how value migrates within an industry architecture:...
TRANSCRIPT
How value migrates within an industry architecture:
Kingpins, bottlenecks, and evolutionary dynamics
C. Jennifer Tae
London Business School
Regent’s Park London NW1 4SA United Kingdom
+44 20 7000 8770
Michael G. Jacobides
London Business School
Regent’s Park London NW1 4SA United Kingdom
December 17, 2011
Keywords: industry architecture, value migration, capability heterogeneity, value chain dynamics
2
How value migrates within an industry architecture:
Kingpins, bottlenecks, and evolutionary dynamics
Abstract
This paper explores the dynamics of value distribution within a sector. It provides exploratory
quantitative evidence on how conditions within the segments of a sector’s value chain affect the
relative profitability of those segments. We consider how value shifts from one part of the value
chain to another by linking two different but causally connected levels of analysis: the inequality
of capability within a segment and the segment’s share of total sector profit. We show that the
presence of superior firms (‘kingpins’) in a segment increases the segment’s share of total sector
value, and establishes the segment as a bottleneck. Although kingpins exert a positive externality
on their direct competitors, a ‘bottleneck’ segment displays more internal inequality, making the
presence of ‘kingpins’ a double-edged sword.
INTRODUCTION
Understanding the foundations and evolution of profitability and value is at the core of strategy
research. The existing literature, drawing on Industrial Organization (I/O) economics (Rumelt,
1991; Schmalensee, 1985) on the one hand, and on the Resource-Based View (RBV) (Barney,
1991; Teece et al., 1997; Wernerfelt, 1984) on the other, has identified several factors that drive
profitability. Related research on patterns of profit accumulation (Dierickx & Cool, 1989; Pacheco
de Almeida & Zemsky, 2007) or on the game-theoretic foundations of competitive advantage
(Bradenburger & Stuart, 1996) has extended our understanding. But, while the questions of how
profits emerge and why they are sustained are still debated (see Lipmann & Rumelt, 2003;
Jacobides, Winter & Kassberger, 2011), our understanding of how profit and value change over
time is even sketchier. In particular, the current literature leaves two issues underexplored: first,
the systematic connection between different parts of a sector or value chain1; and second, the
dynamic aspects of profitability – particularly the factors that dictate how profit and value evolve
by shifting from one part of the value chain to another. This paper primarily focuses on the first
issue, and in so doing, extends the growing literature on dynamics of profitability.
The evolution of the computer sector provides an excellent illustration of the way profit and
value2 shift from firms specializing in one activity (computer assemblers) to those specializing in
other activities (software developers and microprocessor manufacturers). Our focus is on how
profit migrates not only from one firm to the next within a particular setting (i.e. among computer
1 Throughout this paper, we use the words ‘sectors’ and ‘industries’ interchangeably. The term ‘value chain’, (Porter,
1985) was originally used for linked activities or functions within a single firm, but has also come to refer to the
different activities involved to produce a final good / service across firm boundaries. Different segments make up a
vertically related value chain/ecosystem: Each of the various value-adding activities is referred to as ‘a segment’. 2 So far we have used the words ‘profit’ and ‘value’ synonymously, but they are not coterminous. We leave the
manifold problems of the definition of ‘profit’ aside (see Lippman & Rumelt, 2003, for a discussion). Instead, we use
an intuitive sense of profit and value based on market capitalization, which, for all its shortcomings, is a useful starting
point. Market cap is the NPV of future profits (or cash flows), and corresponds to a sense of value that is appropriated
by the firms’ owners, and as such is close to the normal sense of the word ‘profit’ (see Jacobides, Winter & Kassberger,
2011 for a detailed explanation).
2
assemblers), but also across different part of the value chain (i.e. from computer assemblers to
software developers). Popular business analysis has caught on to this phenomenon, and provided
some hypotheses on the dynamics of ‘value migration’ (Slywotzky & Morrison, 1997) and the
idea of ‘profit pools’ (Gadiesh & Gilbert, 1998). However, lack of research has limited our
empirical and theoretical understanding of this process.
In addressing this issue, we draw on the conjecture that the architecture of a sector – that is,
the ‘rules and roles’ that shape relationships between firms – determines its profitability. This
approach allows us to argue that changes in one segment will influence the entire sector, since the
parts of a value chain are interdependent. The conditions within one part of the sector are linked to
that sector’s profitability in relation to the overall industry ecosystem (Jacobides, Knudsen, &
Augier, 2006). We develop this theory further, and undertake an exploratory quantitative analysis
using a large dataset to examine intra-sectoral relationships. Our approach, though exploratory,
could offer a fresh perspective on strategic dynamics. It contributes by broadening the focus from
the conditions of individual firms and the particular segment in which they operate to the broader
value chain or ecosystem (Pisano & Teece, 2007; Iansiti & Levien, 2004).
Returning to the computer example, we look not only at the conditions within the software
development, computer assembly, and microprocessor manufacturing segments separately, but also
at the relative condition of such segments compared to the overall PC sector. In other words, our
dependent variable is the value captured by a particular segment in relation to the overall sector,
while our independent variables describe the different conditions within each segment. This set of
profitability dynamics relates two different but causally connected levels of analysis.
The primary goal of this paper is to explore the relationship between these two levels. We
consider the differences in capabilities between firms within each segment and argue that they
affect conditions not only within a segment, but also between segments. We then test and identify
3
the conditions within each segment that continue to influence its relative share of value vis-à-vis
other segments, both at a given point in time and over time. We look at segments that consist of
firms directly competing among themselves with substitutable products, and those that comprise an
industry’s value chain3.
We begin by reviewing existing research on profitability and recent work on industry
architectures. We then propose and test hypotheses on how competitive conditions observed at
segment level might affect the relative share of value captured by segments along the value chain.
We examine the evidence in the computer sector, which was the inspiration for this research; and
briefly contrast these findings with the automotive sector, which exhibits much less value migration,
to consider boundary conditions of our analysis. Based on the findings, we conclude by linking
back to the literature, outlining limitations, identifying avenues for future research, and discussing
implications for theory and practice.
THEORETICAL BACKGROUND
There has been much debate on the nature and drivers of profit and value. We cannot offer a
comprehensive review, but instead provide a selective analysis of research relevant to our thesis.
Sources of profit and value
Early work on the source of profits drew heavily on ideas developed in I/O economics, which
attributed sustained high levels of profit to either i) firms positioning themselves in a naturally
profitable market or activity (Bain, 1951) or ii) firms changing the structure of the market by
establishing high entry barriers to deter entry by potential competitors (Porter, 1980)4. In other
3 A segment that comprises an industry value chain can also, in and of itself, be an industry in its own right. For
example, mobile handset manufacturing comprises a segment in the mobile telephony industry, but mobile handset
manufacturing itself is an industry with different sets of activities handled by co-specialized firms in segments (e.g.
handset chip manufacturers, handset display manufacturers, and handset assemblers). While recognizing this nested
nature of industry architectures, we focus on the former case in this paper. 4 One notable exception to this is the studies on successive oligopolies and double marginalization (Perry, 1978). In this
literature, the principal concern is whether or not the imperfect market conditions in either the upstream or the
4
words, the market power of firms has been highlighted as the source of profits. Those studies shed
light on the price-setting and strategizing behaviors of firms under oligopolistic conditions, which
lead to non-competitive profits. This approach was complemented by later work, at least partly
implied by Demsetz’s observation (1973) that concentration and profit may be the result of
heterogeneity of firm capabilities. Furthermore, it has become better understood that there are
evident profit differences among firms competing within the same market (Nelson, 1994), as well
as among top-performing firms in nominally unattractive markets. This issue is addressed by
examining how much performance variance can be explained by attributes at the level of the market
(industry), firm, or line of business (Porter & McGahan, 1997; Rumelt, 1991; Schmalensee, 1985).
A stream of research under the Resource-Based View (RBV) heading has shifted attention
from imperfections in the product market to the inputs – specifically, resources and capabilities –
that explain differential profitability even in the context of full competition. Wernerfelt (1984)
viewed a firm as a bundle of resources that can be used to generate above-average returns. The
RBV’s main argument is that a unique combination of specialized and complementary resources
and capabilities leads to superior performance (Peteraf, 1993; Wernerfelt, 1984) – an insight
originally developed by Penrose ((1959) 1995). In particular, Barney (1991) argued that profits
emerge when a firm is able to acquire and maintain resources that are valuable, rare, difficult to
imitate, and non-substitutable. Another stream of literature looked at the role of firm heterogeneity,
as opposed to ‘owned’ resources: what makes firms different in the first place (Nelson, 1994;
Nelson & Winter, 1982). It considered how idiosyncratic ‘routines’ and firm-specific ‘recipes’ or
technologies allow for profits that are only gradually eroded as capacity expands and as firms try to
build on what they deem valuable (Denrell et al., 2003; Winter, 2003). Extending this idea further,
downstream market induce firms to vertically integrate. It also looks into how upstream concentration affects
downstream concentration. Although this is related to the concern of this paper, the earlier work does not directly
address the evolution of profit along the value chain per se.
5
Teece et al. (1997) introduced dynamic capabilities: firms’ ability to nurture, assess, and reform
their own capabilities. By incorporating this evolutionary aspect into the RBV framework, this
work broadened the focus from static profits to dynamic profitability, suggesting that developing
the competency to change competencies is what enables sustainable profit.
While the I/O economics-based approach and the RBV may differ in their views on the
mechanisms through which firms generate profits, both share one important assumption: that
markets, or segments, are independent of each other. Firms compete only within a market, and it is
their performance within that market, relative to other firms, that determines their profitability. As a
result of this assumption, existing literature principally addresses the question ‘what determines the
profitability of firms within a given setting?’5 It is also worth noting that most of the existing
research focuses on statics. Even the limited research on dynamics (Dierickx & Cool, 1989;
Lippman & Rumelt, 1982) focuses on the issue of ultimate sustainability of profit (Barney, 1991;
Ghemawat, 2005; Peteraf & Barney, 2003). Issues of profit / value migration, in the sense of which
factors drive changes in profits and how those profits shift within a value chain, remain largely
neglected.
Industry architecture and industry bottlenecks
More recent research has acknowledged that many modern industries are characterized by different
firms undertaking different activities along the value chain. For a good number of sectors, changes
taking place in one part of the value chain profoundly affect other parts (see Bresnahan &
Greenstein, 2000, for a verbal account). Sectors such as telecommunications (network operators,
software developers, device manufacturers), pharmaceuticals (biotech firms and drug
manufacturers), and financial services (loan originating and loan servicing) amply illustrate the
5 In microeconomics, general equilibrium models assume that the prices and production of all goods, including the price
of money and interest, are interrelated. However, its focus is on explaining the supply, demand, and price of an entire
6
point. This challenges the assumption that settings are static, and raises the question of how changes
in one segment affect its relative value capture in the sector. Yet while this phenomenon may be
acknowledged, research directly exploring it is still fragmented. This section provides a summary of
some of these fragments.
