where p = the ratio of process R&D to total R&D. The numerator represents the proﬁt
maximizing level of process R&D as a ﬁmction of ex ante output q. The second term in
the denominator represents the proﬁt maximizing level of product R&D as a function of q
and as a function of K, the additional output derived from new product innovation. The
parameters f and g represent industry level technological opportunities for process and
product innovation, respectively. The parameters ﬂi represent the marginal returns to

innovation and can differ between product and process innovation.

27

The implication of equation 1.1 is that the derivative of p with respect to q is
positive “.. as long as the marginal return to process R&D declines sufﬁciently slowly
relative to product R&D” (p. 235). Their equation implies that a process innovation will
reduce costs and increase the price-cost margin. Both of these factors will increase the
returns to a given product innovation which itself increases the price-cost margin but now
can be applied to a larger quantity, thanks to the ex-ante process innovation. Given an
innovation in one dimension, any improvement in the other will allow an increase in the
ﬁrm’s ability to appropriate the returns to the second. These revenue complementarities
provide incentives to allocate R&D resources to both types of innovation, but not
symmetrically. As the ﬁrm’s sales grow, the incentive to process innovate begins to
surpass the incentive to product innovate, thus implying a positive relationship between
ﬁrm size (sales) and the returns to process innovation intensity.

Unfortunately, neither Quirmbach nor Cohen and Klepper consider organizational
factors in the R&D allocation decision. Although it may be true that improvements in
one area lead to more revenue as another area improves, this does not necessarily imply
that improvements in one area will make improvements in the other area less costly. Not
only are there the usual diminishing returns to R&D investment within a particular

dimension, but the rate of dirnirrishment could be affected by the other dimension.

28

2. Measurement

The value of innovative activity is determined by a variety of factors. Not
surprisingly, measuring the value of innovative activity is a complex issue. Regardless of
methodology, measuring the returns to innovation requires choosing an appropriate proxy
for the innovative activity itself. There are two common candidates: research and
development expenditures (R&D) and patents.19 Both are considered excellent, although
not perfect, indicators of the level of innovativeness. Patents are theoretically appealing
because they are considered a true output. Griliches (1990) has argued that patent
statistics, which are very detailed, are a superior indicator of innovative ability.
However, many innovations are not patented. For example, Cordes, et. al. (1987) ﬁnd
that many high technology ﬁrms do not patent their innovations because of the short
technological-life—cycles in their industries.20

R&D, on the other hand, is a better indicator of the effort put forth by the
innovator. However, R&D expenditures do not necessarily translate into successful
innovations, and R&D spending may not be the only source of innovation. Moreover,
R&D expenditure data “are subject to considerable error in reporting; ﬁrms are given

considerable latitude in classifying activities used for ﬁnancial reporting” (Doyle,

 

‘9 Not surprisingly, large corporations are responsible for the majority of both. In 1988, for instance, 73%
of the patents granted to US inventors were to corporations (82% of foreign patents were to corporations),
with 25% going to individual inventors (Griliches, 1990).

2° See Griliches (1990) and Hall (1998) for a summary of the advantages and disadvantages of each.

29

1994).21 Thus, determining the micro or macro impact of R&D spending on innovation is
empirically difﬁcult. However, the situation appears to be improving. In the US,
Standard and Poor’s Compustat data compiles extensive R&D information primarily
from 10-K reports to the Securities and Exchange Commission. The National Science
Foundation conducts annual R&D surveys, and surveys by others are conducted
periodically.22 Beginning in 1989, the UK. began requiring ﬁrms to report R&D
expenditures as a standard accounting practice. In summary, the choice between using
patents, R&D spending, or some other metric as a proxy for innovation boils down to
methodology and data availability.

Another common issue involves the choice of an appropriate measure for the
returns to innovation. Several measures have been used, including proﬁtability itself,
proﬁt margin (rate of return), total factor productivity, and market value. For example, a
stock market event study measures the change in the market value of a stock before and
after a speciﬁed event has occurred. The problem is determining the appropriate points in
time for the each measurement. This approach often is used to study the impact of
patents, although it is important to choose the patent(s) careﬁrlly (Austin, 1993). Many
patents are redundant because often there are “many viable solutions to a technical

problem, and other ﬁrms can ‘invent around’ a given patented solution” (Scherer and

 

2' Mansﬁeld (1986) and Hulten and Robertson (1984) both ﬁnd that the introduction of the R&D tax credit
in 1981 resulted in a considerable amount of non-R&D related spending being reclassiﬁed as R&D
investment to take advantage of the credit.

22 Examples include the Industrial Research Institutes R&D survey annually conducted by the Center for
Innovation Management Studies (which I use in this study), the survey conducted by Levin, Bohen, and
Mowery, 1985, and the European Innovation Monitoring System (EIMS).

30

Ross (1990) p. 624), and many patents are simply not commercially viable. It is known
that the distribution of the value of patents is extremely skewed, and thus many patents
contribute little to economic growth.

Other methodologies include creating a metric for innovation success and
regressing it against proﬁts (McGrath and Romeri, 1994), or regressing various ﬁrm and
industry characteristics against the ﬁrm’s rate of return. Geroski et. al. (1993) use this
latter approach to estimate the impact of innovation on the proﬁt margin of a large sample
of UK ﬁrms. Using the number of signiﬁcant innovations produced by a ﬁrm as their
measure of innovativeness, they found that an additional innovation contributed
anywhere between 6.1% and 16.5% to a ﬁrm’s proﬁt margin.

The most common approach to measuring the returns to innovation involves
relating total factor productivity to some measure(s) of innovation data (R&D often is the
more natural choice because of its input nature). For instance, another study by Geroski
(1989) examined the effect of innovation and other characteristics on total factor
productivity growth using a sample of 79 industries in the United Kingdom spanning a
three-year period. He found that innovation was by far the most important factor in the
study, accounting for as much as 70% of observed productivity growth. He also found
evidence that the impact of a single innovation is more durable over time than the impact
of other factors, such as entry by domestic or foreign ﬁrms. Despite the important results
from this methodology, this approach and others like it may be difﬁcult to implement for
several reasons: ﬁrst, long lags often exist between R&D spending and the realization of

the returns ﬁom R&D. The lag introduces uncertainty in the appropriate speciﬁcation of

31

the model, and disqualiﬁes many sources of data that are not long enough to cover the
appropriate time span. Second, a signiﬁcant lag time between action and the observation
of results limits the ﬁrm or policy maker’s use of the relationship as a tool for planning.
Third, the measurement of innovative activities at the ﬁrm or industry level can be
difﬁcult to match with other inputs, partly as a result of these same lags, and partly
because of data limitations. Therefore, it can be difﬁcult to incorporate other inputs, such
as other costs, into the analysis (Hall, 1998).

One popular alternative to measuring the returns to innovation, used in this study,
is a stock market valuation approach using R&D expenditure data. This approach has
gained popularity since it was ﬁrst introduced by Griliches (1981). The market valuation
approach regresses R&D expenditures, other intangible assets, and tangible assets against
the market value of the ﬁrm. The advantage to this method is that it avoids the problems
presented by the lag between R&D and its returns. This method is not a panacea,
however, because it is limited to evaluating private ﬁrms that are traded on a well
functioning ﬁnancial market, and therefore it does not evaluate the public returns to
innovative activity. In addition, it limits the researcher’s ability to incorporate line of
business data into the study. Nevertheless, it is particularly well suited to measuring the
value of intangible assets because it incorporates these assets directly into the valuation
equation, thereby leveraging the idea that publicly traded corporations are bundles of
assets (such as knowledge, reputation, and tangible assets).

The market valuation approach is basically a variation of the hedonic regression

method, which attempts to “determine the marginal value of a particular intangible asset

32

by regressing the market price for ﬁrms that possess the asset on various characteristics of
the ﬁrms” (Hall (1998), p. 4). In the market valuation of R&D application, one measures
the marginal value of an additional dollar of investment in a given type of corporate asset,
namely R&D. A central issue with this methodology is the method of valuing the ﬁrm.
The method most often employed follows traditional ﬁnance literature. The market value
of the ﬁrm is determined by the present discounted value, or net present value, of its

expected cash ﬂow into the future.

V = NPV(discounted cash ﬂow) . [1.2]

In the ﬁnance literature, cash ﬂow to the common stock owner usually is represented by

current and future expected dividends. In turn, the dividends paid out by the ﬁrm will be

a direct function of its earnings, or income stream.

V13, = NP V(income stream (dividends)) . [1.2’]

A more explicit representation of Equation 1.2’ is the following:

V1,, = Incomet +Incomet+ 1 /(1 +k)+1ncomet+2 /(I +k)2 + + Incomet+n /(1 +k)” [1.2”]

where k is the required rate of return on the common stock. For a depreciable asset, n

will equal the number of periods before the asset’s value is fully depreciated. All things

33

equal, higher depreciation rates will imply smaller n, reduce the number of terms in the
net present value computation, and therefore lower market values. Likewise, higher
required rates of return also will lower market value.23

Finally, in considering the market value effects of ﬁrm assets, it is necessary to
control for the expected level of investor risk. Risk comes in two varieties, systematic
and unsystematic. Unsystematic risk refers to risk unique to the ﬁrm, such as
management capabilities, R&D uncertainty, competition, or regulation. Systematic risk
refers to risk the variability of a stock’s return attributable to factor’s affecting the market
as a whole, such as interest rate changes, inﬂation, and consumer conﬁdence.
Theoretically, all unsystematic risk can be eliminated through diversiﬁcation, and thus
the reward to bearing risk should depend solely on its systematic risk (Ross et. all. (1995)
p. 347). Therefore, the value of an asset depends on its expected cash ﬂow discounted
appropriately, where systematic risk is a necessary factor in the discounting process, and
inﬂuences the investor’s required rate of return. In fact, following Doyle (1994, p. 59)

the required rate of return, k*, for a ﬁrm can be represented as

k‘ = rf+ ﬂf(km — rf) [1.3]

where rf is the risk free rate (e.g. treasury security yield), km is the market’s expected

 

2’ Given some simplifying assumption regarding income (or dividend) growth over time, two special cases
of equation 1.2” can be determined. First, if a zero growth rate is assumed, equation 2.1 ” simpliﬁes to
Vi, t = Income /k. Second, assuming a constant growth rate in income, equation 2.1” sirrrpliﬁes to

34

return on the overall market portfolio, and ,8] is the ﬁrrn’s beta. The beta is a traditional

ﬁnancial measure of risk relating the volatility of an individual common stock to the
volatility of a broad market index, such as the Standard and Poor’s 500, the DOW, or the

NASDAQ. Higher volatility relative to the index implies higher risk, but also larger
returns.24 Instead of the interest rate, k. is used as the discount rate in equation 1.2” to
obtain the net present value of the ﬁrm’s cash ﬂow/income, and hence its market
valuation. Thus, the higher the perceived risk by investors, the higher the beta, the higher
the required rate of return, and the lower the ﬁrrn’s market valuation.

In order to apply the above analysis toward the goal of measuring the retums to
innovation, the typical model of market value hypothesizes that the discounted value of
current and future income, and hence the value of the ﬁrm, is a function of its n number

of assets:

V“ =ﬁ(Income stream, (A ,1, A1,, ...... A n, t )), [1.4]

or more simply,

Vi', =ft(A,.,, A1,, ...... Any) [1.4’]

 

V13, = lncome(l +g) /(k-g), where g is the growth rate.
2‘ In some cases, rather than the beta, the Standard & Poor’s ranking of earnings and dividend stability is
used (Hirschey, 1982).

35

In the case of constant returns to scale of the proﬁt function and a single asset, we obtain
the familiar result that the market value of the ﬁrm is a product of the book value of its

assets times a ‘shadow price’, commonly referred to as Tobin’s q.

V, = q,g(A,, A2, ......) . [1.5]

Several assumptions underlie this method: ﬁrst, it is assumed that ﬁrms invest in these
assets in a value-maximizing manner. Second, appealing to the efﬁcient markets
hypothesis, it is assumed that the ﬁnancial market on which the ﬁrm trades is efﬁcient,
and that all optimal asset trades have occurred. Finally, although the true functional form
of the function f (or g) is unknown, it is assumed that a satisfactory approximation exists.
Typically, and primarily for econometric considerations, these approximations assume a
linear or a traditional Cobb-Douglas form.

The assets within the function f generally are classiﬁed as tangible or intangible.
Tangible capital usually is represented by the book value of the ﬁrm at time t, and
intangible capital reﬂects the stock of knowledge capital and advertising capital at time t.
Advertising capital, typically represented by the sum of the weighted ﬂow of advertising
expenditures over the previous x number of years (typically 5 years), reﬂects the value
imputed ﬁ'om the product’s reputation. Likewise, the ﬂow of R&D expenditures over the
previous x number of years is used to represent knowledge capital. The method of
weighting previous expenditures varies from study to study. Occasionally, current proﬁts

or cash ﬂow is included in the valuation equation to account for market power or long-

36

run proﬁtability not related to advertising or R&D (Hall, 1993 and Hirschey, 1982). Also
occasionally, the grth rate of sales is included as a proxy for expected future growth of
the ﬁrm, which some view as an important factor in common stock valuation (Hall,
1993)

Given the above, several speciﬁcations for estimating market valuation are
possible. After taking the natural logarithm of both Sides and applying approximations,

the linear model and the Cobb-Douglas model as speciﬁed by Hall (1998) are

Log VIJ =10g qr + at Log Air + 7: Ki,/Ai,! ’ [1-6]
and

Log Vt, =10g q, + 0? Log Ar, + 6!. Log (Kr/Au). [1-7]

respectively, where V“ represents the market value of the ﬁrm at time t, q, is Tobin’s q,
which in the regression represents the intercept term, A i, , represents tangible assets, and

Kt: represents intangible assets. The last term in each speciﬁcation is expanded

additively to include each type of intangible asset: advertising, R&D, risk, and any other
included terms.

Alternatively, some researchers, notably Hirschey (1982) and Doyle (1994)
speciﬁed valuation equations without the log transform. Like Hall, Hirschey used the
ratio of intangible assets to tangible assets as independent variables, while Doyle

refrained from using ratios, instead preferring to apply a Taylor expansion to equation

37

1.5. One advantage to Doyle’s methodology is that it more easily lends itself to working
with ﬁrst-differenced equations. If the time period over which the data extends is of short
duration, as is often the case, it may be difﬁcult to construct a reliable measure of the
stock of intangible capital, and transforming the data through ﬁrst-differencing allows the
researcher to work with ﬂow variables instead of stock variables. FurthermOre, ﬁrst-
differencing allows more ﬂexibility in the treatment of risk.

According to Hall (1998), researchers using United States data on manufacturing
ﬁrms typically ﬁnd that each dollar invested in R&D leads to an additional market
capitalization of between $2.5 and $8, with the stock of R&D valued between 0.5 to 2
times the value of ordinary assets. Despite expectations to the contrary, the addition of
industry speciﬁc dummies does not change the estimates on average. Another interesting
result is that the shadow value of R&D is not stable over time, and is sensitive to the
period examined. For example, the late sixties and seventies was a period of high
valuation of R&D, while the late eighties and nineties was a period of low valuation.
Hall offers two explanations for this: ﬁrst, variations in the value of ordinary assets
caused the value of intangible assets to vary in an inverse manner. Second, beginning in
the 1980’s, declines in the value of R&D assets occurred in high-tech industries where
the returns to R&D are short-lived. Another possibility, which Hall does not consider, is
that ﬁnancial markets value some types of R&D more than other types. Thus, changes in
the composition of R&D over time will change the valuation of intangible assets as a

whole.

38

II. THEORETICAL FRAMEWORK AND MODEL

Geroski (1995) stated that “entry appears to be relatively easy, but survival is not”
(p. 435). Clearly, the ﬁrrn’s ability to adapt to changing market conditions will determine
whether it survives. In my view, an essential component of this process of adaptation is
the efﬁcient allocation of R&D resources between the product dimension and the process
dimension within the constraints of the organization. Organizational design is an
essential ingredient for analyzing innovation. In the theory that follows, the purpose of
deﬁning organization-related concepts is not necessarily to build a model using each
concept, but to provide a reasonable explanation for observed R&D allocation strategies
and the value placed on them by ﬁnancial markets.

This section develops a theoretical model that describes the ﬁrm’s allocation of
innovative resources between product R&D and process R&D. I show that investment in
organizational design is not necessarily without hidden costs or beneﬁts; organizational
strategies in one dimension could interact with strategies in the other to create
extemalities from one dimension to the other. I ﬁrst discuss and develop a general
framework of the model and deﬁne important concepts. Then I develop a corresponding

mathematical theory from which I derive some speciﬁc hypotheses.

39

A. Interactions Between Product and Process Innovation

Recall that Acs and Audretsch (1987) and Cohen and Klepper (1996) discuss the
idea that the returns to process innovation are greater relative to product innovation as the
level of ﬁrm output increases. I refer to this as the appropriability eﬂ‘ect. Speciﬁcally,
the appropriability effect states that a given process innovation will generate more net
revenue than a given product innovation as output increases (when this occurs, I refer to
the appropriability effect as positive -- if the reverse is true, I refer to the effect as
negative). Thus, there are apprOpriability incentives to focus innovative efforts on
process R&D as output increases. However, incentives do not always translate to
behavior. Internal factors necessary for the ﬁrm to capitalize on the appropriability effect
inﬂuence ﬁrm allocation strategies, but have not been explored in previous research. For
example, a product innovation may require different manufacturing techniques than the
ﬁrm currently uses, and impose adjustment costs on the process side. In addition,
organizational factors likely will inﬂuence the effectiveness of innovative efforts in and
across dimensions. For these reasons, the existence of the appropriability effect may not
provide enough incentive for the ﬁrm to allocate R&D resources in accordance with it.

In the model that follows, the basic input that determines the ﬁrm’s ability to
produce commercial innovations, in addition to direct R&D expenditures, is
organizational design. I deﬁne organizational design as a collection of organizational

design practices (ODP’s). An ODP refers to a part of the organization with a distinct

40

form of its own and which has a speciﬁc ﬁrnction or purpose within the ﬁrm. It could be
a training program, a speciﬁc management structure, an R&D center, a ‘total quality
management’ strategy, a team, an operational learning and control system, etc.
Organizational design practices are mechanisms for establishing ﬁrm speciﬁc
capabilities, particularly as they relate to the two innovative dimensions within the ﬁrm --
product and process. A well functioning, well-managed organization comprises a
collection of ODP’S in each dimension, and is highly ﬂexible.25

Anderson and Schmittlein (1984) and Milgrom and Roberts (1990) describe a
tendency for ODPs to be “cluster ” in ﬁrms, thereby providing evidence that many
ODP’s are complementary.“ This is consistent with Athey and Stem’s hypothesis that
product innovation and process innovation are complementary. Another plausible
hypothesis for clustering may be that some ODP’s are more suited for product innovation
than process, and vice versa. Those ﬁrms with a strategic orientation towards process
(product) innovation would naturally tend to have a collection of ODP’s more suited to
process (product) innovation.

