Enemies at the Gate? New Entry Threats and Innovation

in the U.S. IT Industry

Yang Pan, Peng Huang, Anandasivam Gopal

Robert H. Smith School of Business

University of Maryland, College Park, MD 20742

Abstract

We examine how firms in the IT industry adjust their innovation strategy as a response to the potential

entry threats they face in their product markets. While turbulence in the product markets caused by

startups affects an incumbent firm’s investment decisions, prior research has provided contrasting

theoretical predictions on the relationship between new entry threats and innovation strategy. In addition,

the absence of acceptable industry classifications for startups, as well as the inability to accurately gauge

when they represent a credible threat has limited empirical research into this question. In this paper, we

first contribute a new measure of identifying these threats through text analyses based on product

descriptions provided by incumbent firm 10-K filings and business descriptions provided by start-ups.

This new measure differs significantly from approaches that use static industry classifications, which are

backward-looking and do not fully account for industry evolution over time. We show the text-based

measure captures entry threats from the startup space through a series of validation tests. Second, using a

sample of U.S. IT firms over the period 1997-2013, we show that incumbent firms react to new entry

threats by systematically reducing innovation investments. We show these effects to be robust to a series

of regression specifications addressing the endogeneity of new entry threats. Interestingly, we also find

that in the face of intensive new entry threats, firms with diversified product portfolios are less aggressive

in terms of reducing their innovation spending, compared to specialized firms. We discuss the research

and practice implications of the new text-based measure for new entry threats as well as the responses to

these threats by incumbents in the paper.

Keywords: Innovation, R&D, new entry threats, disruptive technology, text mining

1

1. Introduction

It is accepted wisdom that the Information Technology (IT) industry plays a central and critical role in

driving economic growth over the last two decades, especially in the United States (Jorgenson and Stiroh

1999; Jorgenson et al. 2000). This contribution to economic growth is partly a product of the constant

technological changes and fast clockspeed (Fine 1998) that is a significant characteristic of the IT

industry in general. The rate of change in the industry’s external environment, including the development

of new technologies, shifts in consumer preferences, and fast-moving market dynamics far exceeds those

seen in other industries (Brynjolfsson and McAfee 2011; Mendelson and Pillai 1999). The result is a

compression of product life-cycles (Mendelson and Pillai 1998), volatility in market structure and a

hyper-competitive context where advantages, if any, tend to be short-lived (Wiggins and Ruefli 2005). In

such contexts, the ability to develop new product and process innovations remains a key driver of a firm’s

competitive advantages (Kleis et al. 2012).

A significant part of this fast-moving dynamic is fed by the high rate of new entry in the form of

entrepreneurial ventures; technology startups are constantly introducing new products and business

models to the market (Giarratana 2004). Indeed, over 70% of all venture-capital funded startups tend to be

associated with the IT industry (Gompers and Lerner 2001), thus contributing significantly to the threat of

new entry experienced by incumbent firms. The presence of intense entrepreneurial activity in the product

market of an incumbent can cause turbulence in the market – entrepreneurs incorporating new and

innovative technologies, backed by influential venture capitalists and their deep networks, can affect the

future market potential of incumbents very quickly in the IT context. Incumbents may respond to this

constant threat of new entry in many different ways, but since innovation remains an important source of

competitive advantage in technology markets, we ask: In the presence of such threats of new entry, how

do incumbents respond in terms of their investments in innovation? This forms the central research

question we address.

Extant literature suggests two mechanisms at play in how incumbents may respond to new entry

threats. On the one hand, firms may increase their investments in their innovative activities, in the hope of

2

building technological and process capabilities to successfully compete in the future as the perceived

nascent threats materialize (Lukach et al. 2007; Reinganum 1983). Consider, for example, the situation

facing Intel upon the anticipated entry of ARM processors into the mobile computing markets. Intel

initially underestimated the mobile processor markets and the strengths of ARM processors - RISC

(Reduced Instruction Set Computing) based processors that reduce costs, heat and power use, making

them perfect for light, portable, battery-powered devices, leading them to be licensed by firms such as

Apple, Marvell, Broadcom, and Samsung. However, rather than eschew the threats to their market that

ARM represented, Intel chose to invest heavily in innovation, developing its own low-power x86-

compatible Atom processors.1 Thus a strategic response to the threat of new entry on the part of the

incumbent may be to increase investments in innovation-related activities, so as to compete head on with

the new entry, if and when it does materialize.

On the other hand, an arguably logical choice would be to eschew risky investments in innovation

that, at the best of times, have uncertain payoff and instead choose to either conserve their cash (Hoberg

et al. 2014; Klepper and Simons 1997) or invest in complementary capabilities that are vital in the

commercialization of new innovation such as in manufacturing, distribution, or marketing (Teece 1986),

as a way to differentiate themselves should the anticipated threats materialize (O’Connor and Rafferty

2012). In addition, an incumbent may choose to delay its investment in the new innovation until the

technological and market uncertainties settle, and then cherry-pick the winners that emerge from the new

entries – either through strategic alliances or acquisitions. Consider, for instance, Instagram, a mobile

photo-sharing app developer. As Instagram steadily increased in popularity within the social media

ecosystem and positioned itself as a social network in its own right, it was perceived as a potential threat

by Facebook. Rather than wait for the competition to fully materialize, Facebook acquired the company

for approximately $1 Billion in cash, while also abandoning its own efforts towards developing a

standalone mobile photo-sharing app.2 Effectively, Facebook chose to not invest further in its own

1 http://appleinsider.com/articles/15/01/19/how-intel-lost-the-mobile-chip-business-to-apples-ax-arm-application-processors

2 http://techcrunch.com/2012/04/09/facebook-to-acquire-instagram-for-1-billion/

3

innovation activities, but to conserve cash and deploy this in the form of an acquisition to deal with the

threat of new entry. These alternative predictions reflect an important tension in the literature regarding

the specific manner in which firms respond to new entry threats that remains empirically unresolved. In

this paper, we present empirical evidence that helps resolve this tension specifically in the IT industry,

thereby generalizing beyond individual cases such as those described above.