Some research has argued that interdependence among firms engaged in different parts of
the value chain stabilizes over time and results in one or a few rival ‘platforms’: co-specialized
‘business ecosystems’ each with their own sponsors, orchestrators, and keystone members (Gawer
& Cusumano, 2002; Iansiti & Levien, 2004). Existing studies introduced such ideas by analyzing
real-world observations in depth. Second, and building on those observations, a somewhat more
analytical approach has been advanced under the ‘industry architecture’ concept (industry
architectures being defined as sector-wide templates that circumscribe the terms of the division of
labor).
Recognizing the interconnectedness of the different segments that constitute a value chain,
Jacobides, Knudsen, & Augier (2006), drawing on the literature of innovation (Teece, 1986) and
cooperative game theory (Brandenburger & Stuart, 1996; Dixit & Nalebuff, 1991; Jovanovic &
MacDonald, 1994), argued that the conditions within a segment of the value chain affect that
segment’s share in the total profit within the sector, and determine how profitable it is to operate in
that stage compared to other parts of the chain. Moreover, it is sometimes possible for a firm or a
small number of firms engaged in a particular value-adding activity to shape the industry
architecture to their advantage. The authors referred to this phenomenon as an ‘industry bottleneck’
(2006: 1208)6. They suggested that firms use capabilities not just to compete within a segment, but
economy. The general equilibrium theory was not, to the best of our knowledge, empirically examined in the way this
paper addresses the issue. 6 The term ‘industry bottleneck’ was proposed by Jacobides et al. (2006), but the phenomenon has been mentioned by
others (Baldwin & Clark, 1997; 2000; Ferguson & Morris, 1993; Iansiti & Levien, 2004). The initial discussion of
‘bottlenecks’ is found in the discussion of technological progress, notably in Rosenberg (1969).
7
also to increase the value of that segment within the sector. To the extent that they succeed, profits
will migrate to their segment. This view is consistent with peripherally linked studies on sectors
with complex value chains: Ethiraj (2007) investigated the concentration of R&D efforts and also
identified that such efforts accrue to the ‘bottleneck’ within a modular complex system.
This ‘architectural’ approach brings together components of existing studies to show how
profit gravitates toward a set of firms engaged in one particular activity. The intuition here is that
superior capabilities in one activity enable a firm to change the competitive conditions of the
segment it belongs to – which also could change the industry architecture to its advantage. This is
different from the RBV in that the competition of capabilities takes place not only at
market/segment level (e.g. among mobile handset manufacturers) but also at the value-chain level
(e.g. among mobile handset manufacturers, network providers, content providers and so on).
Insights developed in collaborative game theory (Brandenburger & Stuart, 1996; Dixit &
Nalebuff, 1991; Jovanovic & MacDonald, 1994) have considered similar dynamics. These works
suggest that dominant firms within one segment become more of a bottleneck, since they can
leverage their position of strength over all other participants, who must cooperate with them to
create value. The main finding in collaborative games is that the less replaceable a firm is, the
greater the share of value it can appropriate. This avenue of theory thus offers a formal complement
to the work on industry architectures.
Finally, some work on the emergence of Global Value Chains (e.g. Gereffi et al., 2005) and
the changing behavior in the Port wine supply chain (e.g. Duguid, 2005) has noted the importance
of power along the value chain as a driver of profit and value. These authors found that some firms
became more important, and reaped higher profits, by guaranteeing the quality of the final good
produced by the value chain. The firms, but also the particular segment that act as guarantor of
quality by driving customers' perceptions charge accordingly, like Schneider-Kreuznach in the
8
camera lens sector. Which set of companies, and what part of the value chain becomes the
guarantor is a matter of path-dependency: For Port wine, it is shippers, but it is grower/bottlers for
Bordeaux and intermediate commerçants for Burgundy wine. At times, strategic battles for this
quality control may erupt: The ‘Intel Inside’ marketing campaign, which was not only an effort of
Intel to outcompete AMD, but of microprocessors to overshadow computers (and their makers) as
the hallmarks of quality provides a classic example from our setting.
In sum, despite substantial evidence suggesting strong interdependence among different
activities within a sector and structured causal patterns to ‘share of value captured’, most existing
theories are based on the implicit assumption that markets corresponding to different activities
along the value chain are independent. While some recent research has looked at the importance of
profit migration, it has rarely been accompanied by empirical results. Some work has emerged at
the qualitative level (e.g. Dupeyre & Dumez, 2009; Ferraro & Gurses, 2009), but the argument that
conditions within a part of a value chain affect the share of profit to the whole remains more of an
intriguing conjecture than a documented finding.
THEORY DEVELOPMENT
In this paper, we want to both formalize and advance theory, and perform some exploratory
quantitative research. We examine whether different conditions within each segment of an industry
will affect the relative proportion of value (i.e., the NPV of future profits) that each segment
captures. For example, we examine how the competitive conditions within the microprocessor
manufacturing segment affect that segment’s share of value captured within the computer industry.
Specifically, we consider how conditions in value (market capitalization) and technological
investment (R&D expenses) within a segment affect the segment’s share of market capitalization
within the sector. This allows us to identify the mechanisms through which firms are able to reap
more profit for their segment relative to the whole industry (and not just for themselves).
9
The starting point for our argument is heterogeneity between and within segments.
Beginning with segment-level value (market capitalization), we would argue that a firm’s capability
is a key driver of its profitability. A firm with superior idiosyncratic capabilities can outperform its
competitors in the segment, enjoying higher profitability. The novel feature of our argument is that
such a firm uses those capabilities to manage the way firms in other segments are ‘connected’ to
itself, which also affects its segment (Jacobides et al., 2006). At the micro level, we have evidence
that firms do deliberately shape their ecosystem and affect the ‘rules of the game’ (e.g. Duguid,
2005; Ferraro & Gurses, 2009). At a more ‘ecological’ level, we argue that specific distributions of
capabilities, and in particular substantial inequality in terms of capabilities in a segment, is likely to
create benefit for a segment, as it allows one dominant firm (or a few) to shape the rules of the
game to the advantage not only of itself, but also of its segment.7 Given imperfect indicators of
capability difference within the segments, we consider how inequalities in value and in R&D
investment within a segment are connected to that segment’s ability to capture value.
More specifically, we argue that a high degree of inequality in value (market capitalization)
is a good correlate of heterogeneity of capabilities. (We contrast this with inequality in terms of
market share, which we measure and control for – and discuss in the final part of this section.) We
further argue that segments that are less ‘equitable’ in terms of the value distribution are likely to
have key firms (hereafter referred to as ‘kingpins’) that are in a position to turn those segments into
bottlenecks. The kingpin in a relatively unequal segment can turn its segment into the guarantor of
quality by providing the ‘certification function’ (Duguid, 2005) to the end consumers of the value
chain’s product. As the kingpin and its function becomes the certificator of quality, it makes other
segments and their participants less important. This creates an externality, allowing other firms in
7 Of course, even if this were true, though, reliable (let alone comparable, multi-segment) data on capabilities is nearly
impossible to obtain. Given such data limitations, our (exploratory) theory development focuses on the features we can
observe, and thus conduct a ‘reduced form’ analysis, looking at implications of the theory for observable correlates.
10
the segment to benefit to a degree. Leveraging superior capabilities, a kingpin can also set the
industry standards that ensure that the entry barriers around its segment stay high. Symmetrically,
intra-segment competition is intense when there is little difference in capabilities among firms, and
they are likely to concentrate their efforts on competing within the segment and pay little attention
to shaping the industry architecture. In other words, competitive and equitable structures do not
afford many opportunities to shape the sector, and potentially benefit from it. Thus:
Hypothesis 1a. The degree of inequality in value among firms within a segment will be
positively associated with that segment’s share of value within the industry.
Firms’ superior capabilities contribute to the emergence of a new or reshaped industry architecture
from which they and their segments benefit the most. One manifestation of this is kingpins setting
the rules of interface between two adjoining segments to their advantage (Jacobides & Winter,
2005), by leading in standards negotiations or institutional arrangements, as Lew Wasserman and
MCA did for the motion picture sector (Ferraro & Gurses, 2009). With positive feedback, others in
the focal segment follow the precedent to avoid transactional investments – although they cannot
fully replicate the rule-setting firm’s advantage. Over time, the interfaces evolve into a de facto rule.
The lock-in (Schilling, 2000) to a specific relationship by dint of high switching costs can give the
same result. Even with little or no switching cost, firms may be reluctant to seek alternatives due to
the lack of information that would help them evaluate ability to serve the need (Shapiro & Varian,
1998). Stakeholders’ view of who guarantees quality becomes resilient to change due to the
inherent information problem, despite efforts of firms in other segments to take over the role. This
increasing ‘embeddedness’, coupled with the efforts of the designer of the current architecture to
maintain the status quo, e.g. by shaping the regulatory regime to its benefit, enables the segment
with more inequality in value to sustain its position. Thus:
Hypothesis 1b. The above relationship will hold with positive lags.
11
The huge difference in value among computer assemblers, e.g. IBM and Sperry Rand, in the early
stages of the sector’s history, and the dominance of IBM in the sector through its ability to shape
collective outcomes during that period illustrates the hypotheses above.8
The analysis above suggests that kingpins should be able to extract value from the
ecosystem that not only accrues to themselves, but also creates some real externalities for the other
firms in the segment. Thus, our empirical testing will consider whether kingpins make their
segment (themselves included) better off; and whether, even when we exclude the kingpin from the
analysis of relative value, the firms in the segment are still seen to benefit from its presence. We
will refer to these tests as the ‘weak test’ and ‘strong test’ of H1a and H1b.
In addition to measuring differences in capabilities, as proxied by the value they can create,
we can also look at some of the inputs that give us a sense of differential capabilities. Technological
investment is one such input. By examining variations in the distribution of such investment,
measured by R&D spending, we can get a sense of unevenness in the distribution of capabilities.
Inequitable distribution means that a kingpin (or set of kingpins) can benefit by exerting an
externality that allows it to make its own part of the value chain into a bottleneck. So, in line with
Hypothesis 1a, we can expect that:
Hypothesis 2a. The degree of inequality in technological investment among firms within a
segment will be positively associated that segment’s share of value within the industry.
Inequality in technological investment is not just an indication of inequality in capabilities. Because
it affects the future productivity landscape, technological investment also changes the dynamics of
the segment itself. So the more unequal the technological investment, the more that segment will
8 Of course, the dispersion in market capitalization (as contrasted with dispersion in market share) can be an indication
and a consequence of the fact that some firms (in this case, IBM; in the software sector of the 1990’s, Microsoft) is an
indication of the power of one firm. This is consistent with our view, and helps to sharpen the argument of how
conditions within a segment (e.g. a kingpin’s share) are related to how that segment participates in the sector-wide
distribution of value.
12
benefit in the future, all else equal. Given the time-lags of technological investment, we would
expect that the effects of heterogeneity in capabilities will take some time before the segment can
establish itself as a ‘bottleneck’. We can thus expect that:
Hypothesis 2b. The above relationship will hold with positive lags.