It is also possible that ODP’S afford a pathway through which one dimension can
interact with the other. This could happen if those ODP’s more suited to one dimension
are functionally incompatible, or functionally complementary, with ODP’s more suited to
the other dimension. For an example of ODP incompatibility, suppose the ﬁrm has a

product innovation ODP comprised of teams organized around speciﬁc product lines. A

 

‘5 By ﬂexible I refer to how easily ODPs change and respond to outside stimuli. See footnote 4 for
deﬁnitions of ﬂexibility.

41

prod

on

are
existi
CIITB‘.
had.

ﬂexib

subsequent adoption of a process oriented ODP, such as an operational control system,
may interfere with the creative ﬂow of the design teams and limit their effectiveness.
Therefore, to the degree that these ODPs interfere with each other, the efﬁciency of the
overall organization will be impaired, and the ﬁrm will exhibit trade-offs between
product and process innovation. In contrast, complementary ODP’s would have the
opposite effect; they would provide incentives to invest in both dimensions and enhance
the effectiveness of the overall organization.

The existence of an interaction may also affect the ﬁrm’s ability to respond to
external conditions. The more likely it is that potential ODP’s are incompatible with
existing ODP’s, the more limitations placed on the ﬁrrn’s ability to change or augment its
current OD as market conditions dictate, thereby impairing its ﬂexibility. On the other
hand, if ODP’s are overall complementary, then the ﬁrm should have a higher degree of
ﬂexibility, all else equal.

If incompatibilities exist, the ﬁrm conceivably could choose to invest primarily in
a single dimension. An exclusive collection of ODP’s in one particular dimension may
allow the ﬁrm to reap the beneﬁts of their within-dimension complementarities while
avoiding any negative organizational extemalities. This will provide it a more efﬁcient,
albeit one—dimensional, organization while retaining a high degree of ﬂexibility in that
dimension. Of course, there may be adverse revenue consequences down the road.
However, a ﬁrm may augment its internal R&D with external R&D, thus avoiding

negative interactions while maintaining a proﬁtable two-dimensional structure. In

 

26 Athey and Stern (1998) develop a method to test for these relationships.

42

contrast, complementary ODP’s would have the opposite effect; they would provide
incentives to invest in both dimensions and eventually lead to positive revenue
consequences, and provide more ﬂexibility for R&D allocation. The point is that product
innovation and process innovation are not conducted independent of each other. It is
quite possible that interactions between them, via organizational considerations or some
other mechanism, produce either trade-offs or complementarities between dimensions and
affect the optimal allocations of product and process R&D.

In smnmary, it is clear that the appropriability effect will tend to create incentives
for higher process R&D activity relative to product R&D activity as the size of the ﬁrm
increases. However, it also is possible that organizational requirements for successful
product innovation may interact with the organizational requirements for successful
process innovation in one of two ways: in a complimentary fashion that enhances
innovation in each dimension, or as a trade-off that reduces innovation efﬁciency in each
dimension. The implication is that a large enough interaction between product and
process innovation, either positive or negative, may interfere with the appropriability
effect. For example, a trade-off between product and process innovation could induce the
ﬁrm to become one-dimensional; that is, abandon innovation in one dimension altogether
and focus solely on the other, and look to increase external R&D. Alternatively, an
asymmetric spillover from one dimension to the other may signiﬁcantly change the
optimal allocation strategy from one that supports the appropriability effect to one that

supports a negative appropriability effect. The following model, which builds upon

43

Cohen and Klepper (1996), incorporates appropriability incentives and interactions

between product and process innovation into a formal structure.

B. The Model

The model that follows is based on Cohen and Klepper’s (1996) product and
process R&D incentive model. I extend their model to incorporate interactions between
product and process innovation while retaining their basic revenue incentive framework.

The goal of the R&D decision-maker is to maximize proﬁt generated from
innovative output in each innovation dimension i, product and process.27 Suppose at time
t the ﬁrm introduces a new product variant. The product variant will increase the price
buyers are willing to pay for the product, and thus increase the price-cost margin on each
unit of output produced. The introduction of a process innovation also will increase the
price-cost margin, but through a decrease in the average cost per unit of output. The total
revenue generated from either type of innovation will be a direct function of the price-
cost margin times the level of ex-ante output.

Following Cohen and Klepper, let qt denote ex-ante output, measured as the
number of units of output sold. In the product innovation dimension, let ht denote the
fraction of the ﬁrrn’s existing buyers that purchase the ﬁnn’s new product variant, Z, the

units of output produced (and sold) of the new product variant that results from product

innovation, and A D,t the time until the market imitates the product innovation (the
parameter A 1),, reﬂects the level of competition in the industry). Therefore, the total net

revenue generated from product R&D expenditures in time t can be represented as

Vt(Dt) = AD,t(htqt + 29 Ut(Dt) , ' [2.11

where Ut(Dt) represents the price cost margin attained from product R&D expenditures
(D1). The price cost margin is an absolute measure of the increase in the price attained
from the product enhancement. It is the difference, attributed to product R&D, between
the price of the new variant and the price of the old variant. In an industry that was
perfectly competitive prior to product innovation (price equals marginal cost), U(D,)
would be equal to the post-innovation price minus average/marginal cost.28 To reﬂect the
idea that more product R&D expenditures increases the price cost margin but at a
declining rate, I assume U '(D) > 0 and U ”(D) < 0.

In the process dimension, h=1 and 2:0, and thus the revenue generated from

process R&D expenditures can be denoted,

 

’7 In the model that follows, I will abstract from any moral hazard that would provide incentives for the
decision-maker to invest in R&D in a non proﬁt maximizing manner.

2’ I could have deﬁned 2 as a function of ex ante output, q. However, Z is deﬁned as the number of new
customers attracted to the enhanced product that did not previously buy the obsolete version. Although
there may be a relationship, I do not see it as a direct, ﬁrst order relationship, but as an indirect inﬂuence
(neither do Cohen and Klepper). Therefore, the parameter b reﬂects an indirect relationship, and there is
no need to deﬁne Z as a function of q. Furthermore, this speciﬁcation does not detract from the basic
intuition of the model nor affect the directional irrrpact of the results of the analysis below, but it would
complicate the analysis.

45

Vt(Rd = AR,rqut(Rd [22]

where R, represents process R&D expenditures. Analogous to product R&D, I assume
U ’(R) > O and U "(R) < 0. Also analogous to product R&D, A R,t represents the time until
the cost advantage is matched by the industry.

Given 2.1 and 2.2, total ﬁrm net revenue generated from R&D activities at time t,

17;, can be represented as,

17t = AD,t(htqt + Zr) Ut(Dr) + AR,ttht(Rt) - Dr - Rt. [23]

This is the basic revenue-incentive framework used by Cohen and Klepper.29 It implies
revenue incentives are greater for process R&D than product R&D as ﬁrm ex ante sales
increase. However, rather than specify an explicit functional form for U(*), they specify
a general form for its derivatives: U ’(R) = mRR—I/ﬁ and U ’(D) = mDD‘I/ﬂ, where ,6
deﬁnes the rate at which the marginal return to process R&D and product R&D decline.
In other words, the efﬁciency with which R&D is performed is a direct function of the
value of ,6. The parameters mR and mu represent industry level technological
opportunities for process and product R&D. They chose this speciﬁcation for the
following reasons: “it is convenient because it allows ‘R’ and ‘D’ to be solved for

explicitly as a function of q. This makes it possible to characterize the conditions that

46

need to h

 

prim.
Irvin III

R&D. vi

 

Sllllt I'll?

need to hold to obtain the simple intuition suggested above as well as to derive detailed
predictions that allow further testing of the model” (p. 234). Although they allow ,6 to

vary in theory, they impose the restriction that ,6 is equal for product R&D and process
R&D, which implies “the marginal returns to process and product R&D decline at the
same rate as a function of the level of process and product spending” (p. 235).

In many respects, Cohen and Klepper’s speciﬁcation of U(*) is extremely
powerful in its simplicity, but it is still somewhat limited for my purposes for several

reasons. First, they do not consider a possible interaction between product and process

innovation. Second, their assumption that ,6 is equivalent in each dimension is somewhat
suspect. It amounts to the assumption that each dimension conducts innovative activity
with equal efﬁciency. However, as the previous section demonstrated, this efﬁciency is
dependent upon the organizational design of the ﬁrm. Therefore, to the extent that
organization design is asymmetric between dimensions, this assumption is unwarranted.
To extend Cohen and Klepper’s framework to incorporate the organizational
design aspects of innovation, I alter their original speciﬁcation by deﬁning the price cost

margin attained ﬁ'om product R&D as

U(D) = mDDaU +1) , [2.4]

 

2” All variables represent current period and thus the subscript t can be dropped from now on with no
confusion.

47

whit
prod
deter
assun

proce-

u y

where m D represents the level of technological opportunities in the industry in the

product dimension. The parameter or is analogous to Cohen and Klepper’s ,B(D); it

determines the efﬁciency with which the ﬁrm employs its product R&D expenditures. I
assume 0 f m D f 1 and 0 f a f 1. The parameter 2 represents a spillover effect ﬁ‘om

process innovation. This parameter lies between negative 1 and one; -1 < 2 < 1.

Likewise, the price-cost margin function for process R&D can be deﬁned as

U(R) = mRR‘SU +9) [2.5]

where OSmR,6§1and0_<_ 651.

The discussion in the previous section suggests the parameters or and 8 are
functions of organizational design. If 0]) represent the level of organizational design in
the product dimension, then a = (#00). Extending this idea further, assume the

efﬁciency of product R&D increases as the level of organizational design increases, but at
a decreasing rate, i.e., a’>0 and a”<0. Likewise, for the process dimension, 6 = 6 (0R)
and6’>0 and6"<0.

Finally, the parameters 2 and 6 represent the interactions between product and
process innovation and also are dependent upon organizational design; 2:2(0R) and

9=9(OD). Thus, if organizational design increases or decreases in one dimension, there

are spillovers into the next dimension through these interaction terms. To reﬂect the idea

48

that the effect of the spillover into one dimension increases as the level of organizational

design in the other dimension increases, but at a decreasing rate, and that the increase is

 

in the same direction as the initial sign of the term, I assume |2’ >0, l2 ”|<0 and |6’|>0,

|6’ ’|<0. In addition, assume the interaction is equally dependent on irmovation efﬁciency
for each dimension; i.e., 0R = 01) implies 2 = 6 and 2’ = 6’. The purpose of the
interaction terms is to capture the basic intuition behind the organizational design aspects
of irmovation outlined in the previous section, and this assumption in no way detracts
ﬁom this basic intuition. I also make the simplifying assumption that the signs of 2 and 6
are the same. Therefore, 2,6 > 0 implies complementarities between product and process
innovation, 2,6 < 0 implies a trade-off, and 2,6 = 0 implies there is no interaction.”

This speciﬁcation is convenient because it deﬁnes an explicit ﬁrnctional form for
U(*) while retaining the useful properties of Cohen and Klepper’s original speciﬁcation —
i.e., it allows D and R to be solved for explicitly as functions of q and the exogenous
parameters by solving the ﬁrst order conditions for D and R. It also characterizes the

impact of organizational design on the effectiveness of R&D expenditures in a manner

 

’° Fixing the signs of 2 and 6to be the same amounts to the assumption that there is roughly an equivalent
spillover effect on each dimension. For most ﬁrms, I see no reason for this assumption not to hold.
Furthermore, this assumption does not detract from the basic intuition of the model nor affect the
directional impacts of the interactions in the analysis below, but it would complicate the analysis.

49

T1

or

consistent with the intuition outlined in the previous section.31

Equations 2.4 and 2.5 imply that equation 2.3 now can be written as:

17=AD (hq+Z)mDD“(1+") +AquRR5(1+9—D—R. [2.3’]

The relationships between proﬁt, revenue, cost, and each type of R&D are depicted in
Figures 1 and 2. It is clear that the optimal (proﬁt maximizing) levels of R&D occur
where the slope of the revenue curves equal 1, the slope of the total cost curve. The
position of the revenue curve depends on the exogenous parameters in the model. If
A Dm D > A Rm R then industry effects favor product innovation over process innovation,
and total revenue from product innovation will be greater than that from process
innovation, all else equal. Likewise, the ﬁrm speciﬁc efﬁciency and interaction
parameters may favor one dimension over another. In addition, within the same industry,
ﬁrm speciﬁc parameters will determine relative positions of revenue curves. The model
allows for the proﬁt maximizing level of R&D to differ for each type of R&D, and for

total proﬁt ﬁ'om each type of R&D to differ.

 

3' In the analysis that follows, the parameters a, A, 5, and 9 are assumed exogenous to the current period
allocation decision. This assumption may not be realistic in that it removes the ability of the ﬁrm to conﬁol
the interaction. However, it is doubtful that organizational changes occur quickly. Athey and Stern (1998)
consider such organizational changes as long-run decisions. Therefore, in a static framework like we have
here, treating these parameters as exogenous is appropriate. While a static framework limits the power of
the model, it sinrpliﬁes the analysis dramatically, and still allows meaningful results to be derived. A more
comprehensive analysis would endogenize these parameters within the framework of a dynamic model. I
leave this task to future researchers.

50

Revenue (5),
Total Cost (8)

A

 

 

 

 

 

‘
v-‘
v
. '
”
'
p
”
ﬂ

,,,,,,,, V(Dz)
/ ''''' [a(I+4)<5(1+9)]

’

TC
(slope=1 )

V(R)

——————————————— V(DI)

[a(1+2)<6(1+6)]

 

 

’
Process R&D (SR),

Product R&D (8D)

Figure 1. Relationship between revenue, cost, and R&D expenditures.

51

Profi'

Profit (SH)

A

b----——- -----—--—-—

L.— --—-----————

h---‘----

 

 

—>
D*, R* D*2 Pr ess R&D (SR),
Prod ct R&D (8D)

11(1):) H(R) 11(1):)

Figure 2. Relationship between R&D proﬁt and R&D expenditures.

52

An important feature of the model is the dependency between process and product
innovation through the interaction terms. By deﬁnition, if organizational design is
enhanced in one dimension, it increases the efﬁciency of R&D in that dimension, but it
also has an impact on the efﬁciency of R&D in the other dimension. This implies the

following hypothesis:

Hypothesis 1:
The eﬂiciency of product (process) R&D inﬂuences the eﬂiciency (and revenue) of

process (product) R&D according to the Sign of the interaction terms.

To derive further predictions from the model, I ﬁrst need to derive the proﬁt
maximizing levels of product R&D and process R&D. This is determined by setting
marginal revenue equal to marginal cost for each type of R&D and solving for D and R.

The ﬁrst-order conditions for equation 2.3’ with respect to R&D expenditures are:

£11 =17D = a(1+2) m1) A D (hq + Z)D“(1+")-1 _ 1, [2.6]
d)

and
a: =IIR = 5(1+9)mR AR qR5(1+9)-1 — 1, [2.7]
on

To prevent the marginal revenue schedules from increasing, I impose the restrictions that

6(1 +6) <1 and a(1+2)<1. This ensures downward sloping marginal revenue curves.

53

Graphical representations of the marginal cost curve and alternative marginal revenue
curves are depicted in Figure 3. As the graph shows, higher ﬁrm speciﬁc parameters in
one dimension imply higher proﬁt maximizing levels of R&D in that dimension. Thus,
higher optimal spending on product R&D (D*2) versus process R&D imply exogenous
parameters favoring product innovation. Conversely, unfavorable parameters for product
innovation lead to a lower optimal product R&D allocation (D*,). In other words, for an
equivalent amount of R&D, marginal revenue can be different for different types of
R&D. The actual curves will vary from ﬁrm to ﬁrm based on the values of the
exogenous industry and ﬁrm speciﬁc parameters in the model. Figure 3 demonstrates
that if the parameters favor one dimension over the other, the optimal (proﬁt maximizing)
level of R&D will be higher in that dimension.

Setting equations 2.6 and 2.7 equal to zero amounts to setting marginal revenue
equal to l (marginal cost). Doing so and solving for D and R yields the net revenue

maximizing levels of product and process R&D:

19* = [a(1 +2)A 1)»:qu + 2)] 1/(1-a(1+t)) , [2.8]

and

12* = [6(1+6)ARqu] 1/(1-5(1+9)) . [2.9]

54

MR,MC

 

 

 

 

 

 

 

Figure 3. Relationship between marginal revenue and marginal cost.

55

Using this speciﬁcation, I can determine if and under what conditions the
appropriability effect is positive. Recall, the appropriability effects states that the
appropriability of process innovations increases relative to product innovations as ﬁrm
sales increase. That is, larger ﬁrms can generate proportionately more net revenue from
process innovation than product innovation. This implies that net revenue from process
R&D should be an increasing ﬁmction of sales. It also implies that the magnitude of the
effect of sales on net revenue generated from process R&D should be greater than the

effect of sales on net revenue generated from product R&D. Formally,

50743] - 1749]] > 0. [2.10]
361

Equation 2.3’ implies:

17(R) - 11(D) = M}; AR qR 56W - mD AD(hq + Z)D“(1+") . [2.11]

Taking the derivative of this expression with respect to ex ante output q yields,

5mm] — 17(1)]; = mR AR R 5(1+9) - mD AD h D‘WT"). [2.12]
é’q

The case in which we are most interested is the sign of this expression at the optimal

levels of product and process R&D. Inserting R* and D*, the net revenue-maximizing

56

values for process and product R&D, respectively, into expression 2.12 and simplifying

yields,

6(17LR*2-17(D*)) = ARmR[5(1+6)ARqu]6(1+9)/(1-5(1+9)) [2.12’]
é’q

- hADmD[a(1+/1)Apmp(hq+22 ]“(1+")/(1"‘(1+’W

Other than the dimension speciﬁc parameters, the primary difference between each term
in this expression is the inclusion of h in the second (product R&D) term. Since h is
always less than one, this expression will tend to be positive, supporting the
appropriability effect. Therefore, the more successful the product innovation at retaining
existing customers and attracting new ones, the less likely the appropriability effect will
hold. Under the assumptions that all parameters in each dimension are held equal, with
the exception of h and Z, the net revenue curves with respect to ex ante sales for each
type of R&D are depicted in Figure 4.32 As the ﬁgure demonstrates, the difference in the
slopes is determined by h, the proportion of existing accounts retained after product
innovation. The intercept (I) for marginal revenue of product R&D occurs at
mDADZD“(1 +1). The appropriability effect is supported in the region to the right of the
intersection of the two curves. Notice however, if h=1, the curves never cross and the

appropriability effect is never positive.

 

’2 There are two exceptions to this conclusion: ﬁrst, if the exogenous parameters unduly favor the product
dimension, and second, if hq+Z is signiﬁcantly larger than q. Either of these situations could make the
second term in equation 2.12’ greater than the ﬁrst. I exarrrine these situations in more detail below.

57

H(R) 9 11(1))
A

/

 

 

fun): (h=1)

/
/

TI(R)

    
 

      
 

 

(D): (0<h<1)
___________________________________ I1(D): (h=0)
Negativ Appropriability Positive Appropriability
Effect Effect
+
q

Figure 4. Relationship between marginal net revenue (proﬁt) and ex ante sales for each

type of R&D assuming all parameters

except h equivalent in each dimension.

58

Based on the above analysis, one would fully expect the ﬁrm’s allocation strategy
to reﬂect the appropriability effect. In other words, if the appropriability effect exists and
is positive, one would expect the expenditures on process R&D to rise with ﬁrm sales at a
rate greater than that of product R&D expenditures. Therefore, another way to determine
if the appropriability effect exists at the Optimal levels of R&D expenditures is to
examine the relationship between the ﬁrrn’s optimal allocation strategy and ex ante sales.
Deﬁne process intensity, P, as the ratio of the optimal level of process R&D (R*) to the

optimal level of product R&D (D*):

P = [anemm WW“) . [2.13]
[a(1+/I)ADmD(hq + 2)] ”""’”*"”

A positive derivative to this expression with respect to q implies that process intensity
rises as ﬁrm size increases, and the ﬁrm’s allocation strategy reﬂects a positive
appropriability effect. In contrast, a negative derivative with respect to q would imply the
opposite, and a derivative equal to zero would imply the optimal allocation strategy is
independent of ﬁrm size. Figure 5 depicts the shapes of the optimal R&D expenditure
curves with respect to sales. At zero ex ante sales, it still is possible to have positive
optimal product R&D. Generally, the slope of the process R&D curve is larger than that
of the product R&D curve, and all else equal, it has more curvature. Figure 6 depicts the
relationship between the marginal change in optimal (proﬁt maximizing) R&D
expenditures and ex ante sales assuming all exogenous parameters are equal to their

counterpart in other dimension.

59

R*,D*

 

 

R9:
Di:

 

Figure 5. Relationship between optimal (proﬁt maximizing) R&D expendihrres and ex

ante sales.

60

BR‘Vaq, 6D*Iaq

 

 

 

 

A
Negative Process ] Positive Process 6R*I6q
Intensity wrt Sales 5 Intensity wrt Sales /
§ aD*/aq
5 >
q

Figure 6. Optimal allocation strategy with respect to sales.

61

After taking the derivative of this expression with respect to q, one derives the

following condition that must hold for the derivative of expression 2.13 to be positive.

(1 + Z/hq) > (1- 6(1+6))/(1-a(1+2)) . [2.14]

(The proof is provided in appendix A). When this condition holds, ﬁrms allocate
resources in accordance with a positive appropriability effect.

Expression 2.14 has several interesting implications with regard to the optimal
allocation strategy. To simplify things for the analysis below, assume the interaction
terms are equal; i.e., 6 = 2. This assumption means that the spillovers ﬁ'om one
dimension to another are symmetrical. This has an implicit assumption that
organizational design levels in each dimension are roughly equivalent. Although this
may be a strong assumption, it allows me to isolate the impact of ﬁrst-order changes in
the other parameters, and still allows for meaningful analysis of interactions in general.

In expression 2.14, notice that the sign does not depend on industry level
parameters Ai and mi. This implies that industry effects are not critical in determining the
optimal allocation strategy with respect to ﬁrm size. In addition, unless Z=0, the left-
hand-side of the equation is always greater than one. Therefore, this expression will
always hold if 6(1+6) 3 a(1 +2). Since we’re assuming the interaction effects are
roughly equivalent, this implies positive appropriability effects when the process R&D
efﬁciency parameter is greater than or equal to the product R&D efﬁciency parameter (in

this case the marginal return schedule for process R&D will be larger and less steep than

62

 

the schedule for product R&D). Although this is a sufﬁcient condition, it is not a
necessary condition. Even an efﬁciency parameter for product R&D greater than that of
process R&D does not preclude a positive appropriability effect. Therefore, I expect the

appropriability effect should be positive in general.

Hypothesis 2:
As ﬁrm size increases, the marginal returns to process R&D will be larger than the

marginal returns to product R&D at the optimal allocation.

It is clear that the appropriability effect will be zero or negative if and only if
5(1 + 6) 3 a(1 +2). What is the minimum level of a for which the appropriability effect

remains positive? Rearranging equation 2.14 yields:

a(1+2) < 1-@q{1-6{1+6)l. [2.14’]
[M + Z]
If this condition does not hold, then the appropriability effect is not positive. First, note
that as Z increases, this condition is more likely to hold. Therefore, the more new
customers obtained by the product innovation, the more likely the appropriability effect
Will be positive. Next, examine the term hq+Z. This term represents the number of
CUStol'ners retained after the product innovation, plus the number of new customers

acquired via the product innovation. Therefore, it represents total demand in the current

period. If I let g represent the growth rate of sales in the current period, then hq+Z s

63

(1+g)q. That is, demand in the current period will equal demand in the previous period

times the growth rate. Solving for Z and inserting the result into equation 2.14’ yields

a< 1+g+h(6(1+Q-1). [2.14”]
(1 +2) (1 +g)
This expression states that the likelihood that the appropriability effect is positive or
negative can be determined by examining the following factors: the efﬁciency parameters
of product and process R&D, the interaction terms, the proportion of existing customers
retained, and the growth rate of sales.

Any change in the parameters on the right-hand-side of condition 2.14” that
increases the RHS will support a positive appropriability effect, whereas any change to a
parameter that decreases the RHS will support a negative appropriability effect. It can be
shown that as the efﬁciency parameter of process R&D ((2 increases, the RHS increases
(as one would expect). In contrast, as the proportion of existing customers retained (h)
increases, the RHS declines, which does not support a positive appropriability effect (as
we concluded from equation 2.12’). In addition, higher growth rates for sales (g) implies
a smaller RHS. If 6 ¢ 2, then larger spillovers onto process R&D will increase the RHS
whereas larger spillovers onto product R&D will decrease the RHS. However, the effects
of the interaction terms are somewhat more complicated when we assume 6=2, or the
somewhat weaker assumption that the interaction terms move concurrently. It can be
shown, under the assumption that the interaction terms move concurrently, that the

condition h<(l+g) ensures the RHS will increase as the interaction terms increase. This

64

condition should always hold. The above analysis implies the following propositions:

Proposition 1 :

A positive appropriability effect is less likely to exist when the ﬁrm retains a high

proportion of its ex ante sales.

Proposition 2:

A positive appropriability eﬂect is less likely to exist in the presence of high sales growth.

Proposition 3 :
A positive appropriability effect is more likely to exist in the presence of trade-oﬂfs than

in the presence of complementarities between product and process R&D.

65

C. Summation of Hypotheses/Propositions

Hypothesis 1 :

The eﬁ‘iciency of product @rocess) R&D inﬂuences the eﬁ‘iciency (and revenue) of
process @roduct) R&D according to the sign of the interaction terms.

Hypothesis 2:

As ﬁrm size increases, the returns to process R&D will increase relative to the returns to

product R&D.

Proposition 1:

A positive appropriability effect is less likely to exist when the firm retains a high
proportion of its ex ante sales.

Proposition 2:

A positive appropriability eﬂect is less likely to exist in the presence of high sales growth.
Proposition 3:

A positive appropriability effect is more likely to exist in the presence of trade-oﬂfs

between product and process R&D than in the presence of complementarities.

66

III. EMPIRICAL FORMULATION AND RESULTS

This section applies empirical analysis to the above hypotheses. First, I summarize my
data sources and structure. Then, 1 specify the empirical models for each hypothesis and

estimate them using statistical methods.

A. Data

This study uses a relatively recent set of survey panel data developed by the
Industrial Research Institute/Center for Innovation Management Studies (IRI/CIMS)
ﬁ'om their Annual R&D Survey. The survey encompasses a wide array of information on
R&D expenditures from a large cross section of industrial ﬁrms. One hundred six
corporations participated in the survey, collectively accounting for a large percentage of
formal R&D in the United States. The information was collected at the ﬁrm, industry
segment, and laboratory level beginning in 1992. The data used for this study is annual
information covering ﬁscal years 1992 through 1997, reported at the ﬁrm level. It
contains information on total ﬁrm R&D expenditures and on the type of R&D
undertaken. This includes product R&D, process R&D, basic research expenditures,
applied research expenditures, support R&D, and external R&D. The data also includes
ﬁgures on annual sales.

In addition to the survey data, Compustat market valuation data was appended for

each ﬁrm in the survey. This includes each ﬁrm’s stock market value at the end of ﬁscal

67

year, the value of its tangible assets (book value), annual advertising expenditures, and
annual sales. Compustat also supplied total annual R&D expenditure data for each ﬁrm
in the [RI survey. In all but a small number of cases, ﬁgures for R&D expenditures and
sales as reported from each source tied-out to within a small degree of error.

A number of ﬁrms surveyed did not provide complete responses in any given
year, and many did not respond at all in some years. Therefore, the survey data contains
a number of missing data points. This raises the question of whether the non-responders
were chosen randomly. If not, estimates of coefﬁcients on any regression will be biased.
Fortunately, Compustat supplied information on total R&D, sales, market value, and
asset value for many of the missing and non-missing observations in the survey. A
comparison of each variable from each source revealed, with one exception, a similar
distribution for data reported in the survey with that of data not reported in the survey but
supplied by Compustat. The one exception is total R&D. The average value of R&D
expenditures supplied by Compustat is 26.5% higher than survey R&D. This may
indicate that ﬁrms with higher R&D expenditures were less inclined to respond to the
survey, which would imply that estimated coefﬁcients on R&D-type expenditures using
survey data alone might be biased downward. However, the survey data includes ﬁrms
with a wide range of R&D expenditures, with a standard deviation twice as large as the
sample mean. This suggests the sample of survey R&D expenditures adequately
represent the R&D expenditures of ﬁrms in general. Summary statistics of survey
responders and non-responders are depicted in Appendix E.

In order to arrive at a consistent sample data set to use for each analysis below, a

68

number of adjustments were made to the original data. First, obvious reporting errors
were corrected. Second, two ﬁrms were omitted from the sample because of reported
average sales in excess of $100 billion, which were nearly twice as high as the next
highest sales ﬁgures. Omitting these two ﬁrms considerably mitigated heteroskedasticity.

Third, advertising expenditures were missing for a large number of ﬁrms for some
or all years. Therefore, instead of excluding these observations, I employed an
instrumental variables approach to estimate the missing values. Advertising expenditures
were predicted based on the ﬁrm’s sales, total R&D expenditures, and tangible assets.
Sales was chosen as a variable because it is assumed that the purpose of advertising is to
increase brand awareness and therefore sales. Total R&D expenditures and tangible
assets were chosen based on Doyle (1994, p. 86), who employed a similar technique to
estimate advertising expenditures. First, I run regressions including only those ﬁrms

reporting advertising expenditures using the following speciﬁcation:

ADV*i,t = 0'13: + ,3] Sale-St: + '52 Sale-92m + ’63 RDi,t + .54 TAi,t+ 6i,t , [11}

where ADV“ represents advertising expenditures of ﬁrm i at time t, Sales“ represents
sales revenue of ﬁrm i at time t, Sales2 1"; is sales revenue squared, RD“ represents R&D
expenditures of ﬁrm i at time t, and TA i, t represents tangible assets of ﬁrm i at time t.
Then, I use these estimated coefﬁcients to compute a ﬁrm’s predicted advertising
expenditures for those ﬁrms not reporting advertising expenditures. I also tried the

following speciﬁcations:

69

ADV}; = at: + .31 Salem + .32 501852 + .63 RDi,t + [34 TAi,t + 81;: . [3-1’]

ADVi,t = ai,t + ﬂ! 501631;: + .32 RDi,t + 6‘i,t , [31”]

Results did not differ signiﬁcantly, although I used speciﬁcation 3.1 because it yielded a
moderately higher adjusted r-squared.

Fourth, 13 ﬁrms did not report disaggregated R&D information for any year.
Although employing an instrumental variables approach to estimate the missing values as
I did with advertising expenditures was considered, there were not enough instruments
available to estimate them all. Therefore, these ﬁrms were excluded. Furthermore, four
ﬁrms were excluded because they reported disaggregated R&D information for only one
year of the survey and reported zero process R&D expenditures for that year and zero
expenditures for at least one other R&D type. This is signiﬁcant because less than 5% of
all ﬁrms surveyed with multiple observations reported zero process R&D expenditures in
each year they participated in the survey. In addition, only a small minority of ﬁrms
reported zero expenditures for more than one type of R&D in any given year. Given this,
I ﬁnd it extremely likely that these four observations are unreliable. Finally, although the
majority of ﬁrms were matched to their Computat market valuation for each year 1992-
1997, a small number did not have market value information appended for either 1992 or
1997. Therefore, since the primary unit of observation in each analysis below is the ﬁve-
year average for each variable included in the analysis (see the next section), a

computation of the average market value would be biased for those ﬁrms that do not have

70

market values recorded in the ﬁrst and last years of the survey. Omitting these ﬁrms
ensures the average market value for each ﬁrm in the regression is computed using the
exact same time frame. Finally, a number of ﬁrms did not have R&D information for
each year of the survey. Some ﬁrms were excluded based on more than one of the above
criteria. However, R&D information is extremely stable from year to year for the
overwhelming majority of ﬁrms. Therefore, to preserve as large a sample as possible,
these ﬁrms were not excluded. What remained was a sample of 71 ﬁrms. Tables 1, 2a,

and 2b furnish summary information for relevant variables before and after exclusions.

71

 

Table 1. Stunmary Statistics: Before Independent Variable Based Exclusions ($MM)

 

 

 

 

 

 

 

 

 

 

 

 

# Mean Std Dev Minimum Maximum
Obs

Market Value 83 $10,816 $15,574 $61 $73,929
Total R&D 83 $268 $429 $4 $2,251
Adverﬁsmg 83 $482 $623 $79 $3 157
Expenditures ’
Sales 83 $8,654 $1 1,461 $6 $63,554
Tangible
Assets 83 $3,038 $3,776 -$322 $17,630
(Book Value)
Process R&D 75 $32 $49 $0 $200
Product R&D 76 $150 $306 $0 $1,624
Applied R&D 75 $55 ‘ $1 14 $0 $634
Basic R&D 77 $9 $25 $0 $134
Technical
Service R&D 77 $27 $47 $0 $255

 

 

 

 

 

 

Notes: Computed from average ﬁrm data over 6 years of survey. All survey ﬁrms
with valid reported aggregate R&D information are included. Only ﬁrms with non-
missing market value data in years 1992 and 1997 are represented.

72

 

 

Table 2a. Summary Statistics — Sample Data after Exclusions ($MM)

 

 

 

 

 

 

 

 

 

 

 

 

# Mean Std Dev Minimum Maximum
Obs

Market Value 71 $9,500 $12,407 $61 $51,681
Aggreg. R&D 71 $225 $362 $4 $1,526
Adve‘ﬁsmg 71 $467 $583 $79 $3111
Expenditures ’
Sales 71 $7,991 $10,593 $6 $63,554
Tangible
Assets 71 $2,869 $3,601 -$322 $17,630
(Book Value)
Process R&D 71 $33 $50 $0 $200
Product R&D 71 $ 124 $250 $2 $ 1,400
Applied R&D 71 $49 $1 10 $0 $634
Basic R&D 71 $9 $25 $0 $134
Technical
Service R &D 71 $28 $48 $0 $255

 

 

 

 

 

 

Notes: Computed from average ﬁrm data over 6 years of survey. All survey ﬁrms
with valid reported disaggregated R&D information are included. Only ﬁrms with

non-missing market value data in years 1992 and 1997 are represented.

73

 

 

Table 2b. Summary Statistics by Year-Sample Data ($MM)

 

 

 

 

 

 

 

 

 

 

 

 

 

1992 1993 1994 1995 1996 1997
Average Aggreg.
R&D Exp. $212 $231 $219 $235 $237 $257
Average
Advertising Exp. $332 $337 $358 $423 $488 $502
3:1:ng Market $6,824 $7,255 $7,407 $9,210 $11,652 $14,916
Average Firm $7,594 $7,494 $7,578 $8,322 $8,859 $8,776
S1ze ($Sales)
Average Ratio
R&D/Sales 5'7 4-9 4-2 3.8 3.8 3.9
Tangible Assets $2,726 $2,657 $2,862 $2,912 $3,098 $3,166
Average Product $104 $124 $103 $98 $106 $191
R&D
Average Process
R&D $29 $44 $29 $28 $34 $38
Average Applied
Research R&D $27 $32 $31 $50 $41 $70
Average Basic
Research R&D $8 $12 $1 1 $9 $10 $15
Average
Technical $37 $38 $37 $28 $30 $30
Service R&D

 

 

 

 

 

 

 

Notes: Computed from ﬁrms included in ﬁnal sample dataset (71 ﬁrms).

74

 

B. Estimation and Results

1. Empirical Framework

The following analysis derives the basic empirical equation used to estimate the
returns to innovative activity. It is based on a stock market valuation approach. The
stock market value of the ﬁrm is an excellent proxy for long-run expected proﬁt because,
based on traditional ﬁnancial analysis, it is assumed that the market accounts for all
relevant information about the ﬁrm’s expected future proﬁtability, including the ﬁrm’s
investment in R&D. I follow common practice and use R&D expenditures as a proxy for
innovative activity. Therefore, the stock market valuation approach relates R&D
investment to the stock value of the ﬁrm. If the stock response to R&D is positive, the
market should believe that the expected proﬁtability of R&D investment is positive.

In the market valuation model, the ﬁrm’s problem is to maximize the discounted
value of its future income streams (proﬁts) while accounting for risk. Its market value is
a function of the assets it holds, both tangible (TA) and intangible (IA). Let V(*)

represent the ﬁrm’s value ﬁmction. Generalizing equation 13’

16512344,, 42. ......) . [13’]

yields,

75

Vi,t =fi,t(TAi,t»1Ai,t: Riskw) . [3-11

Tangible assets refer to physical capital, which can be measured as the book value of all
assets held by the ﬁrm. Intangible assets comprise knowledge stock and marketing stock.
Risk is represented by the ﬁrm’s beta. Therefore, I can derive the following expression

for the value of the ﬁrm:
Vi,t =f( T4131. 4451',» K5231. berattl). [32]

where KS represents knowledge capital, and MS marketing capital.
As discussed in Chapter 1, the true functional form of equation 3.2 is unknown.
For reasons outlined below, I choose to follow Doyle (1994) and use a linear

approximation and apply a linear Taylor expansion to equation 3.2. This yields
Vi,t = ﬁo + ﬂ] T413: + ﬂz M521: + .63 K513: + ﬂ4 b81013: + at + n + 5i,t- [3-31

The coefﬁcients on each asset variable represent the value response to a change in the
level of that asset. The parameter it captures any time series effects that may inﬂuence

affect market valuations of all ﬁrms equally across time. The parameter ai represents
potential unobserved heterogeneity that may result from variables conceivably omitted

from the model, and can be considered an individual effect for each ﬁrm cross section. If

the model is perfectly speciﬁed, at will be zero and will drop out of the equation. If the

76

model is not perfectly speciﬁed, as is likely the case (few models are) then a central issue
is the treatment of (1,, the individual effect from each ﬁrm cross section. If one uses
longitudinal data, as I do in this study, the statistical treatment of the individual cross
section effects is crucial to the analysis.