The landscape of the semiconductor manufacturing segment, comprising manufacturers of memory,
microprocessors, and IC (integrated circuit) chips, illustrates the logic: the heavy R&D by Intel in
microprocessors in the early stages of the sector’s history not only cemented its dominance in
microprocessors, but also positioned the entire semiconductor manufacturing segment to its
advantage along the value chain of the sector through its standards’ leadership. In contrast, the
aggressive yet largely homogeneous level of R&D of chipmakers in memory and IC products has
led to their ultimate commoditization and inability to shape the sector to their advantage.
Per our argument above, kingpins should be able to shape their ecosystem more effectively
when rules are more readily shaped by their activism; and high-technology sectors offer
disproportionate opportunities for shaping ecosystem dynamics. So, all else being equal, we should
expect that inequality in capabilities between firms in a segment that is technologically advanced
‘buys’ that segment an even greater advantage (i.e., turns it into a bottleneck). Issues of standards,
interconnection, etc., become ever more important in these cases, so that the existence of kingpins,
combined with the relatively high technological intensity of the segment, can make their part of the
value chain into a bottleneck – both short-term and long-term. This leads us to predict that:
Hypothesis 3a. The degree of inequality in value will positively interact with the level of
technological investment within a segment, to increase the share of that segment’s value
within the industry.
Hypothesis 3b: The above relationship will hold with positive lags.
13
Finally, we will consider the reverse causal dynamics: That is, we will consider to what extent a
part of the value chain being a ‘bottleneck’ dynamically affects the inequality of value within that
segment. Consistent with the industry architecture speculations, we would expect that the more a
segment becomes a bottleneck, the more power is wielded by the kingpins that dominate it. For
example, the more Microsoft makes software a bottleneck, or the more Google makes online search
a bottleneck, not only do all software makers and all online search firms benefit from the
strengthening of their segment; they also pay a price by living in a segment that becomes
increasingly unequal – more dominated by kingpins. That is, seen from the kingpin’s perspective,
inequality leads to even greater inequality within the segment over time, i.e. there is a kind of
‘Matthew effect’ (Merton, 1968). This leads us to suggest that:
Hypothesis 4a. The share of value each segment captures from the sector will have a
positive lagged correlation with the inequality of value of firms in that segment.
In addition to expecting value capture by segment to affect within-segment value inequality, one
might also expect that such dominance might affect the inequality in technological investment,
leading us to expect that:
Hypothesis 4b. The share of value each segment captures from the sector will have a
positive lagged correlation with the inequality of technological investment of firms in that
segment.
Before describing our empirical design, we note another possible driver of the relationships
between the variables we are interested in, which we will control and thus test for. ‘Traditional’ I/O
economics can lead us to expect, even in the absence of any of the relationships noted above, that
there will be a correlation between market power or oligopoly within segments and the their ability
to capture value within the sector. First, the more firms exist in a segment, the more competitive it
becomes. Likewise, the lower the market concentration and the less likelihood of collusion (Bain,
14
1951), the lower the profits. Thus both market concentration (in terms of sales) and number of
firms should be correlated to the share of value captured – an intuition shared by the qualitative
discussions of oligopolies and oligopsonies along a sector (Bresnahan & Greenstein, 1991). Extant
theory would also lead us to expect that the structural features of each segment that reduce
competition – what Sutton (1991) calls ‘Endogenous Sunk Costs’ (ESC) – should lead a segment to
have a higher share of total value.
EMPIRICAL DESIGN
We conducted an exploratory quantitative analysis. The objective of this analysis was not to
identify and prove/disprove a mechanism. Rather, it was to illustrate the theory advanced in the
previous section by seeing numerical evidence consistent with the discussion of the phenomenon in
the popular press. Our goal was to advance and probe, not test new theory, and see how our analysis
could help explain dynamics in a sector that has attracted much discussion and analysis in the past
decades in academia and popular press alike.
Setting
Our choice of the computer industry as a setting was predicated on its interest, as opposed to its
representativeness (Firestone, 1993). That is, we selected it because we have observed a dramatic
shift in value distribution in the sector. The percentage of market capitalization of firms in NAICS
codes 334111 (computer manufacturing) and 511210 (software developers) as a proportion of total
sector value underwent dramatic change: from 79% to 8% and from 0.01% to 31%, respectively
between 1978 and 2005. This sector thus allowed us to study which segment-level conditions affect
relative value capture, and provide the requisite variation for this exercise to be of interest. That
being said, it is worth adding that this sector is very important to the US economy, accounting for
9.4% of the total manufacturing value add in 2007 (Bureau of Economic Analysis). Also, the end
products are the product of sophisticated manufacturing, involving myriad components and parts,
15
often borne of intense R&D activities, which makes computing similar to other sectors (such as
mobile telecommunications and media) where value is seen to be migrating in today’s economy.
However, as a robustness check, we did compare our findings to a sector whose structure didn’t
change, automobiles, and report our comparisons as a robustness check.
Data
The data cover the period 1978–2005. Our data drew on the dataset originally gathered by Baldwin,
Jacobides, and Dizaji (2006), but was substantially cleaned and checked for COMPUSTAT issues.
The data-gathering process was organized into three different stages. First, by identifying the
relevant NAICS/SIC codes, a model of the industry value chain was constructed. We identified
relevant codes by i) consulting the descriptions of each code listed in NAICS 1997/2002/ 2007
manuals available from the US Census Bureau, ii) tracing the NAICS codes of leading firms in the
industry such as Microsoft, IBM, and Intel, and iii) identifying all NAICS codes of firms that have
the word ‘computer’ in their business descriptions. We consulted industry experts and academics
with the preliminary list of relevant NAICS codes to avoid both Type 1 and Type 2 errors.
Once we had modeled the value chain for the industry, we obtained a list of all firms, both
active and inactive, from COMPUSTAT’s North America database using the conditional statement
section, which allowed us to identify firms belonging to each segment. The combination of the lists
of firms belonging to each NAICS code, therefore, represented the entire population of publicly
traded firms in the computer industry9. The aggregated list was then used in COMPUSTAT North
America’s segment search to extract each firm’s numerical data, which formed the basis for the
construction of our variables. We extracted the following raw data: primary and secondary NAICS
9 Private firms are not included in the data due to data limitations. This is one of the limitations of our study.
16
codes for each firm-year10
, market capitalization, sales, total assets, current assets, current liabilities,
long-term debt, R&D spending (all million USD) and the number of employees (in thousands) for
each firm-year. Except for sales, which are reported at segment level (by NAICS codes), other
numbers are reported at firm level. For firms that participate in more than one segment, we
weighted the firm-level number by each segment’s sales amount. Firm-level data were adjusted in
this fashion before being aggregated up to the segment-level data for analysis.
Variables
The section below explains how we operationalized the constructs. Tables 1a and 1b present the
summary statistics for the variables used to test our hypotheses. The unit of analysis in this paper is
segments.
Dependent variable
The dependent variable is the percentage of market capitalization (segment to the sector)11
for each
segment year. We first adjusted the market capitalization of firms by sales ratio of segments in
which each firm participates. We calculated each segment’s market capitalization by summing the
adjusted market capitalization amount of all participating firms within each segment by year. We
then added the market capitalization amount of all segments comprising a sector by year to derive
the market capitalization of the industry. Dividing each segment’s market capitalization by the
industry’s total market capitalization per year yielded the dependent variable. We also created three
lag variables (representing one, two, and three years past the base year). In addition, we constructed
two other dependent variables (the three- and five-year moving averages of a segment’s market
capitalization) to ensure the robustness of the results. For the final set of hypotheses (H4a and H4b),
10
NAICS was introduced in 1997 and pre-1997 data only have SIC codes. We used the NAICS-SIC correspondence
tables and the descriptions of the business given by firms to ensure consistency in the data. 11
Using the percentage, instead of the amount, is necessary as it highlights the relative nature of value: we are not
interested in each segment’s market capitalization amount, but in its share of value within the sector.
17
we used the inequality of value and technological investment within a segment as our dependent
variable, at time lags of t+1 to t+3.
To consider the ‘strong test’, we excluded the kingpin from our analysis of the share of
value as a dependent variable. That is, we considered how the inequality of the capability
distribution in a segment affected the relative share of the segment over the entire sector, when we
omitted the value of the kingpin. This way, we avoid having the benefit which accrues to the
kingpin being counted as the benefit of the segment, and can see whether the rest of the segment
benefits from the strength of the kingpin. We identified the kingpin in two ways: i) the firm with the
highest sales and ii) the firm with the highest pro-rated market capitalization.
Independent variables
Inequality in value and technological investment. Inequality was calculated using three different
methods: the kingpin’s share of market capitalization or R&D spending in each segment-year; and
the segment’s Herfindahl Index, and Gini coefficient for these measures (and not for sales). We
measured value using the market capitalization of firms in each segment by year. For technological
investments, we used the amounts of R&D spending for firms in each segment by year. High values
for the kingpin’s share, Herfindahl Index, and Gini coefficient mean there is a major inequality in
value or technological investment among participating firms in a segment in a given year.
Interaction term. To test Hypotheses 3a and 3b, we multiplied the average R&D spending of
each segment by the three inequality measures in value of each segment by year.
Control variables
All our models include a set of control variables. The most theoretically significant controls are the
Herfindahl Index of sales and the mean of R&D spending, which can be expected to be associated
with share of value per the ‘traditional’ I/O view, as noted earlier. To account for the size of the
segment, we included both the number of firms and the total number of employees. We also
18
included three additional control variables: asset efficiency (a segment’s total sales divided by its
participants’ total assets); fixed assets (the difference between the sum of the segment’s total assets
and the sum of the segment’s current assets); and cost of entry (the ratio of the sum of the segment’s
total assets to the total number of the segment’s employees). These control variables helped us rule
out the possibility that the (perceived) valuation of some segments (by market participants) might
be driven by i) operational effectiveness and/or ii) capital/asset intensity, which can function as
effective barriers to entry.
PLACE TABLE 1 ABOUT HERE
Robustness checks
We carried out a number of robustness checks. First, we used different dependent variables (three-
and five-year running averages of a segment’s total market capitalization) to confirm that the results
we obtained from our principal dependent variable (percentage of market capitalization of a
segment to the sector) were not driven by our operationalization. We also substituted firms’ market
capitalization with Tobin’s Q in calculating inequality in value to ensure that our choice of
measurement did not affect the results. In calculating Tobin’s Q, we used the approximate Tobin’s
Q formula.12
Finally, we also ran the model using data aggregated at both five- and six-digit NAICS
codes as segments to determine whether the scope of the segment affected our results. The results
were not affected, and as such we only reported results from the models where percentage of market
capitalization was the dependent variable and the five-digit NAICS code data was used.