Unobserved heterogeneity can arise ﬁom many sources. For instance, the
population of interest in this study is all ﬁrms who innovate. However, the sample being
used may not be representative of this population. The sample excludes many innovating
ﬁrms based on several factors: ﬁrms who do not perform formal R&D, ﬁrms not in the
survey, ﬁrms not operating on well behaved ﬁnancial markets, and ﬁrms in the survey but
not reporting complete information (these omissions may indicate self selection bias).
Another, perhaps more important, source of unobserved heterogeneity is the omission of
a relevant variable. For instance, the market may include information in its valuation
equation that is not accounted for in the variables in equation 3.3. Indeed, some
speciﬁcations in the literature have speciﬁed a different and/or broader concept of
intangible assets. In addition to the variables speciﬁed in 3.3, they have included one or
more of the following: earnings, market share, industry concentration (C4), debt, patents,
and ﬁrm growth rate. My speciﬁcation assumes most of the explanatory power ﬁom
these other variables reside in the variables included in 3.3. However, it is possible some

information is not accounted for by the included variables, and thus reside in the ﬁxed

effect parameter ai.

If these unobserved effects are constant over time, and unrelated to the other

variables in the model, then ai effectively can be ignored, and pooled ordinary least

77

squares will provide consistent and efﬁcient estimates. However, more than likely the
effects are related to the other variables in the model, in which case a,- must be dealt with
in another way.

Several statistical techniques are available to estimate equation 3.3. The primary

focus of each method is to transform speciﬁcation 3.3 into one that eliminates the ﬁxed
effect parameter can One common method of doing this is to ﬁrst difference the equation.
If the effects are constant over time, the ﬁxed effect parameter a,- will drop out of the
equation. One advantage to Doyle’s methodology is that equation 3.3 can be ﬁrst-

differenced easily. Doing so yields,

AVi,t = ,5] ATAz‘,t + .32 AMSi,t + ,53 AKSi,t + 13445810“ + Al’t + Vat, [3-4]

where A refers to annual changes in each asset. In addition to the elimination of the ﬁxed

effect parameter ai, ﬁrst-differencing equation 3.3 has several other beneﬁts. First, it
alleviates problems pertaining to the creation of appropriate measures of intangible
capital. In equation 3.3, because marketing and knowledge capital are measured as
stocks, “the amount of knowledge (capital) held by the ﬁrm prior to a given point in time
is unobservable, and therefore the total amount held an any particular time is therefore
unobservable” (Doyle, p. 62). Therefore, using changes in knowledge and marketing
stock in any given year, represented by the amount of R&D and advertising investment in
that year, avoids this problem. In the absence of ﬁrst-differencing, marketing and
knowledge stocks would need to be constructed from current and past observations of

78

advertising and R&D. This involves inferring depreciation and growth rates of intangible
capital ﬂows, which adds another level of complexity to the estimation process and
introduces more opportunity for error.33 A related issue involves the relatively short
history of the dataset, which impairs the construction of reliable measures of the stock of

intangible capital.34 One drawback to ﬁrst differencing is that it assumes the ﬁxed effect

is invariant over time. If there is a time element to (If (such that it becomes a“) then
ﬁrst-differencing will not eliminate it from equation 3.3, and the coefﬁcients will be
biased.

Before proceeding, it is necessary to discuss the treatment of risk in equation 3.4.
Risk is an important issue when valuing assets with the stock market. Although the beta
coefﬁcient is used as a proxy for systematic risk in equation 3.4, it is unavailable for
estimation purposes. However, research has indicated that with regard to the beta, “the
stability increases as the length of the estimation period increases — betas are relatively
stable for periods of four or more years” (Doyle (1994) p. 62-3). This implies that over
the course of 4 or more years, the average change in the beta is expected to be zero. For

this reason, I make one further transformation to 3.3’. I average each differenced variable

in the model over ﬁve years. This results in estimators ,Bj, sometimes referred to as

 

’3 For instance, one such method would be to assume knowledge stock equal to current R&D expenditures
divided by the sum of the depreciation rate and growth rate; STOCK = R&Dt /(6+g). For more details on
estimating capital stocks from intangible flow variables such as advertising and R&D, see Hall (1990).

79

between estimators (W ooldridge, 1995) that use only variation between the cross section

observations. Essentially, it is the OLS estimator applied to time-averaged data:

AV“; = Ar“ + ,6] ATA *1: + ,62 AMS“; + ,83 AKS*,' + ,B4Abeta*i + vi, [3.5]

where the ‘*’ refers to 5-year averaged ﬁrst-differences. The advantage to speciﬁcation
3.5 is that the average change in the beta over ﬁve years is expected to be zero, and thus
the beta drops out of the equation.35 Furthermore, the averaged data may alleviate the
impact of extreme values that are often associated with survey data. After dropping the

beta from the equation, 3.5 becomes:

AV*,~ = Art + [7] ATA *,- + p2 AMS*,- + p3 AKs*,- + vi, [35’]

Unfortunately, if beta is not stable over the 5 year period and it is omitted, then
equation 3.5’ will be misspeciﬁed, and to the extent that beta is correlated with the
variables included in the model, the estimated coefﬁcients of the included variables will

be biased. One would expect the beta to be most correlated with R&D expenditures

 

3‘ Another common treatment for unobserved heterogeneity is to use “ﬁxed effects” estimation. This
technique was employed by Ben Zion (1984), Hall (1993), and Johnson and Pazderka (1993). Instead of
first-differencing, this method essentially demeans each cross section by estimating the difference of each
variable from its cross section mean. The advantage to this method is that it effectively uses time series
data and maximizes degrees of freedom in the regression. However, it still does not correct for time-
dependent unobserved heterogeneity. Furthermore, it would be necessary to derive intangible capital

stocks.

80

because they are subject to great uncertainty. However, as Doyle points out (p. 63), it is
more likely that R&D intensity, as measured by R&D expenditures as a percentage of
ﬁrm size, would be correlated with beta rather than the absolute level of R&D
expenditures themselves. However, I ﬁnd no evidence that the average change in R&D
intensity is signiﬁcantly different than zero.36

Because ﬁrst-differencing the data redeﬁnes marketing and knowledge stock in
terms of annual changes, I can use annual R&D expenditures and annual advertising

expenditures as proxies. Substituting these into equation 3.5’ yields

AV*,‘ = 217* + ,6] ATA *i +ﬂ2ADV*i + ﬂ3RD*,' + vi . [3.6]

This is the basic equation used to estimate the retums to innovation. In this expression,
AV*,° represents the average annual change in market value over ﬁve years for ﬁrm i,
AT A *i represents the average annual change in the book value of the ﬁrm over ﬁve years
for ﬁrm i, ADV*,- equals average annual advertising expenditures over 5 years for ﬁrm i,
and RD"‘i represents average annual R&D expenditures over 5 years for ﬁrm i. As is
customary, an intercept term is included.

The change in knowledge stock, or R&D expenditures, can be disaggregated

ﬁrrther into various types of R&D investment. Let ‘Process’ represent process R&D,

 

’5 Because the beta is omitted, ﬁxed effects estimation would treat it as an omitted variable and add the
variation of beta to the ﬁxed effect parameter ai. For comparison, Appendix C contains ﬁxed effects
estimates for each regression below.

81

‘Product’ represent product R&D, ‘Applied’ denote applied R&D, ‘Basic’ denote basic

R&D, and ‘Service’ represent support service R&D. Equation 3.6 now can be expressed

as:

AV = Ar“ + a1 ATA“ + a2ADV* + a3 Process* + a4 Product“ [3.7]

+ a5 Basic“ + a6 Applied* + a7 Service* + e.

(The ‘i’ subscripts are dropped for convenience.) Again, each variable represents a ﬁve-
year average annual change. In this expression, each coefﬁcient represents the stock
response, or the marginal value, of an additional unit of input.

The data includes ﬁrms from multiple industries, both manufacturing and
nonmanufacturing. Compustat-provided SIC codes range from 1000 to 8700. Therefore,
accounting for industry effects is an important issue. In the analysis below, industry
dummy variables were included to account for industry effects. However, preliminary
analysis indicated that industry terms were not signiﬁcant, nor did they signiﬁcantly alter
the coefﬁcients of other included variables. Therefore, to preserve degrees of freedom
and alleviate collinearity, I omitted industry identifying variables from the ﬁnal
regressions .

Another issue often present in ﬁrm level cross sectional data is heteroskedasticity

related to ﬁrm size. Heteroskedasticity occurs if there is a large disparity between the

 

3" The average annual change in R&D intensity as a percent of the average R&D intensity is very close to
zero (1.0%).

82

largest and smallest observations in the sample dataset. In this case, the error term in an
ordinary least squares regression will have different variances associated with the
different sizes of the observations. F inn size in the sample dataset summarized in Table
2a, as measured by sales revenue, ranges ﬁom $5.7 million to $63.5 billion, a signiﬁcant
disparity. Therefore, heteroskedasticity is likely to be present in the analysis. For each
regression, a Park test was used to test for heteroskedasticity, which consists of ﬁrst
running an ordinary least squares regression, and then regressing the residuals against the
suspected proportionality factor (either sales revenue or tangible asset value). If
heteroskedasticity was detected, a weighting scheme was applied to correct for it. Then,
the Park test was rerun on the weighted regression to validate the weighting scheme. The

weighting scheme is described in detail in Appendix B.

2. Preliminary Analysis

Grabowski and Mueller (1978) estimated after tax returns on R&D expenditures
of between 15 and 20 percent based on depreciation rates of between 5 and 20 percent.
Research that is more recent has used a depreciation rate for R&D capital stock of 15
percent. Other research has estimated that R&D should be capitalized by the market at
between 2.5 to 8 times the investment (Hall, 1998), with most estimates centered at 5 to

6. That is, a dollar invested in R&D should increase the market value of the ﬁrm by

83

between $2.5 and $8. These ﬁgures are based on the use of ﬂow-type variables to
represent intangible capital, i.e., R&D expenditures per year. Alternatively, knowledge
capital stocks can be constructed from current and past R&D expenditures and used in
place of the ﬂow variable. These knowledge capital stocks typically are 4-5 times higher
in magnitude than the corresponding ﬂow variable, and thus yield coefﬁcient estimates
approximately 4-5 times lower than the ﬂow variable coefﬁcients. This is consistent with
an average annual return of 20 percent on R&D capital investment.37 Another empirical
result is that the stock of R&D is valued at between 0.5 to 2 times the value of tangible
assets.

Before estimating the hypotheses derived in Chapter 2, estimation of equation 3.6
will provide a basis for comparing the returns to innovation using my sample to earlier

results. The results of regressing equation 3.6,

AV*,~ = Ar“ + ,6] ATA *,- +ﬂZADV*,- + ,B3RD*,- + vi. [3.6]

are summarized in Table 3.

 

’7 From Chapter 1, market value is deﬁned as the net present value of current and future income.
Therefore, net present value of a $1 investment in R&D, assuming constant annual returns, is equal to
I/(1+k)+I/(l+k)2+I/(1+k)’+I/(l+k)‘+... +I/(l+k)T, where I represents per period income (cash ﬂow)
generated from the R&D investment. In the limit, this simpliﬁes to Mt. Therefore,
V(RDS)=NPV=I(RDs)/k, where RDs represents knowledge stock. A simple computation for knowledge
stock used by Hall (1993) is RDS=RD/(6+g) where 5 represents the R&D depreciation rate and g
represents the R&D growth rate. This suggests V(RD5)=I(RD,)/k(5+g). Following Hall, assume a 15%
depreciation rate and a 5% growth rate. Also assume a required rate of return of 20%. Inserting V=$5,
k=.2, 5=.15 and g=.05 and solving for I yields an annual nominal cash ﬂow/income of $0.2 from a $1 R&D
investment, 3 reasonable cash flow. Note that a larger depreciation rate will result in a higher annual cash
ﬂow.

84

Table 3, regression ii, indicates that the stock response to R&D is 4.7, implying
that a dollar invested in R&D is worth $4.7 to the market value of the ﬁrm. This ﬁgure is
well within the range of previous research. In addition, estimates of tangible assets
between 1 and 2 are also within the expected range. A note of caution; the regression of
equation 3.6 used the standard dataset of 71 observations. If I add back the observations
excluded because of missing disaggregated R&D, I obtain an expanded dataset of 83
observations (the combination of survey data and Compustat data resulted in a fewer
missing observations for total R&D than for disaggregated R&D).

The results of regressing equation 3.6 with this expanded dataset indicate a
smaller stock response to R&D, approximately 3.8 (regression iv), although still within
the range of previous research. The response to tangible assets is moderately higher than
before, and the stock response to advertising expenditures is almost unchanged. These
results add conﬁdence to my chosen statistical methodology. For further comparison, I
ran ordinary least squares estimates on ﬁrst-differenced variables for individual years.

The results of these regressions can be found in Appendix D.

85

 

Table 3. Value of Aggregate Assets

 

Ordinary Least Squares Analysis

 

 

 

 

 

 

 

 

Regression
Standard Data Expanded Data
(# Obs=71) (# Obs=83)
Dependent . . . .
Variable Unwerghted Weighted Unwerghted Weighted
1 11 111 IV
Interc t -400.55 -909.73 -390.92 -349.33
61’ 265.78 356.00 266.80 341.07
ATangible 2.048 b 1.234 0 1.590 b 1.448 a
Assets 0.815 0.669 (0.624) 0.513
5.577 a 4.663 b 4.845 a 3.790 a
“”31 R&D 0.589 0.557 (0.515) 0.433
1.395 a 2.338 a 1.707 a 2.049 3
Ad“ ExPend' 0.495 0.226 (0.398) 0.152
F Value: F Value: F Value: 66.68 gage;
52.11(1 131.72 a a 2121:1223
Adj R2: 0.687 Adj R2: 0.740 Adj R2: 0.706 01823'

 

 

 

 

 

 

" Signiﬁcant at the .01 conﬁdence level.
b Signiﬁcant at the .05 conﬁdence level.
° Signiﬁcant at the .10 conﬁdence level.

 

Notes: Standard errors below coefﬁcients. For each regression, the dependent variable

is the S-year average annual change in market value.

heteroskedasticity-corrected.

86

Weighted statistics are

 

Although numerous studies have examined the returns to aggregate R&D, to my
knowledge no study has examined the stock value of disaggregated R&D. The results of
regressing equation 3.7, both weighted and unweighted, are summarized in Table 4. The
weighted results indicate that the stock response to product, applied, and basic R&D are
signiﬁcantly different from zero. However, the basic R&D response is negative. The
stock responses to process R&D and technical service R&D are not signiﬁcantly different
from zero. Applied R&D appears to be valued the most by the market; a $1 expenditure
indicates an increase in market value of approximately $15. In contrast, basic research is
valued the lowest, with an expected change in market valuation of $-29 for each $1
invested in basic R&D.

These results paint a somewhat different picture than the results from regressing
aggregate R&D. The expectation is that a rational ﬁrm will invest in each asset to the
point where the marginal value of each investment is equivalent. However, this does not
appear to be the case with regard to investment in different R&D types. It appears basic
research and process R&D are not valued by the market, whereas product R&D and
applied research are highly valued. What might explain these disparities? One
possibility is that basic research is considered riskier than later stages of innovation and

the market generally discounts basic research more than other types of

87

 

intangible assets. This also raises the possibility that unobserved heterogeneity related to
risk is much more correlated with basic research and process R&D than other variables,
and thus the coefﬁcients are biased. Therefore, it is possible that an omitted variable is
causing bias. One candidate for an omitted variable is the growth rate of sales and
industry speciﬁc indicators. However, the inclusion of these variable did not
signiﬁcantly alter the results.

Another possibility is that basic research is more characteristic of a public good,
and therefore ﬁrms may have difﬁculty appropriating the returns from basic research.

Finally, reporting errors may by inﬂuencing the results. Public policy measures
such as the R&D tax credit provide incentives for ﬁrms to report non-R&D investments
as R&D, and many government contracts are associated with basic R&D. Therefore, it is
quite possible that R&D expenditures are overstated, or the cost of investing in basic
research is less than the cost of investing in other types of R&D because the government
reimburses basic research expenditures.

Comparing process R&D and product R&D, it is evident that product R&D is
valued more highly by the market. This is not surprising from the standpoint that the
majority of ﬁrms invest more heavily in product R&D than process R&D. As Table 23
indicates, ﬁrms invest on average 4 times more in product R&D than process R&D.
Applied R&D is the next most heavily invested type of R&D, and thus its high valuation
is not surprising.

In summary, it appears the market values more highly those intangible assets that

are closer to the “back-end” phase of the innovation process, such as applied R&D and

88

product R&D, presumably because the returns from these types of R&D are more easily
appropriated.

Although the statistics are corrected for heteroskedasticity, the regression does not
explicitly control for ﬁrm size. Therefore, it does not test for the Schumpeter hypothesis;
i.e., a relationship between ﬁrm size and R&D value. If the market incorporates ﬁrm size
effects into their valuation, equation 3.7 would not capture it. In the next section, I test

for size effects.

89

 

Table 4. Value of Disaggregated Intangible Assets

 

Ordinary Least Squares Analysis -- # Observations = 71

 

 

 

 

 

 

 

 

 

 

 

 

Regression
Independent Variable Unweighted Weighted
i ii
Imme t -338.913 , 577.292
1’ 183.143 251.023
. 1.58385 , 0.85815
ATang‘blc ASSC‘S 0.65486 0.65 843
-14.81394 b -6.37705
Process R&D 6.68049 6.81091
4.30335 , 4.91704
“0‘1““ R&D 1.21591 1.33121
. 19.18513 , 15.00427
Apphed R&D 2.50408 3.14081
. 3.91657 7.34025
Tm" gem“ R&D 5.40542 5.86822
. -14.22789 29.34434
Ba“ R&D 8.95988 10.57552
2.04972 , 2.06155
Ad“ ExPend‘ 0.39300 0.29935
F Value: 58.5 7 a F Value: 88.29 ‘1
Adj R2: 0.852 Adj R2: 0.897

 

 

 

’ Signiﬁcant at the .01 conﬁdence level.
" Signiﬁcant at the .05 conﬁdence level.
° Signiﬁcant at the .10 conﬁdence level.

 

Notes: Standard errors below coefﬁcients.

annual change in market value, AV“.
corrected.

90

The dependent variable is the average
Weighted statistics are heteroskedasticity-

 

3. Appropriability Eﬂect

In this section, I test for the existence of the appropriability effect. According to
hypothesis 2, process innovations should generate more net revenue than product
innovations as the sales of the ﬁrm increase. This should translate to higher valuations
for process R&D than product R&D as sales increase. In order to test for this effect, it is
necessary to estimate the impact of R&D on ﬁrm value at different levels of ﬁrm sales. If
the stock response to process R&D is larger than the stock response to product R&D, as
sales increase, then I will have evidence to reject the null hypothesis that the
appropriability effect does not exist.

To accomplish this empirically, I incorporate sales-intangible asset interaction

terms into equation 3.7 for each intangible asset variable. This yields,

AV = Ay+ 6!] ATA + a2 ADV + ,62 Sales*ADV + (13 Process [3.8]
+ ,B 3 Sales *Process + and Product + ,64 Sales *Product + a5 Basic
+ ,65 Sales *Basic + a6Applied + ,66 Sales *Applied + a7 Service

+ ,67 Sales *Service + s.

This methodology is based on Doyle (1994), with the difference that I am using

disaggregated R&D data. The use of interaction terms allows me to determine the

inﬂuence of sales on each intangible asset variable by examining the cross partial

9l

derivative of the market value change with respect to sales and the intangible asset of

interest:

62mm = 6], [3.9]
oTSales)o"j

where j represents a particular intangible asset, and 6]: its sales interaction coefﬁcient.