ANALYSIS
Methods
12
Whereas the original formula to calculate Tobin’s Q (Lindenberg & Ross, 1981) is very challenging, Chung & Pruitt
(1994)’s formula allows an approximation using more easily available information. It is defined as (market
capitalization – current assets + current liabilities + long-term liabilities-book value)/total assets-book value. The
difference between the Qs calculated from the two methods was empirically tested and the results show that the
approximate Tobin’s Q formula yields largely similar numbers.
19
We specify a segment’s value as a linear function of the explanatory variables: the share of a
segment’s market capitalization in a sector = f (inequality in value, technological investments, and
sales of each segment). Because we are using panel data, it is possible that the error terms will not
be independent across time or within segments (Greene, 2008). There are potential time-dependent,
macro-level factors that could affect the profitability of each segment. Likewise, because several
firms in the sample are active in more than one segment, the errors could be correlated between
segments if some firms perform differently from others due to systematically better management.13
Because we are unable to identify and measure the effects described above, there is potential for a
systematic component to be embedded in the error term, which violates OLS assumptions (Kennedy,
2003). Fixed or random effects may be used to correct for violations of this sort (Greene, 2008).
Because we are interested in how the relative value share of segments changes over time, we use
fixed-effects models (fixed by segment) through which we conduct within-segment estimations.
Not only does the research question point to a fixed-effects model, it also offers an efficient means
of dealing with non-constant variance of the errors, i.e. heteroskedasticity, stemming from the
cross-sectional and temporal aspects of the pooled data. The Hausman test results also supported the
use of the fixed-effects model in place of the random-effects model.
RESULTS
Table 2 reports the results from Fixed Effects GLS estimators for the computer sector. Table 3
reports results from the testing of lagged relationships. Note that each table contains one
(contemporaneous) to three (lagged) results, and as such we report results from four different sets of
regressions – which substantially increases our confidence in our findings. The results show that
there are regularities in the relationship between different competitive conditions within a segment
13
We construct independent variables using the segment level data instead of overall firm data: however, this does not
completely rule out the possibility that the unobserved characteristics affect those segment-level observations to
systematically differ from other firms within segments.
20
and the segment’s relative share of value within the sector. When we only included the control
variables, only two variables were significant: the number of firms and the sum of all employees for
the weak test, and the number of firms and fixed assets for the strong test.
Inequality in value. We find support for H1a, in which we predicted the positive relationship
between the inequality in value among firms within a segment and the segment’s value within the
industry. Kingpin’s share is the strongest predictor of the segment’s share of value, followed by
Herfindahl Index. Gini coefficient does not show any relationship with the segment’s share of value.
We also find support for H1b, which examines the above relationship over a lagged period. We find
support for H1b from t+1 to t+3 with the kingpin’s share and Herfindahl Index. In contrast, Gini
coefficient is never significant and the signs of coefficient are positive only in t+1.
In terms of the ‘strong test’ of H1a and H1b, the correlations reported in Table 4 show that
when we exclude the kingpin from the dependent variable, it is not clear whether the remainder of
the firms in a segment can benefit from the existence of a kingpin – that is, there is no clear sign of
externality. This may be due to the fact that the kingpin’s efforts help the segment including itself
more than they do so excluding itself.
Inequality in technological investment. We find partial support for H2a. All three variables
have the positive sign, as expected, but only the kingpin’s share is statistically significant. H2b
yields similar results. Only the kingpin’s share remains significant throughout the period. Moreover,
the sign of coefficients turns negative from t+2 for Gini coefficient.
In terms of the ‘strong test’ of H2a and H2b, the correlations reported in Table 5 show that
when we exclude the kingpin from the dependent variable, the remainder of the firms in a segment
can still benefit from the existence of a kingpin – that is, there is a clear sign of externality.
Regardless of whether kingpins were identified using their pro-rated market capitalization or their
sales, kingpin’s share and Gini coefficient have a positive externality on the segment. The effect
21
continues from t+1 to t+3. Herfindahl Index, however, is never significant. The contrast in results
between H1a/H1b and H2a/H2b may be attributed to the way in which kingpins turn their segments
into bottlenecks, e.g. more with their technological prowess than profit per se.
PLACE TABLES 2, 3, 4 AND 5 ABOUT HERE
Interaction effect. We find weak support for H3a where we examine how the inequality in value
interacts with the mean of technological investment in affecting the segment’s share of value. We
find support for H3a with only one variable: one that uses the kingpin’s share. As for H3b, we see
different patterns in lagged relationships. We find strong support for H3b from t+1 to t+3 with two
variables: one that uses the kingpin’s share and another using Herfindahl Index. However, we find
opposite results to what we expected with the variable that uses Gini coefficient from t+1 to t+3.
Feedback loops. Table 6 shows the results for the impact of value capture within a segment
on the inequality in value for that segment over time. We find weak support for H4a. Only the
kingpin’s share of value within a segment in t+1 has a positive correlation with the segment’s value.
Similarly, we find continued and consistent, albeit weak, support for H4b, for which we predicted a
positive relationship between a segment’s share of value at t=0 and the inequality in technological
investment in subsequent time periods. It is only the kingpin’s share that has a positive sign and is
statistically significant from t+1 to t+3.
In addition, we analyzed the effect of changes in the number of participants on the value
capture within a segment, as changes to value capture of a segment can induce firm entry or exit.
The results show that the changes in the number of participants have a positive effect on the value
capture of the segment, conversely to what traditional theory would predict. Similarly, the
Herfindahl Index of sales (which measures concentration) is not statistically significant.
PLACE TABLE 6 ABOUT HERE
22
Summing up, Table 7 shows the impacts of inequality in value and technological investment
in the segment, and the interaction between inequality in value and the mean of technological
investment as a predictor of the share of a segment in the entire sector. We observe distinct patterns
in how inequality observed among firms within a segment affects the segment’s share of value. The
choice of the measurement (the kingpin’s share, Herfindahl Index, or Gini coefficient) did not
significantly affect our results, although some differences were present. The fact that the kingpin’s
share is the most significant and robust result is consistent with our theoretical expectations; and the
fact that multiple measures of inequality all seem to point to the same direction demonstrate
robustness. The coefficients of one of our controls, mean of R&D spending, were positive and
significant in the weak test as the standard theory predicts. However, they turned negative in the
strong test, implying that there really is a kingpin effect. Furthermore, the fact that the sales
concentration neither plays the expected role nor becomes statistically significant increases our
confidence in the results.
From value migration to value stability: Putting our theory to a stringent test
As mentioned earlier, the choice of this sector was predicated on the analysis of value that had
migrated. Yet we also wanted to consider a sector with the opposite features – i.e. a sector where
value distribution hardly budged. The automobile sector fit that particular bill, and while we knew
that this was a sector with a much slower pace of structural evolution, we ran the same analyses.
Perhaps unsurprisingly, given the lower variance in terms of changes in inequality in the sector over
time, the inequality in value or technological investment within a segment did not affect the
segment’s share of value along the value chain. Arguably, in sectors where change is slow, or where
a kingpin cannot shape the environment, and where we do not see substantial changes to conditions
within a segment, we should not expect to see the relative value move around in the sector (between
segments). This result, consistent with our theory, places some boundaries on where we expect the
23
results to hold. Perhaps more interestingly, though, H3a and H3b are confirmed. This seems to
further support our view that when technology is important, or more broadly when firms can
reshape their sector, kingpins can help their segments benefit. Finally, even in the automobile sector
we find evidence of the Matthew effect – suggesting that inequality can at least help the dominant
firms to achieve an even better position in their segment.
PLACE TABLE 7 ABOUT HERE
Illustrating our Results
While our paper focuses on the quantitative evidence in the computing sector, we wanted to
illustrate the mechanisms we referred to through one concrete example from our sample. Consider,
in particular, NAICS 334112, which consists of firms primarily engaged in manufacturing computer
storage devices such as hard-disk drives (HDDs), CD-ROM drives, and floppy-disk drives that
allow the storage and retrieval of data. The case of computer storage device manufacturers, and
HDD manufacturers in particular, illustrates how the degree of inequity in capabilities among direct
competitors affects the segment’s share of value in the sector. Manufacturers of HDDs compete on
features such as data density and latencies, as well as smaller form factors that enable the reduction
of physical sizes in computing devices – all of which require intense technological knowledge. The
level of R&D investments among firms has also remained largely homogeneous, lest they put both
their profitability and survival at risk. Due to the intense competition among firms within the
segment, even those with somewhat superior capabilities (e.g. Western Digital and Seagate) could
not use their skills to establish an industry standard, or interfaces that could help shape the sector to
their advantage. For example, the ATA/SCSI interface has remained resolutely unchanged for the
past three decades, which benefits only computer assemblers. The relative homogeneity in R&D
investment, which hinders the emergence of kingpins by engendering relative homogeneity in
future capabilities, forced the incumbents to gradually shift their focus from technological prowess
24
to scale and price. It led to continued consolidation in the segment and, as of October 2011, only
three participants remain: Toshiba (10.8%), Seagate (40%), and Western Digital (49.2%). But the
high concentration could not help the segment, since it was the result of the segment’s relative
impotence, and not, per the more traditional economic rationale, an opportunity for pricing power.
The relative share of value of firms belonging to NAICS 334112 has remained low and hardly
changed over time, ranging between 1% and 7%. So while heterogeneity and concentration at the
level of technology or capabilities, and in particular a kingpin’s dominance, would have helped the
segment, the concentration in the segment (as traditionally measured by sales) did not, and instead
concentration was the result of the segment ‘losing out’. Sales concentration was a symptom of
malaise, as opposed to a predictor of success – the phenomenon known as ‘defensive concentration’.
DISCUSSION
Our findings suggest there is a systematic connection between the conditions within a part of the
value chain and the share of value it can capture. While the ‘traditional’ IO explanations fail to
explain the data, we do find that segments with kingpins tend to become ‘bottlenecks’ that capture
more value.
Our findings indicate that capability differences among firms (as proxied by inequality
within a segment, and mostly the dominance of a ‘kingpin’) have a fairly robust relationship with
the share of value each segment captures in its sector. These findings lend support to the emerging
industry architecture literature, and to speculations on how firms shape their sectors, although we
do not and cannot directly test for the underlying mechanisms.14
14
In line with the exploratory nature of this paper we considered if any other explanations could account for this result.
The only explanation that could be consistent with the data is an alternative causal pathway to the same pattern. There
exists a possibility that our results come from a repeated pattern of exogenous innovation that both increases the value
of a segment and affects the degree of inequality in it. For this to happen such innovation would need to arise from the
incumbents --else, entrants would reduce inequality by entering the fray of competition. Furthermore, the innovation
would have to be both beneficial for the segment and differentially beneficial to the most capable or technologically
advanced players. The assumptions inherent in this alternative explanation seem taxing, and they are also broadly
25
We find that the degree of inequality in value within a segment has the strongest positive
effect on the segment’s share of value, and is robust both contemporaneously and over time. This is
in contrast to the effect of market share or average technological intensity, neither of which affects
the share of value each segment captures. This suggests that it is not a market power story, and
makes the findings on kingpins (in terms of capabilities or technology) more noteworthy. Our
analysis also indicates that greater value accruing to a segment dynamically begets even greater
within-segment inequality in value and technological investment in the future. This suggests the
presence of a Matthew effect, and is worthy of further study.
Collectively, our findings show that inequality of capabilities, and not market concentration
or market share dominance, and inequality of technological prowess rather than technological
intensity is what makes a segment more valuable along the value chain. Furthermore, the more a
segment becomes a bottleneck, the more unequal that segment becomes. This demonstrates that
heterogeneity in capabilities matters, not only in terms of individual firms’ profitability, but also in
terms of the segment’s profitability compared to the sector as a whole. This indicates that the
presence of a firm with superior capabilities in a segment can exert positive externalities on its
peers by ‘growing the pie’ that a segment can attract. Inasmuch as a firm with superior capabilities
can pursue tactics such as turning its segment into the guarantor of quality or establishing industry-
wide interfaces, other firms in the segment can gain, although not as much as the kingpin. The
analysis of the ‘feedback loop’ suggests that positions of power along a value chain serve to
enhance the dominance of a few firms, so that the bottleneck bestows on the kingpins the ability to
further tighten their grip, even in such fast-moving sectors as computers. So while other firms, in
the short term, see their plight improved by a kingpin, with their segment growing in importance,
inconsistent with the strong finding that the more firms (and entrants) exist in a segment, the greater the share of the
segment on the value of the sector.
26
the kingpin over time takes on more and more of the value of the segment, making it a double-
edged sword. Such a view result contrasts subtly with the ‘winner takes all’ hypothesis (e.g. Arthur,
1989; Kelly, 1998).15
However, this is consistent with qualitative studies (Ferraro & Gurses, 2009;
Depeyre & Dumez, 2010) and it adds flesh to the anecdotal discussion in the popular press about
how profit pools migrate (Gadiesh & Gilbert, 1998; Slywotzky & Morrison, 1997).
This paper has provided a qualitative empirical analysis that aspires to complement existing
theory and empirical studies. We focused on a sector where substantial value migration had
happened so as to explore the factors underpinning value migration. The juxtaposition with the
automobile sector suggests that there are sectors whereby structures shift more slowly, so that
kingpins cannot ‘buy’ so much advantage in their sector. When (in a cross-sectional sense) kingpins
can lead to value migration, or when they cannot, emerges as a fascinating area for future research,
well outside the scope of this study.
Contributions
Our contribution is to highlight previously unexplored patterns and to advance one particular theory
consistent with the patterns. This complements existing research in a number of ways.
First, our analysis of how the changing conditions within different segments affect the value
distribution within a sector advances the existing studies on industry evolution. Our claim is that
while we know a lot about the segment-by-segment dynamics of entry and exit (Klepper, 1996,
1997), ‘shakeouts’ (Abernathy & Utterback, 1978; Greenstein & Wade, 1998; Klepper & Simons,
2005), and industry structure (Jovanovic & MacDonald, 1994; Malerba & Orsenigo, 1996; Nelson,
1994), we have only a vague and often impressionistic account of the entire sector, or ecosystem.
We thus hope that our consideration of the entire industry architecture, and the relationships within
15
According to winner takes all argument, a firm with superior capabilities, whether intentionally or unintentionally,
drive out its direct competitors in the segment both in terms of sales and profits, inducing their exit, and eventually
27
it, provides an additional angle on the understanding of industry demographics. Our explicit focus
on issues of heterogeneity within segments as a driver of sector-wide dynamics is aligned with
research that looks at not only the drivers, but also the implications of differences of capabilities in
a sector (Jacobides & Winter, 2011; Syverson, 2011).
Second, this study complements work on industry evolution, which has mainly looked at
changes in ‘who does what’ (e.g. Sturgeon, 2002; Cacciatori & Jacobides, 2005; Gibbon & Ponte,
2006) by looking at the dynamics of ‘who takes what as a result of the dynamics of segments along
the value chain’. Langlois (1992) has looked at how the transactions in which firms engage affect
what the firms do and how the industry will evolve. Similarly, Argyres & Liebeskind (1999) have
proposed how contractual obligations resulting from different transaction costs in the past can
impose different levels of difficulty in what firms do in the future, thus affecting the future shape of
the industry. We view our research as an extension of this tradition. By exploring how different
competitive conditions in segments (which are the result of different transactional conditions faced
by individual firms in those segments) affect the share of value it gets in the sector, we offer a first
step to what we think is a promising research program, on both the theoretical and empirical levels.
Third, we complement recent work on industry architecture and global value chains (Duguid,
2005; Ferraro & Gurses, 2009; Gereffi et al., 2005; Teece & Pisano, 2007), by shifting from the
individual, micro-level analysis of how firms shape and change their entire ecosystem to a large-
scale analysis. We offer the large-scale quantitative counterpart to these studies, and provide a
template for further investigation of the elusive but important dynamics of value migration.
become a de facto monopoly in its segment. Although we cannot rule out its possibility in a more distant future, at least
in our setting, the presence of a firm with superior capabilities in a segment seems to benefit its direct competitors.
28
Future research
Our initial findings on the relationships between a segment’s condition and its share of value open
up exciting new avenues for research. First, while our paper was focused on sector-wide dynamics,
we clearly need to have the corresponding qualitative analysis in the same setting to shed more light
on the micro-mechanisms. The question here becomes, how exactly do strong players in one
segment of the value chain exert a positive externality over other firms in their own segment?
Second, although we speculate that differences in the relationship between conditions within
a segment and its share of value exist depending on the industry setting, we do not have a clear
understanding of the mechanisms that underpin these differences. As the literature on modularity
has shown (Baldwin & Clark, 2000; Baldwin & Woodard, 2007; MacDuffie, 2008; Sturgeon, 2002),
the ease with which a product or service can be ‘modularized’ and replaced in a value chain affects
profitability. Degrees of modularity or other structural characteristics differ from one sector to
another. Thus, the industry specificity of some effects will influence the degree to which firms can
either free-ride on the fruits of the superior firm’s labor or imitate/replicate its behavior. It will be
interesting to look into whether, and if so, how industry settings moderate the relationships
discovered in this paper.
Third, it would be possible (once the qualitative micro-mechanisms are explicated) to
complement this research with some formal modeling, whether in the CGT tradition (MacDonald &
Ryall, 2004), or through other models of multiple-segment industry evolution that are in the making.
It is possible to create an econometric model and test it on the data as well, although we feel that
such an approach should follow the initial, exploratory phase we currently engaged in. And finally,
it would be good to complement the pilot ‘sectoral’ study with other interconnected sectors – even
though one quickly comes up against data limitations.
29
Limitations
This study has a number of limitations. On the theoretical level, although we identified that
heterogeneity in capabilities among firms is the key driver of value distribution along the value
chain, the question of whether this effect is the result of a conscious strategy remains. We cannot
say whether a particular part of a value chain with a bigger share of value is an unintended
consequence of an individual firm’s pursuit of profit, or something that firms in a segment
deliberately coordinate. By treating industry architecture as given, we also avoid the question of its
endogeneity.
Conceptually, we consider each segment as one entity, and look at the aggregate resolution
of the competitive battle as proxied, indirectly, through the inequality of participants’ capabilities.
Doing so takes our focus away from the struggle within each segment, such as the battle between
different potential solutions, or even industry-wide architectures and the related ‘platform wars’
(Gawer & Cusumano, 2002). Our paper is also agnostic on the sources of capability differences, as
we treat heterogeneity as a given, focusing on its implications rather than its antecedents.
On the empirical level, there are additional limitations. First, as we mentioned, we chose
computers on the basis of theoretical interest, as opposed to generalizability; we looked at a sector
where the phenomenon of value migration did occur, and tried to explain it. This leaves the
question of boundary conditions open. While our inescapably brief discussion of the dynamics of
the automobile sector, notable for its lack of migration, partly addressed this topic, the broader
questions of when we expect kingpins to be able to change their sector remains.
Second, as we consider this sector’s specificity, it is worth noting that the computer sector
has fairly clearly delineated boundaries. This allows us to test the hypotheses in a constrained
setting, characterized by relative stability in terms of the segments that constitute it. These very
merits, which enable large-sample analysis, also entail limitations. In many sectors, such as
30
telecommunications and media, where we are witnessing very substantial value migration, the
nature of the constituent segments changes and evolves over time. This makes empirical analysis
elusive, but also adds a further element of structural change that was absent from our setting, as it is
hard to pin down the segments that constitute the sector. Whether we were wise to choose the
‘fruitfly’ of industry architecture, whose boundary stability facilitates econometrics, or whether we
should be criticized for limiting ourselves with an overly stylized setting remains a matter of taste.
Third, our data also comes with its own limitations. Our analysis does not cover the entire
population of the computer industry, since we only look at publicly listed firms in the US market.
This affects the comprehensiveness of our data: We leave out both i) private firms and ii) firms that
are publicly listed in countries other than the US. Our data include non-US firms with ADRs (e.g.
TSMC), but exclude other firms such as Samsung Electronics. We do not have prima facie concerns
that these exclusions bias our results. And, last but not least, no record in action or secondary data
we could think of would easily redress the problem.
Fourth, we weighed market capitalization and other measures that are identified at firm level
with the sales data to transform and construct the measures to segment-level data. We recognize
that this has shortcomings, as it arbitrarily prorates firms’ value. We think that this arbitrary choice
does not invalidate our results and analysis, since we focus on fixed effects. That is, we consider
how changes in the market capitalization, prorated by sales (even if we assume an arbitrariness in
prorating as a baseline), over time, links to changes in the share of value captured. If anything, a
rough measure in terms of pro-rating should introduce more noise. This makes results in terms of
the fixed effects we find (especially as they are robust in terms of lags and specifications) all the
more interesting. On a pragmatic level, there does not seem to be any other way to study such
interesting empirical phenomena absent a pro-rating rule. Hence, we do use caution, but note that
our results (on inequality measures) are robust.
31
Fifth, in our data, we did not account for diversified firms that are present in multiple
industries, as opposed to being present in multiple segments within a single industry. Texas
Instruments, for instance, manufactures semiconductors as well as mathematical calculators. We
could not control for such firms. We only included the sales data from relevant segments and
partitioned other relevant measures, only observed at firm level, by weighing them with sales.
However, we could not completely rule out the possibility that these firms active in multiple
industries might influence the segments they belong to in an unobserved fashion.