The parameter 6]- represents the stock response to an intangible asset controlling for ﬁrm
size. In other words, it represents the change in the value of an intangible asset in
response to an increase in sales. For example, if j represents process R&D, then the
impact of an increase in sales on the value of process R&D would be represented by the
coefﬁcient 63. A positive value of 63 implies that the value of process R&D increases
with ﬁrm size, while a negative coefﬁcient implies ﬁrm value declines with ﬁrm size.

The results of regressing equation 3.8 with ordinary least squares are summarized
in Table 5. They indicate that the addition of sales interaction terms changes the
interpretation of several intangible asset relationships. For example, the sales-process
R&D interaction is positive; as sales increases, the value of process R&D increases. The
coefﬁcient is signiﬁcant, so I can reject the hypothesis of no relationship between sales
and process R&D. In contrast, the sales-product R&D interaction term is negative,
although insigniﬁcant. Therefore, I cannot reject the hypothesis that there is no
relationship between ﬁrm size and product R&D. Taking these two results together
implies that the value of process R&D increases with ﬁrm sales at a higher rate than

product R&D, exactly as the appropriability effect predicted. Based on this result, I can

92

reject the null hypothesis that the appropriability effect does not exist.

However, the large negative coefﬁcient for process R&D implies investment in
process R&D may not be rational for all ﬁrms, depending upon their size. The responses
to process R&D investment and other types of intangible investments at the mean of ﬁrm
size are depicted in Table 5b. The Table indicates that both process and basic R&D have
negative responses at the mean. In the case of process R&D, ﬁrms with sales below
$12,403 million will not see positive returns to their investment. For basic R&D, ﬁrms
with sales greater than $503 million will not see a return on their investment.

Examining the other coefﬁcients, with the exception of advertising expenditures,
the value of all other assets decline as sales increases. This implies that, in general, there
are decreasing returns to R&D with regard to ﬁrm size. This contradicts the Schumpeter
Hypothesis. Previous research on the Schumpeter hypothesis has been inconclusive. For
example, Doyle found evidence in support of Schumpeter’s hypothesis in some industries
but not in others. One explanation for these ﬁndings is that Schumpeter is supported only
for some types of R&D, but not for every type. Therefore, research that has used
aggregate R&D to test for the Schumpeter hypothesis were dependent on the composition
of R&D in the sample used.

My results also may be inﬂuenced by the period I am examining. The 1990’s saw
high valuations for small and growing high tech ﬁrms, which may have increased the
value of intangible assets for smaller ﬁrms. However, another explanation is likely.
Notice that, with the exception of basic research, the stock response to process R&D, at

just under $8.4, is lower than other R&D responses. Therefore, the fact that

93

Schumpeter’s hypothesis is not supported is not surprising given that, as ﬁrm size

increases, the only R&D type that beneﬁts is the type with the lowest valued R&D.

94

 

Table 5. Value of Disaggregated Assets Controlling for Firm Size

 

Ordinary Least Squares Analysis: # Observations = 71

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Regression
Independent Variable Unweighted Weighted
1 11
Intercept 2:632: £33.83:
ATangible Assets 3:25:22; 3233)::
Process R&D -3223?) £3326:
(Process R&D)*(Sales) 3333,21 3333?
Product R&D 323:2: 33333?
(Product R&D>*<Sales> 33331333? 31333313?
Applied R&D 22123332 3323333
(Applied R&D)*(Sales) 000003293? 000022822
Tech. Service R&D 23:33: 1:383
(Tech. Serv. R&D)*(Sales) 333333; 0.0535(5)???
BasicR&D $213333 131i???)
(Basic R&D)*(Sa1es) $3333) -3333;
Adv. Expend. 8.30866) 3.3832
(Adv. Expend>*<Sa1es) 333333322 323333333
F Value: 49.49 a F Value: 88.09 a
Adj R2: 0.900 Adj R’: 0941

 

 

 

’ Signiﬁcant at the .01 conﬁdence level.
b Signiﬁcant at the .05 conﬁdence level.

° Signiﬁcant at the .10 conﬁdence level.

 

Notes: Standard errors below coefﬁcients.

annual change in market value, AVi, t-

95

The dependent variable is the average

 

 

 

 

 

 

 

 

 

 

 

 

Table 5b. Stock Response at the Mean - Weighted Regression

Independent Variable Mean Response Sales Level at “(Blah
response — 0

Process R&D -20.03 12,403
Product R&D 10.55
Applied R&D 16.78
Tech. Service R&D 10.47
Basic R&D -13.85 503
Adv. Expend. 0.44

 

* If mean response less than zero.
Notes:

I also ran a similar speciﬁcation using total R&D rather than R&D types to test

whether the Schumpeter hypothesis is supported with regard to total R&D. Table 6

summarizes the results for the standard and expanded datasets. Regardless of the data

included or weighting, the results indicate that for this sample Schumpeter is not

supported. The sales-total R&D interaction term is negative and signiﬁcantly different

from zero.

Table 6b indicates the stock response to aggregate R&D at the mean of ﬁrm size

is just over $6.6, and that of advertising is closer to $2.

In summary, process R&D is of increasing value relative to product R&D as ﬁrm

size increases, and therefore I can reject the null hypothesis that the appropriability effect

does not exist. The next section investigates the interaction between product and process

innovation.

96

 

 

Table 6. Value of Aggregate Assets, Controlling for Firm Size

 

Ordinary Least Squares Analysis

 

 

 

 

 

 

 

 

 

 

Regression
Independent Standard Data Expanded Data
Unweighted Weighted Unweighted Weighted
i ii iii iv

meme t -972.478 3 4323.900 3 -797.435 b 4029.773 3

p 332.395 415.110 328.577 337.330
ATangible 1.389 c 0.995 2.558 a 2.562 a

Assets 0.726 0.741 0.604 0.568
11.354 8.350 a 8.709 a 8.398 a

Tom R&D 1.330 a 1.516 0.971 1.014

-0.0005 -0.00025 000024 000022

*

Tom] Sales 0.0001 3 0.00010 b 0.00005 3 0.00005 3
2.426 3.252 a 1.749 c 2.209 a

Ad“ ExPend' 1.019 b 0.809 0.985 0.690

A dv*Sa1es 0.00006 a 0.00002 c 0.00003 0.00002

0.00002 0.00001 0.00002 0.00001

F Value: F Value: F Value: F Value:

45.53 a 71.08 a 53.49 a 76.26 a
Adj R2: 0.761 Adj R2: 0.834 Adj R2: 0.762 Adj R2: 0.821

 

 

 

 

 

 

‘ Signiﬁcant at the .01 conﬁdence level.
b Signiﬁcant at the .05 conﬁdence level.
° Signiﬁcant at the .10 conﬁdence level.

 

Notes: Standard errors below coefficients. For all regressions, the dependent variable

is the 5—year average annual change in market value.

heteroskedasticity-corrected.

97

Weighted statistics are

 

 

 

 

 

Table 6b. Stock Response at the Mean — Weighted Regression
. Mean Response Mean Response
Independent Variable (Standard Data) (Expanded Data)
Aggregate R&D 6.35 6.64
Adv. Expend. 3.41 2.37

 

 

 

 

 

Notes: Mean response computed at a mean sales value of $7,991 MM.

4. Interaction Between Product and Process R&D

So far, I have found evidence to support the existence of an appropriability effect.
The ﬁrst hypothesis in Chapter 2, derived from equation 23’, states:
Hypothesis 1: The eﬂiciency of product (process) R&D inﬂuences the efﬁciency (and
revenue) of process @roduct) R&D according to the sign of the interaction terms.
Essentially, this hypothesis rests on the idea that there are interactions between product
and process innovation such that an increase in the efﬁciency of one will have spillover
effects on the value of the other.

To test for an interaction between product and process innovation, 1 can adapt

equation 3.7 as follows:

AV = Ay+ a1 ATA + azADV + 03 Process + a4 Product [3.10]

+ 0(Pr0cess *Product) + a6 Basic + a7 Applied + 08 Service + 8.

98

The difference between this expression and expression 3.7 is the interaction term,
Product*Process. The interaction term serves a purpose similar to the sales-intangible
asset interaction terms I used before -- it allows me to determine the inﬂuence of product
R&D on process R&D (and vice versa) by examining the cross partial derivative of the

market value change with respect to product and process R&D:

z= 0"2E(AV) = a, [3.11]
UTProduct) oTProduct)

where 0' is the interaction coefﬁcient. Z (or 0') signiﬁes the stock response to process
(product) R&D controlling for product (process) R&D. In other words, it represents the
change in the value of process R&D in response to an increase in product R&D. Or, the
change in the value of product R&D in response to an increase in process R&D. A
positive value of Z implies that the value of process R&D increases as the value of
product R&D increases, and complementarities exist between them. Likewise, a negative
coefﬁcient implies a trade-off between product and process innovation. An implicit
assumption in this methodology is that the interactions on the efﬁciency parameters for
process and product innovation, 6 and /l, are symmetric. In other words, positive
spillovers from product to process innovation are mirrored by positive, and equal,
spillovers from process to product innovation. Although this may be a strong
assumption, it is necessary because there are no direct measures of the interaction terms.

The results for equation 3.10 are summarized in Table 7a. The coefﬁcient on the

99

product/process interaction term is positive and signiﬁcant. This suggests that there are
complementarities between product and process innovation, as Athey and Schmutzler
(1998) and Cohen and Klepper (1996) suggest. Investment in product R&D would seem
to induce beneﬁts to process R&D, and vice versa. The coefﬁcient on the interaction
term suggests that, for every dollar invested in product or process R&D, the
complementarity between product and process R&D increases ﬁrm value by an additional
$0.05 beyond the direct effect.

However, the large negative coefﬁcient for process R&D implies investment in
process R&D may not be rational for all ﬁrms, depending upon their size. The response
to process R&D investment and product R&D investment at the mean of ﬁrm size are
depicted in Table 7b. The Table indicates that process R&D has a negative response at
the mean. Firms with sales below $608 million will not see positive returns to process

R&D investment.

100

 

 

Table 7a. Interaction Between Product R&D and Process R&D

 

Ordinary Least Squares Analysis -- # Observations = 71

 

 

 

 

 

 

 

 

 

 

 

 

Regression
Independent Variable Unweighted Weighted
i ii .

-80.233 -151.883

Intercept 181.458 244.322

ATangible Assets (1):; b 323:
-28.788 3 -25.768 3

Process R&D 7.188 7.913

0.313 0.709

PTOdUCt R&D 1552 1638
0.034 a 0.041 a

Product*Process 0.010 0.01 1
. 18.833 , 15.820 ,
. 11.952 1, 15.609 b

Tech. Serv1ce R&D 5.399 5.888
. -13.828 c -28.786 3

Bas1c R&D 8.184 9.511
2.068 a 2.096 ,

Adv. Expend. 0.359 0.280

F Value: 63.13 a F Value: 89.55 ‘1
Adj R2: 0.877 Adj R2: 0.910

 

 

 

 

‘ Signiﬁcant at the .01 conﬁdence level.
b Signiﬁcant at the .05 conﬁdence level.
° Signiﬁcant at the .10 conﬁdence level.

 

Notes: Standard errors below coefﬁcients. For each regression, the dependent variable
is the 5-year average annual change in market value. Weighted statistics are
heteroskedasticity-corrected.

101

 

 

Table 7b. Stock Response at the Mean - Weighted Regression

 

Sales Level at which

 

 

 

 

Independent Vanable Mean Response response = 0*
Process R&D -20.68 628
Product R&D 2.06

 

 

 

* If mean response less than zero.
Notes:

Next, I respeciﬁed equation 3.10 to control for ﬁrm size, much as I did in the

appropriability effect section. Equation 3.10 becomes:

AV= Ay+ (1] ATA + a2 ADV+ ﬁg Sales*ADV+ a3 Process [3.12]

+ ,6 3 Sales *Process + 524 Product + ,64 Sales *Product

+ 0'] (Process *Product) + 07 (Sales *Process *Product)

+a5 Basic + ,65 Sales *Basic + a6Applied + [36 Sales*Applied

+ 0.7 Service + ,67 Sales *Service + e.

The addition of the sales-intangible assets interaction terms serves the same purpose they

did before; they control for ﬁrm size.

Of particular interest in this equation are the

interaction terms for product and process R&D. With this speciﬁcation, I can determine

the effect of ﬁrm size on the product-process interaction:

102

 

dZ = 02. [3.13]
é’SaIes

 

The parameter 0; represents the stock response to process (product) R&D controlling for
product (process) R&D and sales. In other words, it represents the change in the value of
the interaction between product and process R&D as ﬁrm sales increase. A positive
value of 0'2 implies that the value of the interaction increases with ﬁrm size, and a
negative coefﬁcient implies higher ﬁrm sales has adverse consequences for the
interaction.

The results of regressing equation 3.12 are summarized in Table 7c. The
coefﬁcient on the sales/product/process interaction term is negative, but not signiﬁcant.
This suggests that sales plays no signiﬁcant role in the interaction between product and
process innovation.

However, the negative coefﬁcients for several intangible assets imply investment
in these assets may not be rational for all ﬁrms, depending upon their size. The responses

to intangible investments at the mean of ﬁrm size are depicted in Table 7d.

103

 

Table 70. Interaction Between Product R&D and Process R&D,
Controlling for Firm Size

 

Weighted Least Squares Analysis -- # Observations = 71

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

. Regression
Ind endent Vanable . .
ep Unweighted We1 ghted
Interce t 78.974253 542.077503 c
P 227.226 273.821
. 0.736 0.127
ATanglble Assets 052 0.588
-20.686 -61.728 3
Process R&D 13.655 15.636
0.0001 0.004
* b
(Process R&D) (Sales) 00015 0002
-1.81 8 -l .3 82
Product R&D 4.725 5.887
0.00013 0.000234
*
0’det R&D) (33133) 0.00031 0.000372
0.043 0.059
* c
Product Process 0.028 0.034
-2.5817lE-7 -0.0000021
* 3|:
(Process Pmdu“) (sales) 0.00000216 0.0000024
. 19.166 30.088 3
APPlled R&D 6.389 6.635
. -0.00043 -0.004
at b
. 18.154 27.035 .,
Tech. Serv1ce R&D 9.468 10.875
(Tech. Service R&D) 0.00034 -0.0017
*(Sales) 0.00079 0.0008
. 49.167 24.668
BaSIC R&D 18.261 19.059
. -0.006 -0.000968
* b
(Bas1c R&D) (Sales) 0002 0.000828
0.550 -0.369
0.000036 0.000040
a:
(AdV- Expend) (Sales) 0.000031 0.000025

 

 

 

104

 

 

 

Table 7c (cont’d).

 

F Value: 55.21 ‘1
Adj R2: 0. 921

 

F Value: 86.28 ‘1
Adj R2: 0.948

 

 

" Signiﬁcant at the .01 conﬁdence level.

b Signiﬁcant at the .05 conﬁdence level.

° Signiﬁcant at the .10 conﬁdence level.

 

Notes: Standard errors below coefﬁcients. For each regression, the dependent variable
is the 5-year average annual change in market value.

heteroskedasticity-corrected.

Weighted statistics are

 

Table 7d. Stock Response at the Mean — Weighted Regression

 

 

 

 

 

 

 

 

Independent Variable Mean Response Sales Level é: WhICh
response — 0*

Process R&D -24.53 14,550 (+)
Product R&D -0.07 8,391 (+)
Applied R&D -1.88 7,522 (-)
Tech. Service R&D 13.45

Basic R&D 16.93

Adv. Expend. -0.05 9,225 (+)

 

 

 

* If mean response less than zero. Positive sign indicates larger ﬁrms beneﬁt from
R&D, negative sign indicates smaller ﬁrms beneﬁt.

 

 

5. Unobserved Heterogeneity

As discussed in the empirical framework section, the possibility that unobserved
heterogeneity, or ﬁxed effects, are driving the results is important to consider when
drawing conclusions. Unobserved heterogeneity arises ﬁom two primary sources: sample
selection bias and omitted variables. In my speciﬁcation, the parameter or,- represents
potential unobserved heterogeneity that may result ﬁ'om variables conceivably omitted
from the model, and can be considered an individual effect for each ﬁrm cross section,
and thus represents potential unobserved heterogeneity. There are several methods for
dealing with unobserved heterogeneity. My preferred method was to eliminate the ﬁxed
effects parameter, 61;, by ﬁrst-differencing equation 3.3. This method had the additional
beneﬁt of transforming intangible stocks from levels to changes, and thus avoided the
problem of estimating intangible stocks.

However, another common technique for eliminating ai is to use ﬁxed effects

regression. In ﬁxed effects regression, instead of ﬁrst-differencing, one eliminates ai by
using the difference of each variable from its cross section mean as the unit of
observation. The advantage to this method is that it maximizes degrees of freedom in the
regression because one avoids dropping the initial observation, and it may provide a

better treatment for risk. Since the beta is omitted, ﬁxed effects estimation would treat it

as an omitted variable and add the variation of beta to the ﬁxed effect parameter ai.
Therefore, because ﬁxed effects regression provides potential beneﬁts versus ﬁrst-

differencing, Appendix C contains ﬁxed effects estimates for each regression performed

106

above. To accommodate the ﬁxed effects methodology, speciﬁcations for each
regression are slightly different than those speciﬁed above. In particular, the dependent
variable is the year-end market value of the ﬁrm, rather than the annual change in market
value. This will imply somewhat different interpretations of the coefﬁcients than those
estimated using ﬁrst differenced observations. The estimated coefﬁcients for R&D and
advertising should be larger because they are being regressed against the level of market
value rather than a change in the level. However, one should expect signs and
signiﬁcance levels of the coefﬁcients to be similar to those obtained above. Signiﬁcant
differences will indicate evidence that unobserved heterogeneity is driving the results
obtained above.

Because of the exploratory nature of this research, several permutations of ﬁxed
effects were run for each hypothesis. These include weighted and unweighted
regressions, regressions with and without ﬁrm intercepts, and regressions with and
without a ﬁrst order autoregressive variance structure (to correct for serial correlation). In
general, the ﬁxed effects yielded similar results to those obtained above. Examination of
Tables C1 and C2 indicates that, compared to ﬁrst differencing, ﬁxed effects estimation
yields similar estimates for tangible assets, with stock responses between 1 and 2
depending upon the particular speciﬁcation. Intangible assets are highly valued
regardless of the ﬁxed effects speciﬁcation. The inclusion of ﬁrm intercept parameters
appears to increase the coefﬁcients on intangible assets and decrease the coefﬁcients on
tangible assets, which may indicate ﬁrm or industry effects are important to the

speciﬁcation.

107

Table C3 contains the results of ﬁxed effects estimates testing the Schumpeter
Hypothesis. Again, results were similar to those obtained using ﬁrst differencing. In
general, the sales-R&D interaction was negative. Table C4 depicts results for
disaggregated R&D expenditures. Like those above, coefﬁcients for product R&D and
applied R&D were greater than those for other types of R&D. Table C5 indicates that
ﬁxed effects supports the idea that the appropriability effect is positive, and Table C6
yields a positive interaction between product and process R&D.