Concluding remarks
Limitations noted above notwithstanding, we think that this empirical analysis helps break new
ground in the study of profit evolution and value migration. This exploratory quantitative study
highlights the effect of heterogeneity in capabilities at the individual firm level on value distribution
at the sector level. We show that heterogeneity in capabilities, reflected in particular in the degrees
of inequality in value (market capitalization), positively affect the segment’s share of value in the
broader sector, both contemporaneously and over time. This suggests that a firm’s superior,
idiosyncratic capabilities can not only positively affect its own value (market capitalization) at a
given time relative to its peers; but also can increase the total ‘pie’ available to the segment, since
they increase the relative share of the segment to the entire industry, making it more of a
‘bottleneck’. We demonstrate that there is a real externality that the ‘kingpins’ in a sector can exert
to the rest of their peers in a value chain. We also show, however, that sectors dominated by
kingpins become increasingly more unequal, making the presence of kingpins a double-edged
sword. Finally, our findings on the role of technological intensity, and the brief comparison of our
findings in computers to those in automobiles, suggest that the settings that are more malleable and
easily transformed afford the kingpins a greater ability to extract value, whereas other settings do
not.
32
Our study sheds light on two facets of profitability that have received scant attention. It
looks at how profitability and value evolve within a broader ecosystem or industry architecture,
taking into account the entire value chain rather than focusing on just one part of it. It also helps us
advance the analysis of the mechanisms through which profits evolve over time, and, as such, offers
a first, exploratory quantitative analysis of how profit pools accrete and migrate along the value
chain. With examples such as Microsoft and Intel, who redefined the profit distribution of the
computer sector in the 1990s, and Google’s recent attempts to redefine it yet again, we should start
considering data, both qualitative and quantitative, on how profit and value shifts in the economy.
By expanding the unit of analysis and examining the dynamics of profitability, we will be able to
obtain a more robust and more representative theory of profitability and its evolution, and our study
has offered a step in this direction.
33
References
Abernathy W, Utterback J. 1978. Patterns of innovation in technology. Technology Review 80: 40-7
Akerlof GA. 1970. The Market for "Lemons": Quality Uncertainty and the Market Mechanism. The
Quarterly Journal of Economics 84(3): 488-500
Argyres N, Liebeskind J. 1999. Contractual commitments, bargaining power, and governance
inseparability: Incorporating history into transaction cost theory. Academy of Management
Review 24(1): 49-63
Arthur WB. 1989. Competing Technologies, Increasing Returns, and Lock-In by Historical Events.
The Economic Journal 99(394): 116-131
Bain JS. 1951. Relation of profit rate to industry concentration: American manufacturing, 1936-
1940. Quarterly Journal of Economics 65(3): 293-324
Baldwin CY, Clark KB. 1997. Managing in an age of modularity. Harvard Business Review 75(5):
84-93
Baldwin CY, Clark KB. 2000. Design Rules: The Power of Modularity. The MIT Press: Boston,
MA Baldwin CY, Woodard CJ. 2007. Competition in modular clusters. Harvard Business School
Barney JB. 1991. Firm Resources and Sustained Competitive Advantage. Journal of Management
17(1): 99
Brandenburger AM, Stuart HW, Jr. 1996. Value-based business strategy. Journal of Economics and
Management Strategy 5(1): 5-24
Bureau of Economic Analysis US. Gross Domestic Product (GDP) by Industry, Vol. 2009:
Cacciatori E, Jacobides MG. 2005. The Dynamic Limits of Specialization: Vertical Integration
Reconsidered. Organization Studies 26(12): 1851-1883
Chung KH, Pruitt SW. 1994. A Simple Approximation of Tobin's q. Financial Management 23(3):
70-74
Cohen WM, Levinthal DA. 1990. Absorptive Capacity: A New Perspective on Learning and
Innovation. Administrative Science Quarterly 35(1): 128-152
Demsetz H. 1973. Industry Structure, Market Rivalry, and Public Policy. Journal of Law and
Economics 16(1): 1-9
Denrell J, Fang C, Winter SG. 2003. The Economics of Strategic Opportunity. Strategic
Management Journal 24(10): 977-990
Dierickx I, Cool K. 1989. Asset Stock Accumulation and Sustainability of Competitive Advantage.
Management Science 35(12): 1504-1511
Dixit AK, Nalebuff BJ. 1991. Thinking Strategically: Competitive Edge in Business, Politics and
Everyday Life (1 ed.). W W Norton & Co: London
Duguid P. 2005. Networks and knowledge: The beginning and End of the Port Commodity Chain,
1703-1860. Business History Review 79(3): 453-466
Dupeyre C, Dumez H. 2009. The role of architectural players on coopetition: The case of the US
defense industry. In LR Frederic, Y Said (Eds.), Coopetition: Winning Strategies for the 21st
Century. Edward Elgar: Cheltenham
Ethiraj SK. 2007. Allocation of Inventive Effort in Complex Product Systems. Strategic
Management Journal 28(6): 563-584
Ferguson CH, Morris CR. 1993. How architecture wins technology wars. Harvard Business Review
71(2): 86-96
Ferraro F, Gurses K. 2009. Building architectural advantage in the US motion picture industry: Lew
Wasserman and the Music Corporation of America. European Management Review 6(4): 233-249
34
Firestone, W. A. 1993. Alternative arguments for generalizing from data. Educational Research.
22(4): 16–23
Gadiesh O, Gilbert JL. 1998. Profit pools: A fresh look at strategy. Harvard Business Review 76(3):
139-147
Gawer A, Cusumano MA. 2002. Platform Leadership: How Intel, Microsoft, and Cisco drive
industry innovation. Harvard Business School Press: Cambridge, MA
Gereffi G, Humphrey J, Sturgeon T. 2005. The governance of global value chains. Review of
International Political Economy 12(1): 78 - 104
Ghemawat P. 2005. Strategy and the Business Landscape: Core Concepts (2 ed.). Prentice Hall:
Englewood Cliffs, NJ
Gibbon P, Ponte S. 2006. Trading Down: Africa, Value Chains, and the Global Economy. Temple
University Press: Philadelphia, PA
Greene WH. 2008. Econometric Analysis (6 ed.). Pearson Education: London
Greenstein SM, Wade JB. 1998. The product life cycle in the commercial mainframe computer
market, 1968-1982. RAND Journal of Economics 29(4): 772-789
Iansiti M, Levien R. 2004. The Keystone Advantage: What the New Dynamics of Business
Ecosystems Mean for Strategy, Innovation, and Sustainability. Harvard Business School Press:
Boston, MA
Jacobides MG, Knudsen T, Augier M. 2006. Benefiting from innovation: Value creation, value
appropriation and the role of industry architectures. Research Policy 35(8): 1200-1221
Jacobides MG, Winter SG. 2005. The co-evolution of capabilities and transaction costs: explaining
the institutional structure of production. Strategic Management Journal 26(5): 395-413
Jacobides MG, Winter SG. 2011. Capabilities: Structure, Agency and Evolution. Forthcoming,
Organization Science
Jacobides, MG, Winter, SG, Kassberger, SM. 2011. The dynamics of wealth, profit and sustainable
advantage. Working Paper, London Business School.
Jovanovic B, MacDonald GM. 1994. The Life Cycle of a Competitive Industry. The Journal of
Political Economy 102(2): 322-347
Kelly K. 1998. New Rules for the New Economy: Ten Radical Strategies for a Connected World.
Penguin Books: New York, NY
Kennedy P. 2003. A Guide to Econometrics. MIT Press: Cambridge, MA
Klepper S. 1996. Entry, Exit, Growth, and Innovation over the Product Life Cycle. The American
Economic Review 86(3): 562-583
Klepper S. 1997. Industry Life Cycles. Ind Corp Change 6(1): 145-182
Klepper S, Simons KL. 2005. Industry shakeouts and technological change. International Journal
of Industrial Organization 23(1-2): 23-43
Langlois, R. 1992. Transaction cost economics in real time. Industrial and Corporate Change 1(1):
99-127
Lindenberg EB, Ross SA. 1981. Tobin's q Ratio and Industrial Organization. The Journal of
Business 54(1): 1-32
Lippman SA, Rumelt RP. 1982. Uncertain Imitability: An Analysis of Interfirm Differences in
Efficiency under Competition. The Bell Journal of Economics 13(2): 418-438
MacDuffie JP. 2008. Technological and organizational barriers to modularity: Persistent Integrality
in the global automotive industry. The Wharton School, University of Pennsylvania
Malerba F, Orsenigo L. 1996. The Dynamics and Evolution of Industries. Ind Corp Change 5(1):
51-87
35
Mancke RB. 1974. Causes of interfirm profitability differences: A new interpretation of the
evidence. The Quarterly Journal of Economics 88(2): 181-193
Merton, RK. 1968. The Matthew effect in science. Science 159(3810): 56-63.
Nelson RR. 1994. The co-evolution of technology, industrial structure, and supporting institutions.
Ind Corp Change 3(1): 47-63
Nelson RR, Winter SG. 1982. An Evolutionary Theory of Economic Change. Harvard University
Press: Cambridge, MA
Pacheco-de-Almeida G, Zemsky P. 2007. The timing of resource development and sustainable
competitive advantage. Management Science 53(4): 651-666
Penrose ET. (1959) 1995. The Theory of the Growth of the Firm. Oxford University Press: Oxford
Perry MK. 1978. Vertical Integration: The Monopsony Case. The American Economic Review
68(4): 561-570
Peteraf MA. 1993. The Cornerstones of Competitive Advantage: A Resource-Based View.
Strategic Management Journal 14(3): 179-191
Peteraf MA, Barney JB. 2003. Unraveling the Resource-Based Tangle. Managerial and Decision
Economics 24(4): 309-323 Porter ME. 1985. Competitive Advantage. Free Press: New York, NY
Porter ME, McGahan AM. 1997. How Much Does Industry Matter, Really? Strategic Management
Journal 18: 15-30
Rosenberg N. 1969. The Direction of Technological Change: Inducement Mechanisms and
Focusing Devices. Economic Development and Cultural Change 18(1): 1-24
Rumelt RP. 1991. How Much Does Industry Matter? Strategic Management Journal 12(3): 167-
185
Schilling MA. 2000. Toward a general modular systems theory and its application to interfirm
product modularity. Academy of Management Review 25(2): 312-334
Schmalensee R. 1985. Do Markets Differ Much? American Economic Review 75(3): 341
Schumpeter JA. 1950. Capitalism, socialism, and democracy (3rd ed.). Harper Row: New York NY
Schumpeter JA. (1934) 1955. The theory of economic development. Harvard University Press:
Cambridge, MA
Shapiro C, Varian HR. 1998. Information Rules: A Strategic Guide to the Network Economy.
Harvard Business School Press: Cambridge, MA
Slywotzky A, Morrison D. 1997. The Profit Zone: How Strategic Business Design Will Lead You to
Tomorrow's Profits. John Wiley & Sons: New York
Sturgeon TJ. 2002. Modular production networks: a new American model of industrial organization.
Industrial and Corporate Change 11(3): 451-496
Sutton J. 1991. Sunk Costs and Market Structure: Price Competition, Advertising, and the
Evolution of Concentration. The MIT Press: Cambridge, MA
Syverson C. 2011 What determines productivity? Journal of Economic Literature 42(2): 326-365
Teece DJ. 1986. Profiting from technological innovation: Implications for integration, collaboration,
licensing and public policy. Research Policy 15(6): 285-305
Teece DJ, Pisano G, Shuen A. 1997. Dynamic capabilities and strategic management. Strategic
Management Journal 18(7): 509-533
Tushman ML, Anderson P. 1986. Technological Discontinuities and Organizational Environments.
Administrative Science Quarterly 31(3): 439-465
Wernerfelt B.1984. A resource-based view of the firm. Strategic Management Journal 5(2): 171-80
Winter SG. 2003. Understanding Dynamic Capabilities. Strategic Management Journal 24(10):
991-995
Table 1. Descriptive statistics and analysis (control variables only)
Table 2. Hypothesis testing: weak test
we ak /s tr on g (s ale s /m ar k e t c ap) te s ts Co e f. Co e f. Co e f.
C o n tro l v a ria b le s
- N u mb e r o f firms 2.75E-04 * * * -2.90E-04 * * * -2.90E-04 * * *
(0.00) (0.00) (0.00)
- S u m o f a ll e mp lo y e e s 6.83E-05 * * -2.05E-05 1.20E-05
(0.00) (0.00) (0.00)
- H e rfin d a h l In d e x (s a le s ) 0.005 -0.008 -0.002
(0.01) (0.01) (0.01)
- M e a n o f R& D s p e n d in g 2.66E-05 -1.16E-04 * -1.20E-04 *
(0.00) (0.00) (0.00)
- A s s e t e ffic ie n c y -1.95E-05 1.21E-04 1.10E-04
(to ta l s a le s / t o ta l a s s e ts ) (0.00) (0.00) (0.00)
- F ixe d a s s e ts -6.41E-06 2.30E-05 * * 2.20E-05 * *
(to ta l a s s e ts - c u rre n t a s s e t s ) (0.00) (0.00) (0.00)
- En t ry c o s t 6.09E-07 2.58E-07 6.09E-07
(to ta l a s s e ts / to ta l e mp lo y e e s ) (0.00) (0.00) (0.00)
C o n sta n t 0.026 * * 0.062 * * * 0.060 * * *
(0.01) (0.01) (0.01)
N 616 484 484
F-v a lu e 11.05 * * * 5.05 * * * 4.77 * * *
Mean S.D. 1 2 3 4 5 6 7 8 9 10 11 12 13
1 Segment's share of value in the sector 0.04 0.12
2 Top firm's share (market capitalization) 0.73 0.27 -0.39
3 Herfindahl Index (market capitalization) 0.61 0.34 -0.43 0.99
4 Gini coefficient (market capitalization) 0.72 0.21 0.11 0.03 0.02
5 Top firm's share (R&D expenses) 0.73 0.29 -0.52 0.80 0.81 -0.05
6 Herfindahl Index (R&D expenses) 0.55 0.4 -0.52 0.81 0.83 -0.05 0.99
7 Gini coefficient (R&D expenses) 0.68 0.23 0.06 0.02 0.00 0.61 -0.04 -0.04
8 Number of firms 32 79.4 0.66 -0.58 -0.60 -0.07 -0.69 -0.67 -0.05
9 Sum of employees in a segment 94.9 266 0.87 -0.49 -0.50 0.01 -0.58 -0.57 0.00 0.71
10 Herfindahl Index (sales) 0.63 0.34 -0.41 0.82 0.85 -0.70 0.80 0.82 -0.65 -0.58 -0.45
11 Mean (R&D expenses) 58.3 163 0.04 0.05 0.06 -0.03 0.02 0.04 0.03 -0.02 0.12 0.08
12 Asset efficiency 2.26 22.8 -0.02 -0.02 -0.02 -0.02 0.02 0.01 -0.01 -0.01 -0.02 0.07 -0.02
13 Fixed asset 382 1058 0.01 0.06 0.08 -0.08 0.03 0.04 -0.03 -0.05 0.10 0.09 0.09 -0.02
14 Entry cost 395 1174 -0.05 0.08 0.08 -0.10 0.10 0.11 -0.03 -0.04 -0.05 0.09 -0.05 -0.01 -0.01
Computer
37
+ (p<0.1), *(p<0.05), **(p<0.01), ***(p<0.001)
* Results of all control variables available upon request.
Computer-NAICS5 Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef.
Independent variables
H1 - Inequality in value
- Kingpin's share (market capitalization) 0.079 ***
(0.02)
- Herfindahl Index (market capitalization) 0.027 *
(0.01)
- Gini coefficient (market capitalization) 0.021
(0.01)
H2 - Inequality in technological investment
- Kingpin's share 0.056 **
(0.02)
- Herfindahl Index 0.009
(0.01)
- Gini coefficient 0.013
(0.01)
H3 - Interaction effect
- Kingpin's share * mean R&D spending 2.29E-04 **
(market capitalization) (0.00)
- Herfindahl Index * mean R&D spending 7.14E-05
(market capitalization) (0.00)
- Gini coefficient * mean R&D spending -4.16E-05
(market capitalization) (0.00)
Control variables (not all reported)*
- Number of firms 0.003 *** 2.78E-04 *** 2.68E-04 *** 3.02E-04 *** 2.80E-04 *** 2.74E-04 *** 2.58E-04 *** 2.71E-04 *** 2.71E-04 ***
(0.00) (0.00) (0.01) (0.02) (0.00) (0.00) (0.00) (0.00) (0.00)
- Herfindahl Index (sales) -0.040 ** -0.011 0.014 -0.022 -0.001 0.010 -0.005 0.002 0.004
(0.02) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01)
- mean R&D spending 3.51E-05 2.71E-05 3.25E-05 2.81E-05 2.63E-05 3.02E-05 -1.70E-04 * -3.35E-05 4.46E-05
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
N 605 616 604 601 616 599 605 616 604
F-value 12.38 *** 10.34 *** 9.76 *** 10.75 *** 9.72 *** 9.52 *** 10.62 *** 10.00 *** 9.54 ***
Table 3. Hypothesis testing: strong test
+ (p<0.1), *(p<0.05), **(p<0.01), ***(p<0.001) * Results of all control variables available upon request.
Kingpin-sale/market capitalization Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef.
Independent variables
H1 - Inequality in value
- Kingpin's share (market capitalization) 0.021 0.033
(0.02) (0.02)
- Herfindahl Index (market capitalization) 0.007 0.009
(0.02) (0.02)
- Gini coefficient (market capitalization) 0.029 0.026
(0.02) (0.02)
H2 - Inequality in tech. investment
- Kingpin's share 0.056 0.07**
(0.02) (0.02)
- Herfindahl Index 0.013 0.019
(0.02) (0.02)
- Gini coefficient 0.033* 0.027+
(0.02) (0.02)
H3 - Interaction effect
- Kingpin's share * mean R&D 2.30E-04* 3.5E-04***
(market capitalization) (0.00) (0.00)
- Herfindahl Index * mean R&D 1.10E-04* 1.41E-04**
(market capitalization) (0.00) (0.00)
- Gini coefficient * mean R&D -2.89E-05 -9.62E-05
(market capitalization) (0.00) (0.00)
Control variables (not all reported)*
- Herfindahl Index (sales) -0.019 -0.013 0.004 -0.036* -0.016 0.004 -0.019 -0.013 -0.008 -0.021 -0.008 0.009 -0.037* -0.014 0.008 -0.012 -0.019 -0.008
(0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.01) (0.01) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.01)
- mean R&D spending -1.24E-04** -1.2E-04* -1.2E-04* -1.2E-04* -1.2E-04* -1.1E-04* -3.3E-04*** -2.1E-04** -1.2E-04* -1.3E-04* -1.1E-04* -1.2E-04* -1.2E-04* -1.2E-04* -1.1E-04* -3.7E-04***-4.3E-04***-2.4E-04***
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
N 475 484 475 473 484 473 475 484 484 475 484 475 473 484 473 475 484 475
F-value 4.69*** 4.44*** 4.93*** 5.12*** 4.49*** 4.93*** 5.44*** 5.04*** 4.6*** 4.51*** 4.21*** 4.49*** 5.17*** 4.34*** 4.46*** 6.17*** 5.13*** 4.25***
Table 4. Hypothesis testing: weak test - lagged relationships
+ (p<0.1), *(p<0.05), **(p<0.01), ***(p<0.001) * Results of all control variables available upon request.
Lag 1/Lag 2 Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef.
Independent variables
H1 - Inequality in value
- Kingpin's share (market capitalization) 0.079*** 0.079**
(0.02) (0.02)
- Herfindahl Index (market capitalization) 0.032* 0.032*
(0.01) (0.02)
- Gini coefficient (market capitalization) 0.015 -0.006
(0.01) (0.02)
H2 - Inequality in tech. investment
- Kingpin's share 0.059** 0.056*
(0.02) (0.02)
- Herfindahl Index 0.009 0.009
(0.02) (0.02)
- Gini coefficient 0.008 -0.008
(0.02) (0.02)
H3 - Interaction effect
- Kingpin's share * mean R&D 6.4E-04*** 6.0E-04***
(market capitalization) (0.00) (0.00)
- Herfindahl Index * mean R&D 2.8E-04*** 2.5E-04***
(market capitalization) (0.00) (0.00)
- Gini coefficient * mean R&D -3.5E-04*** -3.8E-04***
(market capitalization) (0.00) (0.00)
Control variables (not all reported)*
- Herfindahl Index (sales) -0.027 -0.003 0.023 -0.013 0.011 0.019 -0.010 0.006 0.011 -0.018 0.006 0.024 -0.016 0.020 0.023 0.001 0.016 0.021
(0.02) (0.02) (0.02) (0.02) (0.02) (0.01) (0.01) (0.01) (0.01) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02)
- mean R&D spending 2E-04*** 2E-04*** 2E-04*** 2E-04*** 2E-04*** 2E-04*** -4E-04*** -4E-05 3.E-04*** 2E-04*** 2E-04** 2E-04** 2E-04** 2E-04** 2E-04** -3.4E-04** -3.8E-05 3.3E-04***
(0.00) (0.00) (0.00) (0.00) (0.00) (0.01) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
N 604 615 604 600 615 598 604 615 604 603 614 603 599 614 597 603 614 614
F-value 10.82*** 9.44*** 8.85*** 9.61*** 8.79*** 8.51*** 15.74*** 12.83*** 13.74*** 6.45*** 5.41*** 4.91*** 5.51*** 4.94*** 4.79*** 9.43*** 7.32*** 9.32***
40
Table 4. Hypothesis testing: weak test - lagged relationships (con’td)
+ (p<0.1), *(p<0.05), **(p<0.01), ***(p<0.001) * Results of all control variables available upon request.
Lag 3 Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef.