In summary, ﬁxed effects estimation yielded similar results to those obtained
using ﬁrst differencing. The results of each major hypothesis tested above were

corroborated by the ﬁxed effects regressions.

108

SUMMARY AND CONCLUSIONS

Public policies towards innovation related activities have met with no small
amount of controversy. For instance, many have argued that accelerated capital
depreciation rules unduly favored investment in physical capital to the detriment of
investment in intangible capital (CED, 1986). Another controversial policy was the
research and experimentation tax credit, instituted in 1981. Critics of the credit argued
that it was not nearly as effective as pr0ponents contended because of various loopholes
that existed in the law. They argued the credit had its largest impact in how ﬁrms
categorized their expenditures for tax purposes; after the credit was introduced, many
expenditures previously not considered R&D suddenly and not so mysteriously appeared
in the R&D expenditures column on corporate tax forms. Therefore, instead of
stimulating R&D, as was its intent, the credit merely induced ﬁrms to rename pre-
existing investment in other assets as R&D (Mansﬁeld, 1986).

The rationale for government sponsored innovation incentives stems from the
belief that the public good aspect of innovation knowledge leads to below-optimal
investment in R&D. Supporters of policies such as the R&D tax credit argue that despite
the problems, these types of policies push investment in innovative related activities
closer to their optimal levels.

One issue often overlooked in these discussions is the relative importance of

different types of innovative activity. For example, Japan’s economic emergence has been

109

'l

attributed in large part to its focus on process innovation, in contrast to the United State's
focus on product innovation (Mansﬁeld, 1988). However, it also is argued that product
innovations are the basis for the creation of new markets and thus policies should favor
product R&D over other types of R&D. The lesson is that the composition of R&D related
activities plays an important role in the performance of the economy, but many aspect of
innovation remain poorly understood. This point is equally true from a microeconomic
standpoint; the composition of intangible assets likely plays as large a role in a ﬁrm’s
success as the amount of R&D itself.

This paper examines the market value of disaggregated intangible assets.
Speciﬁcally, it measures the stock response to advertising, physical capital, and various
classes of R&D, with a focus on product R&D and process R&D. Previous research in
this area has focused on the value of total R&D and its relation to industry or ﬁrm
speciﬁc conditions, particularly ﬁrm size, but has not examined the value of different
types of R&D. Awareness that the composition of R&D is important to ﬁrm value is
growing. Recent studies have indicated that some degree of success in both product and
process innovation is necessary for a ﬁrm to be successful over the long term (Capon et.
al., 1992).

I ﬁrst develop a proﬁt maximizing model that determines the optimal levels of
product R&D and process R&D. The results of the model suggest that at equal levels of
expenditures, the marginal values of different types of R&D can vary, suggesting
different levels of optimal R&D once the proﬁt-maximizing condition of marginal

revenue equals marginal cost is imposed. This implication is supported by the actual

110

 

 

behavior of R&D producing ﬁrms -- product R&D expenditures are on average 4 times
greater than process R&D expenditures. Another implication of the model is that the
value and/or efﬁciency of product (process) innovation can inﬂuence the returns ﬁom
process (product) innovation. Finally, the model suggests a variation on the familiar
Schumpeterian hypothesis of a positive relationship between ﬁrm size and R&D
proﬁtability. Rather than aggregate R&D, ﬁrm size plays an important role in the relative
valuations of process and product R&D. The returns to process R&D should increase
relative to the returns to product R&D as ﬁrm size increases.

To estimate these implications empirically, I employ a stock market valuation
model of the ﬁrm. I ﬁrst estimate the value of tangible and intangible assets. The stock
response to total R&D is estimated to be between 3.7 and 4.7 for each $1 invested, a
response consistent with a required rate of return of approximately 20%, and well within
the range of previous research. Intangible investments closer to the end of the innovation
process, such as applied and product R&D, generally are valued more highly than
investments at the beginning of the innovation process. However, the stock responses to
disaggregated R&D expenditures vary more widely than expected, from a high of 15 for
applied research to a low of -29 for basic research. The disparity in estimates may be a
function of different rates of depreciation for the intangible assets and/or differences in
the ability to appropriate the returns ﬁ'om each type of R&D. However, the most likely
explanation is misreported R&D expenditures for basic research and support services
R&D. Government contracts for R&D support basic research much more extensively

than other types of R&D, and therefore the costs associated with basic research are likely

11]

lower than the costs of other types of R&D. Furthermore, incentives derived from the
R&D tax credit and similar policy instruments provide an incentive to misallocate non
R&D-related investments as R&D. In any case, further research into the ,retums to
disaggregated R&D is called for.

Next, I test the Schumpeter hypothesis that ﬁrm size and R&D proﬁtability are
positively correlated. In testing this hypothesis, I use ﬁrm size-R&D interaction terms
and examine the mixed partial derivative of the change in market valuation with respect
to both ﬁrm size and R&D. The results indicate that Schumpeter is not supported. Firm
size is not a positive and signiﬁcant determinant of R&D proﬁtability. In fact, ﬁrm size
is estimated to have a negative impact on R&D proﬁtability. I extend this speciﬁcation to
examine the relationship between ﬁrm size and different types of R&D. The results tend
to support the idea of a positive appropriability effect; i.e., the proﬁtability of process
R&D increases with ﬁrm size at a greater rate than the proﬁtability of product R&D.
Because process R&D expenditures generally are smaller than product R&D
expenditures, this may explain the lack of support for the traditional Schumpeter
hypothesis.

Finally, I employ interaction terms between product and process R&D to test
whether there is a relationship between the two, and how this relationship affects the
market value of the ﬁrm. I ﬁnd that product and process R&D are complementary,
supporting much of the recent economics literature, but contradicting some strands of
management literature.

The stock market value of the ﬁrm is dependent upon many factors. Therefore,

112

. "I

the potential that unobserved heterogeneity, arising from variables omitted from the
model, is driving the results is a valid concern. Fixed effects regression of each major
hypothesis provided similar results to those obtained in Chapter 3, and thus ﬁxed effects
does not appear to be a signiﬁcant factor in the analysis.

The opportunities for further research are plentiful. First, the results of this paper
should be viewed with caution. It has been shown that the value of intangible assets is

not constant over time. The reasons for this may be several: ﬁrst, variations in the value

 

of ordinary assets may affect the value of intangible assets. Second, declines in the value
of R&D assets may occur in high-tech industries where the returns to R&D are short-
lived. Finally, changes in the composition of R&D over time may contribute to changes
in the measured value of total R&D. Therefore, the possibility that ﬁnancial markets
value different types of R&D at different levels may affect the returns to intangible assets
over time. The period examined here, 1992-1997, was a time of very high growth rates
for the value of common stocks. Therefore, additional research should be conducted in
this area covering different periods. In addition, several implications regarding the
appropriability effect were derived ﬁom the theoretical model that the data did not allow
me to test: (1) a positive appropriability effect is less likely to exist when the ﬁrm retains
a high proportion of its ex ante sales, (2) a positive appropriability effect is less likely to
exist in the presence of high sales growth, and (3) a positive appropriability effect is more
likely to exist in the presence of trade-offs between product and process R&D than in the
Presence of complementarities.

In summary, the ability to measure the returns to different types of R&D will

113

provide new insights into how the allocation of innovative resources affect an
organization’s proﬁts and equity values, and how outside factors may inﬂuence these
choices. The results of these inquiries can lead to improved R&D allocations by
improving the information upon which the organization bases its R&D decisions. The
results also will be a step forward in the quest to improve public policy towards R&D by
providing policy makers with a better understanding of the determinants of ﬁrm
innovativeness. A better understanding of the relative value of different types of R&D
and the different conditions favoring each type may lead to a more efﬁcient structure for

the R&D tax credit or other innovation related public policies.

114

APPENDICES

115

APPENDIX A

116

Es . 32118-3 «8.4% i A «gm + 2

”EN «86355 $63 maﬁa—9:? new EBB 3w:
2: can“ “Exam mm 5888:: 05 5 88. t2 2: wcﬁom 68385. 8388:: 05 3 3.552% E 223298 3: me cm? 2:...

I ll $3816-31 \«N + SWWWEVQ+ CBNII
f8$=x¥£€$+235.34.. EiEQNEEQEeE: QEEEQVQ+ :eiQiﬁékE -

 

 

74 $+te-:\: \«N + WSQEQVNV+ CON Fm

$.8-§Kﬁ+§qmm§+ :3 s§ééisﬁm§+2Q ai§i§§+ 28$: 1%.

 

 

”2

358. 2N + @msxisz
TEE 26:86-2: SEEQ+QS H m

 

.CN 2036.2on 90 c 2 83m“: 53> o>tm>tow 2:.

m _ .N ZOFEQZOU m0 ZOF<>EmQ

< XHQmem<

117

APPENDIX B

118

APPENDIX B

HETEROSKEDASTICITY CORRECTION AND TEST

To control for heteroskedasticity, I used one of the following two weighting
schemes, each assuming that sales is the pr0portiona1ity factor. One scheme uses log

transformations and the other does not. I assume an additive form of heteroskedasticity

in which the estimated variance of the disturbance, 62,, takes the form: , ~.

0'2,(a) = a0 + a, SALESi + azSALESZ, [B1]
or

62.05) = b, + b, log(SALES,) + bzlog(SALESZ,) [132]

where a0, a,, a1, b0, b,, and b2 are constants to be estimated. In some cases, a, = b2 = 0. In
each of the analyses above, I estimated the least squares equations, then apply the least

squares method to the following equations:

e2,~(a) = 60 + c, SALES,- + CZSALESZ, + v, [B3]

ifI assume B1, or

e2,-(b) = d, + d, log(SALES,) + dzlog(SALESZ,-) + vi [B4]

119

 

if I assume BZ. e2i(a) and e2i(b) are the ordinary least squares residuals obtained ﬁ'om

equation B1 and B2, respectively. This yields initial estimates of 62,:

oil-(a) = co + c, SALESi + c,SALEs2, [B1 ’]

021(1)) = d, + 6, log(SALES,) + d,log(SALEs2,) . [B2’]

These initial estimates of co, c,, c2, do, d,, and (12 are not asymptotically efﬁcient because

vi is heteroskedastistic. Therefore, another round of estimators are obtained by applying

least squares to:

 

£3,011): _c2_ +6, SALES,+c,SALES2, +v‘, [B5]
0211a) 021(3) 021(8) 0'21 (a)

and
e_2,{b)=_<_:9_ + c,16g(SALES,)+c,16g(SALEs2,) +v', . [B6]
oil-(b) oil-(b) 0211b) 021-0)

These estimators are asymptotically efﬁcient. Thus, the ‘second round’ estimators of 0'21:

is:

021(3) = c’, + c’, SALES, + c’ZSALESZ, [B1’]

and

120

021(1)) = d’. + d’, 16g(SALES,) + d’,16g(SALEsz,) . [B2’]

Asymptotically efﬁcient coefﬁcients for the speciﬁcations in each hypothesis are

estimated using 021(a) or 021(b) as weights.
After each hypothesis was estimated using the weighted least squares approach, I
tested for any remaining heteroskedasticity using a Park test. This entails regressing the

residuals obtained ﬁom the weighted least squares estimate against sales:

0021: 136 + B: SALES] + 81 [137}

where 002, are the weighted least squares residuals. A signiﬁcant B, indicates the presence
of heteroskedasticity. In each regression speciﬁed in each hypotheses in Chapter 3, little

or no heteroskedasticity was detected after the weighting scheme was applied.

121

APPENDIX C

122

-=:

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

v.82 om. Sow omdemv Ewan: 3.33 3.83 2.82 was“;
283 8.82- 3mm. 8.32- nmdomm- come 3.32-
38: 3.33.. 8.83. 3.2.3 3.52 3.6me $.88. 83“;
29; Shanon. madame cod—on- $.32»- 348mg vodmnm-
o3: omdvmm R. _ Nov aodw: Sdom— 3mg? N932 mmamﬁr
one: 2.32.- $.24; $.38. mmdoom- 34.23 mwdmmm-
:o: 2.6va 3.33 3.2.: 2.232 3.3? 3.32 .388,
22: 3.823- 828$. 3.3%. 2.8mm- $.38 Same—m-
3nd Sb.— Emd :06 3.0m 82. Sqm cued om. ._ mam mm _ ._ >Q<
3:6 m. n... Nae.» vmm.m $0: «3.3 88.2 :m.m_ no; 2 _._ «om.»
m _ mm «a: 89m at: ogd mm: ._ 2.5.. EON NS; omod 3o.— n36”.
maxi E5; Mum—.2 934.. mmmi 2.52 35.2 www.mm mmmé. 95mm 8::— .7309
wnmvd Somd $4de 36—8 :88 c: .o 2 and 2.36 wwb _ .o 3.33 mm _ .o a;
aﬁ : 2: 2 334 88.. ommmd on: £36 Soﬁ— wm$.~ 33.. So.—
8582 woméﬁ 8622:.
38.33. 26.32.
85280 832:3
x x x ._280_m:om 88:88
6.28830 .5
X x x x x x x _m:==<
80285
x x x x X 8.5
x x x x x x 823.63
2 c. o m N. c m w m N _ 6:280
:222wom

 

 

23830 mmwnmELE

ESE covcwaxm 1 mcoimoawox ﬂootm BxE ._ O 8.35

 

 

mZOBmmEOmM mHUmmn—m DmXE

U XEZmAE<

123

 

124

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

._o>u_ ..o 2:
a 8.32.:ch 2865 E8 3925 4:20:38 323 30:0 @3235 .EE 2: mo 02? 6x38 95.3% 2: m_ 033:; 289.3% 2; ”8qu
:22 363% 3.33 3.th m: .2 3.»me $.02: 35>
:2 m 8.2 mm 33.0. 3.33 ~33: wowﬂ 3.2%
38— :6ch ac. _ mmv 33m: fuzz; 3:va 3.32 8:35
mamvm 2.32 now: mm. _ «.2 8.32- wme 3.23.
I o_ m m o w v m 033:3
:Bmmewom Eon—anon:—
Gzzouv .5 2an

 

 

 

.32 3 2:

.a cosmoc.:w.m 286:. £8 962m 45.0508 322. 323 Havana .EE 2.. .o 2...; .829: use-So» 2.. m. 0.2.2:; Eovcoaon 2:- “882

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

mam. v3.3. 5..me 3.6mm. 33m. 3.9.: omdow 358w
2.. mm v8.23 wmvvm modvow madam ..N. m ovdwmm
3.8. 3.32 mmdomv mmdmm. vam. 3.2.: Nag-mm ombwo>
8ch 5.5mm .3... modmmm 3.55- 33.. ”v.3.“
vim. 3:50. 3.33. 36mm. .38. 3.33. 5.3.. 3:3».
32% 8.3m- $3. 8.0%- v. .wvm . - 83-. 2.68-
v. .N. 95-3. 2.83 3.2-N... ..wm: .ndmvv vmdmw was“;
wnmom 3.3m..- ﬂow. exam;- modoom- 3mm. o. 6.3-
38. ...-.03. cvdowm .oNom. ovdv. . mmdnmv 33mm 8.8.»
. Ea. 2.2.91 33.. ....Nmmm- mmdmom- monm. No.33-
mwom. omdmo. 36mm». NNdwN. omNm. . $.82. omdmw Nﬁmu>
o8... Elbow- codobo 8.5 . h- 8.33-- oﬂh. we. . SN-
mmad Eb. mood mmwd .mm.m Sod S..N m2; .2... Ema 98.. >D<
maﬁa. covd ~36 v56 woman .86 mm»... 2:. 30... m5..m mend
nwmd a... owwd SA... ammm mom. «we. mmmd wood cvm.m oz... may.
god 5. .c. 036. .m . .o 306. 3.3. mom... 308 c. ..b. ccwdm v2.2 -.<-..Oh
mmmd 5mm... 3:. m. ..o and an... mead 3:. m3 .o 32. mid <h
o . mod 3...... :6.— ovv. cnmd vmv. .mhd moo. awn .o owe... wow.
mmmﬂmma mmmnhmv .328...
mg 32- moo a2..-
cogotou ointm>
X X X .28“. 25.2.2.2.
Etom -...
8:555
X X X X X X X 32:5...
3322:.
X X X X X E...
x x x x x x 83903
.. o. a w .- c n v m m . H25.50
20.383”.
:vnso ...-nae...
3:3. 65.28% I 20.32qu Bootm. next. .NU Baa-r

 

 

 

125

 

 

 

 

 

 

 

 

 

 

 

 

 

 
  

 

          

 

 

 

 

 

  

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

an. 8%; 8.5;. Nm.§n,_ «2:; 3%? 2.8a
”new 2.5:- .2.$E~ 5...»va- E.c~o_- 93mm 8.38. as;
2:. manna 8,33 033 8.82 2.83. 2.2m
3m: $.85... cams-Nu 8.3%- aévm- $ch $.32- as;
:8. 2.3.8 «35‘ as: 8.82 8.2.3 8.»:
a: 8.8%., 3.94% 93:5. NONE”- $02 58$. 83;
SE 3&8 8.3m... 5...me , 3.2-2 3.22 gm;
~22 $82-- 58: 3.9.8.... 2.3%- Evnm ESQ- 85>
883 880d 8086 883 258.0 S885 0886 882 «885 8890 383 mm4<m
883. «885. 286 c-mwﬁ- 3256 .8886. 883 $83 883 $826 883 33
M33 .33 $3 . .53 23 $3 an? 83 23 m8..- 3:
~32 «8.2 «8.2- 3:: 612.2 Sh: “and X52- Q8.» 5.2 8,2: >o<
. mm4<m
:83 283. 2255 :82. .886 28.3 883 £83 £83 883 :83 55.
3885 383- 883. :89? 8.8.? 883. :8o.o 258.c- Roood- «88.? €59
83. :3 :3 EN 23 Ev.“ Sta 3% 83 92
a: mag-m “3.2 nmwldlm Ewe E3 .38 :93 32a :58
£3 and .23 mm; 23 :28 SS 82 83
23.0 $3 82-. $2 a: 82 $2 8.3 82 E
”8.2 : 833
333. $.88- 38.25 ,
coo-COD U_n_m_-_n>
X X X APEOU «:oﬁcogo?5.
atom
x x x x x x x Eecmm
58.2:—
x x x x x E:
x x x x x x Ems?
: S a a n o w v m m _ “2.230
Same-aux
:vuso ”:nmei
mELE ﬁhwtcgm | mcoﬁmohwom ﬂow-Cm cox; .MU czar—-

 

 

126

 

 

 

 

 

 

 

 

 

 

 

 

 

 

. o>o _.o 2:
E 8:8E:wﬁ 8365 E8 335 456508 323 288 2%:me .85 05 mo 2:? 8x35 95-on 2: m. 032:; 2532.2” 2:_- ”862

:3: $6va 3.2 _m 3&9; 2.83 3.83 wad—o—
mNaOM $.28 wodmmcm ~58: MON-mom 39mm mmdoo. 3.83»

ommm. ovgvmm Emma: ”Add; 2.92 3.3% 3.23

manvm 3.2-\- oc. Scam 3.33. 3.2: mmmom 3.3m- 338»
I S a w o n v m 032:;
:28”:me 2032.85

 

 

 

 

APEoUv .mU Ens-r

 

 

127

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

omwmwo 3.003 ....nmvm ao.~00. 3.3-... 3.3% v....vo.
no.3..- oo.~vvv- no. .N0- .... . mmv- av. .3 . - 8.33 h . .w. .m- $.80.
30.? v0.58 8.00.0 mm. . mm. .v. .om. m0.m~mv 3.3..
om.0vv. - 004.0. 0- 3.00m . - vvdmmw- o. .mmmm- 00.38 86v. m- mo.8>
mmdowo 8.33 3.2 mm 3.2... ., mm.v.~. va. . vov vN.m0..
cavnmv- «Gan..- 8...va- b0dmm0- WEvE-N- .v.v§-w 023...- 9.8%
2.00 $04.. N0m.m .34 vav mom. mmvd wand 3m.— owmd mum.—
oZ-vm 80d. Nada «3.6. 30.5 03... v28. 8nd. mid wmmd. 000.2 >Q<
Sag-N .mwdm mvmdm 000.0“ . .. imam Sag-m n36. ..o..m .vw.mm 0506a Nahum
www.mv- N006. awovv- . .Ndm 0Nv.0v- 0.. ...w wowav .093 03.00 .0. .3 av... . .- 0.95
$06. 2»... v0nd. Ema. avnd. 03.0. o.0.~. v. N... 30.». N30. vomv.
..vm..m >30- o.v.o.. .0m.om- mmvﬁ 000.2.- 000..- vm0.m.- v2.3- wvh..m- .260- 83.0w
«and. m0~d 000.3 056 30.0. maﬁa m0m..- mvvd. 0om.w h. Wm. .20
wmmdm .mNdm 80.2 nmv.0m v3.0.- O0v.wm mam-0m ~.m.>0 ovmdn 3.60 mmm.nn 023......
omav m.n.m .36 Eva and «N0... SQN 0ov.v van m.m.v mmo.m
2....v 00v... . 00w.0 mvms. 300 2.6.0. 2.0.0. vvws m.w.v. 0006 20.3 8:00..
no. .m. 02.... 38.3 0o~.v. mwmv. «Ev. 02-6 2.5... GEN. mom... 0v.-.N.
20.2 man... wvvd. vomd- ovwv. vmm.wm- o0v.0- ...v..- 3.0.3. 090m- omﬁoT 382.
050.0 . RN... 050... m3... 9.0.0 vmmd 0mm... v~0.o ammd 0.0.0 ammo
v3. 8:. 03d .30.. vm0.m w . v. . 9...... ”3.. av0.. mmo. mam. <-_-
~30: . v00. 80
8.88- 52%.- 38.2...
52.00 o.00..m>
x x x .280 .50530
3.0m -...
8.5
x x x x x x x .0352
38.2...
x x x X x 0...“.
x x x x x x Ems...
.. o. a w h 0 n v m N . H22:5
20.3.033—
m man-BO MKNEE...
2:... 0.005% .- 2232me 88.5 08.... .v0 0.0.;

 

 

 

128

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

€3.50. .v0 07....-

..25. .... 2..
... 8:3...Ew5 0.8.0:. 2.8 000.2% .58....08 322. $8.0 0.005% .E.... o... -.o 2...; .805: 0.5-...... o... m. 0.00.2? 250530 2% .882
9.8. 5.22 .38. 8...... was... 2.. . .... .38.
~22... 8.88 8.38 .58. .353 .55. 3.3: 8.8..
.58. 2...... m 2.8% 3.8: 382 mm. . N... ...... ..
3.8.- .38.- ..oie- 8...: .- NS...- 333 2......- was;
.8... 2.28 $.83 2.2... ......N. 5.8% 2......
.863- 3.3%- 3.3.. ..dém- m2...- 3....§ $.28- 85>
_ .. ... a w o w v m 2.2.5
_ 5.32me .505305
.

 

 

129

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

_. 0000.0 00. 00.0 00000.0 00.000 00000.0 00 . 00.0 . 0.00.0 . 0.00.0 00. 00.0 00 . 00.0 00. 00.0 8.00
00000.0 . 0000.0 00000.0 00000.0- 00000.0 00000.0- 00. 00.0 00000.0 0. 000.0- 00. 00.0 00000.0- $0.200
0.0.00 .0000 000.00 000.00 000.0 .00.00 000.00 000.00 .0..00 000.00 000.00 8380
0000.- 000.00- 000.0. 000..- 000.00 000.00- 000.00- 000.00 000.0 00..00 000.0- .
00 .000 00000.0 00.000 00.000 00.000 0. .000 00000.0 0 0 .00.0 0. .00.0 00 .000 0. .000 3.00
00000.0 0-0.00.0- 00000.0 00000.0- 00000.0 00000.0 00000.0 00000.0 0 . 000.0 00000.0 00000.0 .00..0“?
000.00 000.0. 000.00 00040.. 000.00 000.00 0.0.0. .0000 00.0. 000.00 000.0. 02.30.
000.0- 09.00 000.0- 000.00. 000.00- 0.0.00 000.00 000.0- 000.00 000.00- 000.00 .
00000.0 00000.0 00000.0 00000.0 00000.0 00000.0 00000.0 00000.0 00000.0 00000.0 00000.0 m200
00000.0 00000.0 0. .00.0 . 0000.0 00000.0 00 . 00.0 00000.0 00000.0 00000.0 00000.0 00000.0 $2680
000... 000.0 000. .. .000 000... 000.0. 0.0.0 000.0 000.0 000.0 000.0 8:080
0000.- 000.0 000.00- 000.0 0000.- 0.0..- 000.0 000.00- .000 000.0- 000.0
00.000 00.000 00.000 .0.00.0 00.000 00.000 00.000 00.000 .0.00.0 .0.00.0 .0.00.0 8.00
00000.0 00. 00.0 0 . 000.0 00000.0- 00000.0 00000.0- . 0.00.0 00 . 00.0 00000.0- 00 .00.0 00000.0. $88.0
000.00 .00..0 000.00 000.00 0.0.00 000.00 .0. .00 000.00 000.00 000.00 00.00 ...-.0880
000.0- 0.0.0- 000.00 000.00 00.... 000.0 . 0.0.0- .000. 000. . 0 000.0. 000.00
000.0 000.0 000.0 000.0 000.0 000.0 . 000.0 0. 0.0 000.0 000.0 000.0 <0
000.. 000.0 000.. 000.0 000.. 000.0 . 000.0 .00.. 000.. 00... 000..
000.0.0. 000.0.0 2.8.2:.
000.0000- 000.0000-
:ootoo 0.00:;
X x x .280 2605000
.380 -c.
x x x x x x x anecmm.
2.8.8:.
00 x x x x E .. h.
x x x x x x 20.03
: 2 a 0 0 0 0 0 m N _ ”220.0
5.3203.
0 00HBO ..0n2Er.
0:5”. 050520 I 80.08.03. 380.0. .50.“. .00 2000

 

 

 

130

.0 00:000.:06 0.8.0:. 0:8 000000

0:22.008 30.00 0.8.0 0:00:80

.95. ..0 2..
.EE 2.. 00 o:.0> .8008 0:0-.000 2.. m. 200...? 200500 2.0 ”8.02

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

00.0000 00.0000 00.0000 00.0000 00.000. 00.0000 00.000. 00.0.;

.00. 0 00.0000 .0000 00.0000 00.000. 00000 0. 0.0.

.0.0.00 00.0000 00.0000 00.0000 00.000. 00.0000 00.000. 00.08»

00000 00.0000- 00000 0.0000- 000000- 00000 00. . 00-

.0.0000 00.0000 00.0000 0.0.00 00.000. 00.00 . 0 .0.000. 00.080

00000 00.00. 0- 00.00 ...00.0- 00.000- 00.00 ...0000-

0.0000 00.000 00.0000 00.000. 00.000. 0.0000 00.000. 00.0o>

00. 00 .0.0000- 00000 .0.0.00- 00.0.0- 00000 0. 00.0-

0.0000 00.000 00.. .00 00.000. 00.00. .0.0.00 00.000. 00.080

00000 00.00.0- 00000 00.0000- 00.0.0.- 000.0 00.0000-

00.0000 .0.0000 00.0000 .0.0000 00.000. 00. . 000 00.000. 0 0 .00 >

00. .0 00.0000- 00000 00.0000- 00.000 . - 0. 0. 0 00.000 . -

00000.0 0 .0000 00000.0 . .0000 00000.0 0.0000 0.0000 0.0000 . .0000 0 .0000 . .0000 8.00
.0 .0000 0 .0000- . 0000.0 0.0000 00000.0- 0.0000 0 0000.0- 00000.0 00000.0 . 0000.0 00000.0 _.>0.<
_0000. 000.0 000.0 000.0 .000 0.0.0 00.0 000.0 0. ..0 000.0 000.0 >Q<

000.0- 00. .0. 0000.- .00.0 000.0. 000.0 ..0.0 000.0.- 000.0 000.0 00. .0

00000.0 00000.0 00000.0 00000.0 00000.0 . 0000.0 00000.0 00000.0 00000.0 00000.0 00000.0 00.00

00. .0.0- 00000.0- 000.00- 00000.0- 00000.0- 00000.0- 00000.0- 000 . 0.0- 00000.0- 0.0.0.0- 00000.0- 0.0.000.

000.00 000.00 0.0.00 00.00 000.00 0.0.00 000.00 . 0.000 000.00 00.000 0.0.. 0 2000.
.00000. 000.00. 000.0: 000 .00. 000.000 00. .00. 000.00. 000.00 . 000.00 000.00. 000.00 .

. . 0. 0 0 0 0 0 0 0 0 . 0.00.:0>
5.08.03. E00533.

 

3.2.8. .8 0300

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

00.0.0 0. .00 0000.0 000. .. .0000 000.. 00.0.0 0000.0 0000.. 0000.0 .00.. >D<
.000.00 0000.0. .00000 0.00.0 0000.00 00.... 0000.0. 00.0.0. 0000.0 0000.0. 00.0
j000000 0000.00 0000.00 0 . 00 .0 0 00.0.00 000.00 0000.0. 0000.00 0 . 00.00 0000.00 000. .0 2000.
000.00- 0000.0 0000.00- 0.00.00 000. .00- 000.00 0000. . 0 0000.00 0000.00 0000.00 000.00 .
000.0. .0.0.0. 0000.0. 0.0000. 0000.0. 00.0. 0000.0. 0000.0. 000.0. 0000.0. 000.0. 8380
0000.0. 00. 0.0- 0000.0 000000-. 0000.0 000.00- 0000.0- 00 . 0.0. - 0000.00- 0. . 0.0.- 000.00- .
0000.0. 0000.0 0000.0. 0000.0 00.0.0. 000.0 00. 0.0 ..000. 0000.0 0000.0. 000.0 02.00.0
.00000 00. 0.00 0000. .0 0000.00 0000.00 000.00 . .0000 0000.00 0. 00.00 .00000 0 . 0.00 .
00000.0 00000.0 00000.0 00000.0 .00000 00000.0 00000.0 0.0000 0 0000.0 00000.0 00000.0 8:080
.0. 00.0 00000.0 00000.0 0.0. .0 00000.0 0000. .0 000 . 0.0 0000.0 0. 00.0 00000.0 0000. .0 000880
0000.0 000.0 0000.0 .0000 0000.0 000.0 0000.0 0000.0 00.0.0 0.00.0 000.0 8:080
0000.0- 000.0. 00.0.0- 0000.0 0000.0- 000.0 0.00.0. 0000.0- 0000.0 0000.0- .000
00 . 0 .00 0000.0. 0000. . 0 0000.0. 0000.00 .000. 0000.0. 00. 0.0. 0000.0. 0000.0. 000.0. 0080 . 0
000...- 00000.- .00. .0.- 00.0.00- 0000.00- 000.00.- 00000.- 0.00.00- 0000.00- 0000.00- 0.0.00-
00000 0.000 0000.0 0000.0 .0000 0 .00 0000.0 0000.0 00000 5.00 000.0 <0-
0000. . 0000.0 .0000 00.0.. 0000.0 .00.. 0000.0 000. .0 0000.. 000. .0 0 00. .
00.00.. 0.0.0.0 8088:.
.0..000- 000.000-
5880 o.00..0>
X x x .880 800.8000
.0..00 .5
5:0
x x x x x x x .0:::<
.0088...
x x x x x .8...
x x x x x x 0.0.03
: o. a 0 0 0 0 0 m N _ ”25:00
8.8803.
0 00n000 ..0n08.0
0.8.0 0:00:80 1 0:23.803. 088.0. 000.0 .00 0.000.

 

 

 

2

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

.03. ..0 o...
.0 3:02...ch 0.8.0:. 2.8 0009.0 0.5.8.008 30.2. 22.5 0.00.5.0 dz... 2.. ..o o:.m> .8208 .05-.80 2.. m. 200...? 200.5000 2...- ”8.07.
0. .0. 00.0.00 00.0.00 00.0.0. .0000. 00.0000 00.00.. 00.80»
.0.0000 0000.0 00.000 00.0000 0.000 0000. 00.0000
0000. ...00.0 00.0000 00.00. 0.000. 00.0000 .0..00. 00.080
00.000- 00.00. .- 0. .0000- 00.00 00.0.0.- 00.0000 0000.0.
00. .000 00.0 .00 00.0000 00.000. .0000. 00.0000 00.0.0. 00.8>
00.00- 00.00.0- 00.0000- 00.00.- 00.000.- 0.0000 00.000...
00.0000 0.0. .0 00.0000 00.000. 00.000. 00. . 000 00.0.0. 005;
00.0.0- 00.000- 00.000.- 00.0000- 00.000.- .0.0000 00.00. .-
000000 0000.0 .0.0000 00.000. 00.0.0. 00.0000 00.00.. 0023.0
00.000.- .0.0000- .0.000.- .0.0000- 00.000- 00.0000 00. .000-
00.0000 00.0000 00.0000 00.000. 0.000. 00.0000 00.00.. 00Eu>
00.00.0- 00.0000- 00.0000- 00.0000- 00.0000- 0.0000 00.000-
. . 0. 0 0 0 0 0 0 200...;
00.3203. 200.5000...

 

 

 

 

€0.50. .00 0.00...-

 

 

3

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

8288 8888 828.8 828.8 8288 88.8 888.8 828.8 828.8 828.8 8.8.8 3:8
838.8 888.8- 8288 888- 828.8 88.8. 8_8.o- 8288 888.8. 8888 88.8- .332
888. 88.: 88.8 8.: 3.8.8 82: 88.: 88.8 88.: 88.8 88.: 88.2
88.2 88.8 68.8 8.2: 88.8 8.8: 828 88.8 888 88.: 888 .
888.8 888.8 288.8 888.8 88.8 888 888.8 888.8 888.8 888.8 28.8 N ..Smmmm
888.8 :83. 888.8 88.8 888.8 88.0 888.8 8888 388.8 8:8 883 .8885
88.8 888.0 888.8 888.8 88.8 $888 88.8 88.8 8888 888.8 888.8 885
88.8- 88.? :88- 8887 888- 88.8- 888- 82.8- 82.8- 888- 883- .3805
8888 888.8 888.0 888.8. .8888 88.0 3888 888.8 888.8 288.8 88.8 8.8
8.8.8- 888- 88.8- 88.8- 888- 88.8. 888.8- 8888- 88.8- 88.8- 8_8.o- .8805
88.: 28.: 85: 88.: 88.2 8.: 88.: 88.2 88.: 88.: 8:: .885
88.: 8:8 888 8:8 88.; 83.8 88.; 888 88.: 88.8 88.3
_888.o 8:88 8888 8:88 2888 E88 8288 8288 828.8 8:88 888 3.8
3288 «888- 8288 888- 8888 88.8- 8.88- 888.8- 888- _88.o- 888.c- .382;
788.: 88.8 88.8 88.8 888 888 88.8 38.8 88.8 88.8 838 $89:
88.:- 888- 88.2- :88- 888- 8.8.- 88.8- 8:3: 88.8- 88.2 88.-
88.8 88.0 88.8 288 88.8 82 :88 £88 888 8.2 88 5
8:; ENE 88: 888 88; 85 882 m8... 888 88.. 83
882 888 88.35
8 88- 88 m8-
ﬂow-ECU Eng-ax)
x X X ._otou Eon—Egon
3:8 -5
x x x x x x x Edna
E853—
x x x x x 5.8
x x x x x x 2825
: 2 o w 8 o n v m N _ 82:5
coaewom
8&8 EH25;
85E Emu—Sew I 8.28.er $0th taxi so 038.

 

 

 

134

.82 3 8:835 82,; g

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

._o>o_ _.o
05 8 8585ch 88:8: 2.8 3.55 456508 323 823 Havana .EE 05 oo 02? 8th.: 86-on 2: m_ 033:; Enocooou 2F ”8qu
Shook-o $.33 mvdoﬁ. goof €833 m: _ m _ “93>
2 RV vodmmm aﬂov Nvdmv Nomwm 3.85
omomvc mod _ vm 8880 mos-om. Gooom acou— @033»
«when $.m2 m- 3va 3.8.2- omoom 8.2 m
3.3% wodhmm @333 8.22 no. _ oov 3.8: 388,
an? o _ .vam- 3.88 8.33 - oﬁ mm 3.98.
3.5-nun 3.3mm no.3 _ o ovowm _ modoov omooo— 328$
55on Summon- omomm 8.32- gm?- mmdo— .-
868m vooomm ._ c. _ now $.32 8.2-3 coda: no.8;
onmm nip-Ev- ommmm . 3.9.:- oommm ”:5;-
vnooom otmmvm Somoo 3.2.2 369% tumm— _ No.8;
3on o ﬁgme- oooom 3.3 _ m- 3on 3.30-
vmmoooo om _ ooo.o ommoooo _ o _ ooo.o ommoooo o_ooo.o mm _ ooo.o vw _ooo.o _o_ooo.o 2. _ooo.o o _ ooo.o 8.5m
.mooooo omoooo- vmooooo onooooo omoooo- .ooooo mooooo- envoooo 3886 mm _ ooo.o w o ooo.o *>Q<Xm
omooo Sum-m mm: .o :88 32.6 nowam wovmd 33.6 mochm mocha EQN >o<
336. mm. ﬁg mow-ed- ommwo vomodm God 35d of: ._ T om. _.m- vmowi onmo
ownmooo vwomooo $386 on _ mooo onomooo cmmooo mvomooo mi moo-o mummooo t: mood Nomooo 37%
983.? m _ vooo- oco_o.o- nmvvooo mow .o.o- vomooo mmvooo- “Ko— o.o- omovooo ooﬂoo- mvvooo tumm—
EZ-dv $3.3 m2 m5. 089% 2.3.3 oomwm «gown 883% «2 wow 02 wém Soov 28m
$32 23.3 213 momvom Eo—N v8.8. 23:3 3.3: 2 5.3 omoi M: ﬂow .
mwﬁooo m _ ﬂood ooiooo omﬂooo mo_moo.o S—ooo mmEooo ooSoo.o mom—ooo omo_oo.o om_oo.o 88m
Eovooo mocmooo hmwvooo Elooo- ommvooo vooooo- onvmooo mow—ooo Smooo- monoooo Smooo- ...SEom
hmvmﬁv mmmvém mmnmov momodm mommom 55.3” ooﬂdm 3.2.3.. mummdm nonﬁmm mvoom 838m
wmmmdv- 2“.me Swoon- omﬂ .mn- 38.3. 32:.- owtdm- $31. 33.3. wﬂoo moo. _ v- .
I o_ o o N- o n v m m _ 033:;
p.282?”— E28882:
6.258 .8 033.