Independent variables
H1 - Inequality in value
- Kingpin's share (market capitalization) 0.082**
(0.03)
- Herfindahl Index (market capitalization) 0.034*
(0.02)
- Gini coefficient (market capitalization) -0.025
(0.02)
H2 - Inequality in tech. investment
- Kingpin's share 0.062*
(0.03)
- Herfindahl Index 0.014
(0.02)
- Gini coefficient -0.023
(0.02)
H3 - Interaction effect
- Kingpin's share * mean R&D 6.0E-04***
(market capitalization) (0.00)
- Herfindahl Index * mean R&D 2.4E-04***
(market capitalization) (0.00)
- Gini coefficient * mean R&D -4.2E-04***
(market capitalization) (0.00)
Control variables (not all reported)*
- Herfindahl Index (sales) -0.012 0.013 0.024 0.004 0.026 0.026 0.009 0.025 0.029+
(0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02)
- mean R&D spending 2.0E-04**1.7E-04**1.7E-04**1.8E-04**1.7E-04**1.7E-04** -3.4E-04** -3.0E-05 3.5E-04***
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
N 603 613 602 598 613 596 602 613 602
F-value 9.32*** 4.95*** 4.64*** 5.10*** 4.53*** 4.55*** 8.24*** 6.34*** 9.06***
Table 5. Hypothesis testing: strong test – lagged relationships
+ (p<0.1), *(p<0.05), **(p<0.01), ***(p<0.001) * Results of all control variables available upon request.
Kingpin-sale/market capitalization (lag 1) Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef.
Independent variables
H1 - Inequality in value
- Top firm's share (market capitalization) 0.026 0.040+
(0.02) (0.02)
- Herfindahl Index (market capitalization) 0.015 0.016
(0.02) (0.02)
- Gini coefficient (market capitalization) 0.027 0.030
(0.02) (0.02)
H2 - Inequality in technological investment
- Top firm's share 0.057* 0.063**
(0.02) (0.02)
- Herfindahl Index 0.013 0.015
(0.02) (0.02)
- Gini coefficient 0.034* 0.028+
(0.02) (0.02)
H3 - Interaction effect
- Kingpin's share * mean R&D spending 7.37E-05 2.8E-04**
(market capitalization) (0.00) (0.00)
- Herfindahl Index * mean R&D spending 6.65E-05 1.4E-04*
(market capitalization) (0.00) (0.00)
- Gini coefficient * mean R&D spending -5.40E-05 1.40E-05
(market capitalization) (0.00) (0.00)
Control variables (not all reported)*
- Herfindahl Index (sales) -0.019 -0.015 0.006 -0.034+ -0.013 0.007 -0.008 -0.009 -0.005 -0.023 -0.009 0.012 -0.031 -0.085 0.011 -0.012 -0.006 0.001
(0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.01) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02)
- mean R&D spending -1E-04* -1E-04* -1E-04* -1E-04* -1E-04* -1E-04* -2E-04+ -2E-04* -9.E-05 -2E-04** -1E-04** -2E-04** -1E-04** -1E-04** -1E-04** -4E-04*** -2E-04** -2E-04*
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
N 455 461 455 450 461 450 455 461 455 455 461 455 450 461 450 455 461 455
F-value 4.24*** 4.07*** 4.36*** 4.68*** 4.04*** 4.47*** 4.14*** 4.09*** 4.16*** 4.55*** 4.23*** 4.48*** 4.97*** 4.22*** 4.42*** 5.09*** 4.64*** 5.13***
42
Table 5. Hypothesis testing: strong test – lagged relationships (cont’d)
+ (p<0.1), *(p<0.05), **(p<0.01), ***(p<0.001) * Results of all control variables available upon request.
Kingpin-sale/market capitalization (lag 2) Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef.
Independent variables
H1 - Inequality in value
- Top firm's share (market capitalization) 0.033 0.034
(0.02) (0.02)
- Herfindahl Index (market capitalization) 0.019 0.017
(0.02) (0.02)
- Gini coefficient (market capitalization) 0.032 0.030
(0.02) (0.02)
H2 - Inequality in technological investment
- Top firm's share 0.050* 0.054*
(0.02) (0.02)
- Herfindahl Index 0.011 0.011
(0.02) (0.02)
- Gini coefficient 0.040* 0.033+
(0.02) (0.02)
H3 - Interaction effect
- Kingpin's share * mean R&D spending 1.66E-04 1.77E-04
(market capitalization) (0.00) (0.00)
- Herfindahl Index * mean R&D spending 1.13E-04 1.13E-04
(market capitalization) (0.00) (0.00)
- Gini coefficient * mean R&D spending 1.27E-06 -7.40E-05
(market capitalization) (0.00) (0.00)
Control variables (not all reported)*
- Herfindahl Index (sales) -0.018 -0.012 0.013 -0.026 -0.007 0.014 -0.006 -0.005 0.001 -0.015 -0.006 0.016 -0.024 -0.002 0.016 -0.003 -2.E-04 0.004
(0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02)
- mean R&D spending -7.E-05 -7.E-05 -6.E-05 -7.E-06 -7.E-06 -6.E-05 -2E-04* -2E-04* -8.E-05 -7.E-05 -7.E-05 -7.E-05 -8.E-05 -8.E-05 -7.E-05 -2E-04* -2E-04* -4.E-05
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
N 430 436 430 425 436 425 430 436 430 430 436 430 425 436 425 430 436 430
F-value 2.65** 2.56** 2.73** 2.89** 2.45* 3.06* 2.70** 2.71** 2.39* 3.18** 3.08** 3.19** 3.51*** 3.01** 3.34** 3.24** 3.25** 3.04**
43
Table 5. Hypothesis testing: strong test - lagged relationships (cont’d)
+ (p<0.1), *(p<0.05), **(p<0.01), ***(p<0.001)
* Results of all control variables available upon request.
Kingpin-sale/market capitalization (lag 3) Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef. Coef.
Independent variables
H1 - Inequality in value
- Top firm's share (market capitalization) 0.029 0.029
(0.02) (0.02)
- Herfindahl Index (market capitalization) 0.015 0.013
(0.02) (0.02)
- Gini coefficient (market capitalization) 0.018 0.021
(0.02) (0.02)
H2 - Inequality in technological investment
- Top firm's share 5.8E-02* 0.062*
(0.02) (0.02)
- Herfindahl Index 1.41E-02 0.009
(0.02) (0.01)
- Gini coefficient 4.4E-02* 0.039*
(0.02) (0.02)
H3 - Interaction effect
- Kingpin's share * mean R&D spending 7.42E-05 1.45E-04
(market capitalization) (0.00) (0.00)
- Herfindahl Index * mean R&D spending 6.22E-05 8.31E-05
(market capitalization) (0.00) (0.00)
- Gini coefficient * mean R&D spending 1.79E-05 -1.45E-05
(market capitalization) (0.00) (0.00)
Control variables (not all reported)*
- Herfindahl Index (sales) -0.025 -0.018 -0.002 -0.039+ -0.018 0.006 -0.011 -0.011 -0.009 -0.023 -0.014 0.001 0.038+ -0.016 0.007 -0.012 -0.009 -0.006
(0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02)
- mean R&D spending -6.E-05 -5.E-05 -6.E-05 -5.E-05 -5.E-05 -4.E-05 -1.E-04 -1.E-04 -7.E-05 -7.E-05 -7.E-05 -7.E-05 -6.E-05 -7.E-05 -6.E-05 -2E-04+ -1.E-04 -7.E-05
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
N 405 411 405 400 411 400 405 411 405 405 411 405 400 411 400 405 411 405
F-value 2.86** 1.74** 2.75** 3.27** 2.72** 3.40** 2.72** 2.72** 2.96** 4.04*** 3.96*** 3.97*** 4.60*** 3.98*** 4.40*** 4.07*** 4.04*** 3.84***
Table 6. Hypothesis testing: feedback loop
In testing Hypotheses 4a and 4b, we have the same independent variable for every model with different
dependent variables. As such, we indicate the dependent variables on the tables, but the coefficients and the
standard errors reported on the right hand side refer to the independent variable. We only report the coefficients
and the standard errors of the independent variable, i.e. the share of value of the segment, although control
variables for other hypotheses were included in the models.
Lag1 Coef. S. E. F-value N
Dependent variables
H4a - Inequality in value
- Kingpin's share (market capitalization) 0.320 ** 0.12 28.03 *** 572
- Herfindahl Index (market capitalization) 0.177 0.16 17.12 *** 584
- Gini coefficient (market capitalization)
H4b - Inequality in technological investment
- Kingpin's share 0.311 ** 0.12 24.91 *** 578
- Herfindahl Index 0.045 0.17 12.8 *** 584
- Gini coefficient
Control variables (included, but not reported)
Lag2 Coef. S. E. F-value N
Dependent variables
H4a - Inequality in value
- Kingpin's share (market capitalization) 0.208 0.126 16.02 *** 542
- Herfindahl Index (market capitalization) 0.084 0.163 10.73 *** 551
- Gini coefficient (market capitalization)
H4b - Inequality in technological investment
- Kingpin's share 0.329 * 0.133 15.94 *** 561
- Herfindahl Index 0.001 0.19 6.73 *** 551
- Gini coefficient
Control variables (included, but not reported)
Lag3 Coef. S. E. F-value N
Dependent variables
H4a - Inequality in value
- Kingpin's share (market capitalization) 0.150 0.137 8.83 *** 512
- Herfindahl Index (market capitalization) 0.039 0.167 7.97 *** 518
- Gini coefficient (market capitalization)
H4b - Inequality in technological investment
- Kingpin's share 0.358 * 0.146 10.59 *** 549
- Herfindahl Index -0.023 0.202 4.65 *** 518
- Gini coefficient
Control variables (included, but not reported)
no model fit
no model fit
no model fit
no model fit
no model fit
no model fit
45
Table 7. Summary of results
* The names of the variables listed are dependent variables for these hypotheses
Contemporary Year t+1 Year t+2 Year t+3
H1
positive sign (***) positive sign (***) positive sign (**) positive sign (**)
positive sign (*) positive sign (*) positive sign (*) positive sign (+)
positive sign positive sign negative sign negative sign
H2
positive sign (**) positive sign (**) positive sign (*) positive sign (*)
positive sign positive sign positive sign positive sign
positive sign positive sign negative sign negative sign
H3
positive sign (**) positive sign (***) positive sign (***) positive sign (***)
positive sign positive sign (***) positive sign (***) positive sign (***)
negative sign negative sign (***) negative sign (***) negative sign (***)
H4a*
positive sign (**) positive sign positive sign
positive sign positive sign positive sign
positive sign positive sign negative sign
H4b*
positive sign (**) positive sign (*) positive sign (*)
positive sign positive sign negative sign
no model fit no model fit no model fit
Top firm's share
Herfindahl Index
Gini coefficient
Joint effect of inequality in value and
mean of technological investment
Computer-NAICS5
Herfindahl Index (market capitalization)
Gini coefficient (market capitalization)
Inequality in value
Inequality in technological investment
Top firm's share (market capitalization)
Gini coefficient (market capitalization)
Gini coefficient
Feedback - inequality in technological
investment
Top firm's share
Herfindahl Index
Top firm's share (market capitalization)
Top firm's share (market capitalization)
Herfindahl Index (market capitalization)
Herfindahl Index (market capitalization)
Gini coefficient (market capitalization)
Feedback - inequality in value