 

 

135

APPENDIX D

136

 

._o>o_ _.o 2.:

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

a oo=SE=m_m 28:2: m=ou BREW .E2quoo 323 32.5 Emu—Em .EE 2: oo 2:? .819: €0-on E. a 032:? Eat—.38 2:. ”882
Bad , won? 2 mo umvo ommo mwno «MW—o w _ mo ommo hmmo >Q<
on?“ _No..m mono Now.“ 3.6 N26 29o vmmo BEN Nona
wow; Sn; Sod mono mmwo mono Emo nmvo Nmoo mono 00mm
MEN”: «3.2 «San , . ovmw 56.— ohoo Nov; Goo nmmo Eco- 4<+Oh
omwo own; mono Fvo Nmno Evo ammo N_m.o mono Motto 5.
«Rd vmno Now; ovoo So..- 2.0.7 SN..- 33. gm; on

8132 3:63 omoémv Sodom Soomm ommgmm 3.053 oomo: ”:33 mvmsom 385::
www.moE n. Swami- _om._o_. ooooom- 333$ gnomm- SNoE- movon- www.mmm- maﬁoom-

I ooumno I oonmno I Sumac I oonmno r ohnmEO

Wolmwo 3E9?» WWIMHWO BEwEoB Wolwmwo BEBE: whimwwo BEEP: whimwo 32.32: 033”;

o E . 3 E: u E . 3 é: o E . 3 ..5 u E . 3 .2: c E . >9 .:3 E235 No

3.8 3-3 3.3 voéo 8-8 _
2232»;

 

BEE Enos—3m I 50> Eomm 5o mzoawewom moss—um 3.84 92:20 .0 2an

 

 

wow—3mg yam; A<DQ>HQE

D X_OZmnE<

137

APPENDIX E

138

 

Table 1. Summary Statistics-Responders and Non-responders ($MM)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

All 1992 1993 1994 1995 1996 1997
Years
The following variables were supplied by both the IRI survey and Compustat
Avg. R&D Exp.
Survey Resp. 326.5 219.0 354.3 334.6 195.8 386.8 449.2
(# Firms) (46) (46) (5 8) (60) (67) (57)
Non-Resp 483.5 437.0 376.7 436.9 785.9 473.1 417.4
(# Firms) (56) (54) (44) (41) (36) (41)
Advertising
Survey Resp 180.3 150.5 202.0 119.9 76.8 279.8 240.1
(# Firms) (46) (46) (58) (60) (67) (57)
Non-Resp 223.0 213.1 187.7 242.1 383.2 115.7 194.5
(#Firms) (60) (60) (48) (46) (3 9) (49)
Market Value
Survey Resp 1 1878 9048 9174 8949 8393 14279 20352
(#Firms) (40) (42) (53) (54) (61) (50)
Non-Resp 11795 8397 9062 9710 16714 14069 15133
(#Firms) (54) (5 3) (42) (41) (34) (41)
Firm Size ($Sales)
Survey Resp 1 1369 8960 1 1742 1 1865 7923 13632 13474
(#Firms) (46) (46) (5 8) (60) (67) (5 7)
Non-Resp 12262 1 1912 9999 1089 18463 1 1399 1 1474
(#Firms) (56) (55) (45) (44) (3 7) (45)
Ratio R&D/ Sales
Survey Resp 4.4% 3.3% 3.2% 3.0% 3.3% 8.3% 4.1%
(#Firms) (46) (46) (5 8) (60) (67) (5 7)
Non-Resp 11.1% 8.1% 8.2% 23.6% 14.8% 4.7% 7.3%
(#Firms) (5 6) (54) (44) (41) (36) (41)
The following variables were not available from Compustat
All 1992 1993 1994 1995 1996 1997
Years

Product R&D 161.9 107.8 241.5 ’ 171.2 88.1 106.2 265.2
Process R&D 47.1 27.2 46.0 80.4 27.4 36.6 61.0
Basic Research R&D 10.7 8.5 10.7 10.2 8.5 11.1 15.0
Applied Research 46.2 34.2 34.6 46.8 44.3 47.2 68.3
R&D
Technical Service 40.3 34.0 36.7 42.8 26.1 53.2 48.5
R&D

 

 

 

 

 

 

 

 

139

 

REFERENCES

140

REFERENCES

Acs, Z. and DB. Audretsch (1987), "Innovation, Market Structure, and Firm Size,"
Review of Economics and Statistics, 69(4), Nov., pp. 567-74.

eds. (1991), Innovation and Technological Change: An International
Comparison, The University of Michigan Press, Ann Arbor, MI.

Altshuler, Rosanne (1988), "A Dynamic Analysis of the Research and Experimentation
Credit," National Tax Journal, 1988, v41(4), 4453-66.

Anderson, E. and D. Schmittlein (1984), “Integration of the Sales Force: An Empirical
Examination,” RAND Journal of Economics, 15, 385-95.

Arora, Ashish, Ralph Landau, and Nathan Rosenberg, eds. (1998), Chemicals and Long-
Term Economic Growth: Insights from the Chemical Industry, John Wiley & Sons, Inc.,
New York.

Athey, S. and A. Schmutzler (1995), “Product and Process Flexibility in an Innovative
Environment,” RAND Journal of Economics, v26(4), pp. 557-74.

Athey, S. and S. Stern (1998), “An Empirical Framework for Testing Theories about
Complementarity in Organizational Design,” NBER Working Paper No. 6600.

Audretsch, D.B., Albert Menkveld, and A. Roy Thurik (1996), The Decision Between
Internal and External R&D,” Journal of Institutional and Theoretical Economics, V152,
5 19-530.

Austin, David H. (1993), “An Event-Study Approach to Measuring Innovative Output:
The Case of Biotechnology,” American Economic Review, v83(2), 253-258.

Bernstein, J .I. (1986), Research and Development, Tax Incentives, and the Structure of
Production and Financing, University of Toronto Press, Toronto.

Burton, Peter S. (1994), “Product Portfolios and the Introduction of New Products: An
Example from the Insecticide Industry,” RAND Journal of Economics, v25(1), 128-140.

Calantone, R.J., C.A. di Benedetto, and B. Sriraman (1994), “Examining the Relationship

Between Degree of Innovation and New Product Success,” Journal of Business Research,
v30, pp. 143-48.

141

Capon, N., J .U. Farley, D.R. Lehmann, and HM. Hulbert (1992), “Proﬁles of Product
Innovators Among Large US. Manufacturers,” Management Science, v38(2), 157-168.

Chesbrough, H.W. and DJ. Teece (1996), “When is Virtual Virtuous? Organizing for
Innovation,” Harvard Business Review, 74(1), 65-74.

Clark, KB. and SC. Wheelwright (1993), Managing New Product and Process
Development, The Free Press, New York.

Cohen, W.M. and S. Klepper (1992), “The Anatomy of Industry R&D Intensity
Distributions”, American Economic Review, v82(4), 773-799.

(1996), “F inn Size and the Nature of Innovation Within Industries: the Case
of Process and Product R&D,” The Review of Economics and Statistics, v78(2), pp. 232-
43.

Committee for Economic Development (CED) (1980), Stimulating Technological
Progress, The Heffeman Press Inc., New York.

Collins, EL. (1981), “Tax Incentives for Innovation-Productivity Miracle or Media
Hype?," Journal of Post Keynesian Economics, 4(1), 68-74.

Cordes, J ., H. Watson and J. Hauger (1987), "Effects of Tax Reform on High Technology
Firms," National Tax Journal, 40(3), 373-92.

Crawford, R (1991), In the Era of Human Capital, HarperCollins Publishers Inc., USA.

Deiaco, E., E. Hornell, and G. Vickery (1990), Technology and Investment: Crucial
Issues for the 1990's, Pinter Publishers, London.

Dennison, E. (1985), Trends in American Economic Growth, 1929-1982, Washington:
Brookings, p. 30.

Dowrick, S. ed. (1995), Economic Approaches to Innovation, Edward Elgar Publishing
Limited, UK.

Doyle, T. (1994), “Firm Size and the Returns to R&D Investment in High-Technology
Industries,” MSU Doctoral Dissertation.

Ebrill, LP. and D. Hartman (1983), "The Corporate Income Tax, Entrepreneurship, and
the NonCorporate Sector," Public Finance Quarterly, v11(4), 419-36.

142

Encaoua, D, B. Hall, F. Laisney, and J. Mairesse (1996), The Economics and
Econometrics of Innovation: An Overview,” from the 10th International ADRES
Conference and NBER.

Freear, J ., J. Sohl, and W. Wetzel Jr. (1994), "Smaller Company & Inna-entrepreneurial
Issues: The Private Investor Market for Venture Capital," Engineering Management
Review, 22(3), Fall, 91-97.

Geroski, P. (1989), “Entry, Innovation, and Productivity Growth,” The Review of
Economic Statistics, 572-578.

Geroski, P., S. Machin, and J. Van Reenen (1993), "The Proﬁtability of Innovating
Firms,” RAND Journal of Economics, 24(2), 198-211.

Ginn, Martin E. (1995), The Creativity Challenge: Management of Innovation and
Technology, Part B (in Monographs in Organizational Behavior and Industrial Relations),
J ai Press Inc., London, v20.

Gompers, Paul A. and Josh Lerner (1998), “The Determinates of Corporate Venture
Capital Success: Organizational Structure, Incentives, and complementarities,” NBER,
Working Paper 6725, Cambridge.

Griliches, Z. (1981), "Market Value, R&D and Patents," Economics Letters 7, 1981,
183-187. Reprinted in Z. Griliches, ed., R&D, Patents, and Productivity, 1984, 249-
252, Chicago: University of Chicago Press.

(1987), "Patent Statistics as Economic Indicators: A Survey," JEL, 28(4), pp.
1661-1707.

Gruber, William H. (1981), The Strategic Integration of Corporate Research and
Development, Amacom, New York.

Guerard, J .B. Jr., and AS. Bean (1998), R&D Management and Corporate Financial
Policy, John Wiley & Sons, Inc., NY.

Hall, Bronwyn H. (1990), “The Manufacturing Sector Master File: 1959-1987,” National
Bureau of Economic Research, Cambridge, MA, Working Paper No. 3366.

(1993), “The Stock Market’s Valuation of R&D Investment During the
1980’s,” American Economic Review, v83(2), 259-264.

(1998), “Innovation and Market Value” NBER.

143

Hall, H., J affe, A. and Trajtenberg (1998), “Market Value and Patent Citations: A ﬁrst
Look,” NBER.

Hayes, Robert H. and SC. Wheelwright (1984), Restoring Our Competitive Edge:
Competing Through Manufacturing, John Wiley & Sons, New York.

, S.C. Wheelwright, and KB. Clark (1988), Dynamic Manufacturing:
Creating the Learning Organization, The Free Press, New York.

Heertje, A. and M. Perlrnan eds. (1990), Evolving Technology and Market Structure:
Studies in Schumpeterian Economics, The Univ. of MI Press, Ann Arbor, MI.

Hines, J. Jr. (1993), "On the Sensitivity of R&D to Delicate Tax Changes: the Behavior
of US Multinationals in the 1980's," in Tax Issues of the 1990's.

Hirschey, Mark (1982), “Intangible Capital Aspects of Advertising and R&D
Expenditures,” The Journal of Industrial Economics, v30, 375-390.

Hirschey, M. and J. Weyhandt (1985), "Amortization Policy for Advertising and
Research and Development Expenditures," Journal of Accounting Research, v23(1), 326-
35.

Hollomon, Herbert J. (1979), “Policies and Programs of Governments Directed Toward
Industrial Innovation,” Technological Innovation for a Dynamic Economy, Pergamon
Press, NY, pp. 292-317.

Holt, K. and J. Seetzen ( 197 5), “Innovation Research at the International Institute for the
Management of Technology”, OMEGA, The International Journal of Management
Science, v3(2), 219-227.

Horwitz, Paul (1979), “Direct Government Finding of Research and Development:
Intended and Unintended Effects of Industrial Innovation”, Technological Innovation for
a Dynamic Economy, Pergamon Press, NY, pp. 255-291.

Hulten, C. and J. Robertson (1984), "The Taxation of High Technology Industries,"
National Tax Journal, v37(3), 327-46.

Hung, NM. and N. Van Quyen (1994), Dynamic Timing Decisions Under Uncertainty,
Springer-Verlag Berlin Heidelberg.

Johnson, L. and B. Pazderka (1993), “Firm Value and Investment in R&D,” Managerial
and Decision Economics, v14, 15-24.

144

Jorde, M. and D. Teece, eds. (1992), Antitrust, Innovation and Competitiveness, Oxford:
Oxford University Press.

Kamien, M. and N.L. Schwartz (1982), Market Structure and Innovation, Cambridge
University Press, New York.

Klepper, 3- (1996), “Entry, EXit, Growth, and Innovation over the Product Life Cycle,”
American Economic Review, v86(3), pp. 562-83,

Kurdas, C. (1994), Theories of Technical Change and Investment: Riches and
Rationality, St. Martin's Press, New York, NY.

Lee, H. and J. Stone (1994), “Product and Process Innovation in the Product Life Cycle:
Estimates for US. Manufacturing Industries,” Southern Economic Journal, 60(3), 754-
781.

Levin, R.C., W. Cohen, and D. Mowery (1985), “R&D Appropriability, Opportunity, and
Market Structure: New Evidence on Some Schumpeterian Hypotheses,” American
Economic Review, v75(2), 20-24.

Levin, R.C., A.K. Klevorick, R.R. Nelson, and S.G. Winter (1987), “Appropriating the
Returns from Industrial R&D,” Brookings Papers on Economic Activity, 783-820.

Leyden, D. P. and A. N. Link (1992), Governments Role in Innovation, Kluwer Academic
Publishers, Dordrecht.

Mansﬁeld, E. et. a1 (1977), "The Production and Application of New Industrial
Technology”, Norton, New York, pp. 22-32.

Mansﬁeld, 15., “Industrial R&D in Japan and the United States: A comparative Study,”
AEA Papers and Proceedings, 78, 2 (1988), 224—228.

Mansﬁeld, E. (1986), "The R&D Tax Credit and Other Technology Policy Issues,"
American Economic Review, v76(2), 190-94.

McGrath, M. and MN. Romeri (1994), “The R&D Effectiveness Index: A Metric for
Product Development Performance,” Journal of Product Innovation Management, v11,

213-220.

Megnal, P. and M. Klock (1993), “The Impact of Intangible Capital on Tobin’s q in the
Semiconductor Industry,” American Economic Review, v83(2), 265-269.

Mil grom, P. and J. Roberts (1990), “The Economics of Modern Manufacturing:

145

Technology, Strategy, and Organization,” American Economic Review, v80(6), 995-
1 026.

Morroni, M. (1992), Production Process and Technical Change, Cambridge University
Press, Cambridge.

Mowery, David. C. “The Changing structure of US industrial research: Implications for
R&D organization in the Russian Federation,” International Journal of Technology
Management, #5,6,7, 1994.

Nelson, RR. and G. Wright (1992), "The Rise and Fall of American Technological
Leadership," JEL, 30(4) Dec., pp.1931-1964.

Pakes, Ariel (1985), “On Patents, R&D, and the Stock Market Rate of Return,” Journal of
Political Economy, v93(2), 390-409.

Palmer, Karen (1991), “Diversiﬁcation by Regulated Monopolies and Incentives for
Cost-Reducing R&D,” American Economic Review, v8l(2), 266-270.

Pine, B.J. II, B. Victor, and AC. Boynton (1993), “Making Mass Customization Wor ,”
Harvard Business Review, 71(5), 108-121.

Quirmbach, HQ (1993), “R&D: Competition, Risk, and Performance,” RAND Journal
of Economics, v24(2), 157-197.

Ravenscraft, DJ. and FM. Scherer (1987), “Life Aﬁer Take-over,” Journal of Industrial
Economics, v36.

Ricardo, David (1817), Principles of Political Economy and Taxation, 3d ed., 1821,
London: J. M. Dent, 1911.

Roessner, D.J. ed. (1988), Government Innovation Policy: Design, Implementation,
Evaluation, St. Martin's Press, New York.

Rosegger, G. (1980), The Economics of Production and Innovation: an Industrial
Perspective, Pergamon Press, Oxford.

Rothwell, R. and W. Zegveld (1981), Industrial Innovation and Public Policy: Preparing
for the 1980's and the 1990's, Greenwood Press Westport, Connecticut.

Scherer, F.M., (1982), “Inter-industr3I Technology Flows in the United States,” Research
Policy 11, pp- 227-245.

146

, (1984), “Using Linked Patent and R&D Data to Measure Interindustry
Technology Flows,” in Zvi Griliches (ed.), R&D, Patents, and Productivity, Chicago:
University of Chicago Press for the National Bureau of Economic Research.

and D. Ross (1990), Industrial Market Structure and Economic Performance,
3rd ed., Houghton Mifﬂin Company, Boston.

Schumpeter, J .A. (1942), Capitalism, Socialism, and Democracy, Harper & Brothers
Publishers, New York.

Shevlin, T. (1987), "Taxes and Off-Balance-Sheet Financing: Research and Development
Limited Partnerships," The Accounting Review, v62(3), 480-509.

Simon, H.A. (1979), "Rational Decision Making in Business Organizations," American
Economic Review, Vol. 69, No. 4, pp. 493-513.

Slitor, RE. (1966), "The Tax Treatment of Research and Innovative Investment,"
American Economic Review, v56(2), 217-31.

Smith, Adam (1994), The Wealth of Nations, The Modern Library, New York, 1776.

Sobek, D.K., J .K. Liker, and A.C. Ward (1998), “Another Look at How Toyota Integrates
Product Development,” Harvard Business Review, 76(4), 36-49.

Solow, Robert (1957), "Technological Change and the Aggregate Production Function",
Review of Economics and Statistics, 39(3), 312-320.

Swenson, CW. (1992), "Some Tests of the Incentive Effects of the Research and
Experimentation Tax Credit," Journal of Public Economics, v49(2), 203-218.

Tillinger, J. (1991), "An Analysis of the Effectiveness of the Research and
Experimentation Tax Credit in a Q Model of Valuation," Journal of the American
Taxation Association, v13(2), 1-29.

Tirole, Jean (1990), The Theory of Industrial Organization, The MIT Press, Cambridge,
MA.

Utterback, James M. (1979), “The Dynamics of Product and Process Innovation in
Industry,” in Technological Innovation for a Dynamic Economy (Hill & Utterback eds.),
Pergamon Press, NY, pp. 40-65.

Vijverberg, Chu-Pong, and J. Fei (1996), “Production Functions with Factor-Oriented

147

Scale Sensitivity,” The Review of Economics and Statistics”, 309-320.

Zaltman, G., R. Duncan, and J. Holbek (1973), Innovations in Organizations, Wiley-
Interscience Publication, New York.

 

148