Studying new entry threats empirically poses several challenges. By definition, threats from

entrepreneurial startups represent forward-looking estimations of the extent to which the potential entry of

new competition may influence cash flows or product market performance; therefore, these threats have

not manifested as yet but may materialize at some point in the future, compared to traditional measures of

observed entry, which are contemporaneous. Thus, extant empirical work studying new entry threat have

typically focused on de alio entry (diversifying entry) instead of de novo entry (entry by startups), and

extrapolated from these measures of present market competition to evaluate potential new entry threats

(Aghion et al. 2009; Becker-Blease 2011). In addition, these measures typically define competition or

industry concentration based on static SIC or NAICS codes (Aghion and Howitt 1992; Becker-Blease

2011), and suffer from multiple shortcomings: a firm is rarely reclassified even when it diversifies or

transitions into a different industry; and they lack temporal variations both within and across industries.

Furthermore, for start-ups, there are no ready-to-use and accepted industry classifications, causing

difficulties in identifying any potential threats from de novo entries. Since new entry threat is often a

central construct in theoretical analyses of competitive dynamics (Spence 1977), the absence of an

established measure represents a significant gap in the literature.

Given this lack of an existing measure of the threat of new entry from start-ups, one of the

objectives of our work in this paper is to develop and validate such a measure. While we formally define

the new entry threat measurement later in the paper, intuitively we measure the extent to which an

incumbent firm’s description of its product markets in its 10-K filings overlaps with the descriptions of

new entrepreneurial firms that receive first-stage funding from VCs during the same period. In order to

understand where de novo entry might occur for an incumbent, we focus on extracting business

4

descriptions of startups receiving first round VC funding from VentureXpert in that year. The focus on

very early-stage startups is particularly appropriate here since these firms collectively represent

movement in the incumbent firm’s product market towards possible future competition. In addition, the

fact that they were funded by VCs makes these threats particularly relevant because they tend to be highly

innovative and have a higher likelihood of becoming real threats to incumbents in the near future.

Since we are interested in evaluating the effect of these threats on the incumbent firm’s

innovation strategy, we build on prior research in IS and innovation (Kleis et al. 2012; Tambe et al. 2012)

by focusing on R&D spending by the incumbent. R&D expenditures are empirically attractive because

R&D efforts can be undertaken relatively quickly by the firm and can be more easily linked to specific

external events. In our study, more specifically, we use R&D intensity, defined as R&D expenditure

standardized by total asset in the same year, as a measure of the extent to which the firm responds to new

entry threats through investments in innovation. We estimate our models using panel data methods on

firms in the IT industries over the period of 1997 - 2013; these years are selected based on joint data

availability on R&D spending, new entry threats and firm-level data on industry classifications. Our

results show that greater new entry threats are associated with a reduction in R&D investments, all else

being equal. While R&D spending represents one form of innovation spending, we also test our theories

on an alternative measure of innovation that has been used extensively – patent counts and citations (Hall

et al. 2001). While patent data pose several challenges in terms of data quality and availability, our results

from patent counts are fully consistent. Finally, we also examine the extent to which these relationships

hold in the case of diversified incumbent firms – our results show that new entry threat has a larger effect

on specialized firms, suggesting that incumbents operating in specialized markets are more sensitive to

new entry threat and adjust their investments in innovation more swiftly.

Our work provides several contributions to the IS literature on innovation. The rapid rate of

technological change observed in the IT industries and the associated volatility in product markets

(Brynjolfsson and McAfee 2011) has brought into sharp focus the role of the firm’s response to such

threats. In addition, a societal focus on innovation-related IT entrepreneurship and the development of

5

entrepreneurial ecosystems (in the forms of incubators and accelerators) have radically improved the

abilities of these new ventures to threaten incumbent firms. In these contexts, it is important to understand

how new entry threat may be accurately measured, and to understand how incumbents respond to these

threats. Our work here directly contributes to these gaps in the literature. The implications of our work

extent beyond understanding the impact of new entry threats to innovative investments, to also evaluating

how these responses may vary depending on whether the firm is diversified, and the effects on patenting

strategies adopted by incumbents. We thus contribute to the broader literature on IT and innovation in a

significant manner.

Beyond the theoretical implications, we also contribute by devising and validating a new way of

measuring new entry threat from entrepreneurial ventures using text analysis. We build on existing work

in other disciplines such as finance and marketing that has used text to construct measurement schemes

(Hoberg and Phillips 2010; Tetlock et al. 2008) by creating text similarity scores between incumbent

firms and broad movements in the VC-backed entrepreneurial space. These similarity scores are more

suitable for capturing emerging threats from the collective entrepreneurial space, rather than focusing on

single entrepreneurs that are easy to ignore or discount, thereby representing considerable value within

the fast-moving IT context. We believe that many more empirical questions in the literature relating to

new entry threats can be addressed through the usage of the measures we provide.

2. Theoretical Background

2.1 New Entry Threats

All firms face lateral competition and new entry threats in their native markets. However, these are

particularly pronounced and influential in the high-tech sector, specifically in the IT industry, (Fontana

and Nesta 2009), where firm survival is often in question as a result of these competitive dynamics.

Collectively, these are referred to as product market threats, defined as incipient instability and

uncertainty in a firm’s product market, which can threaten the sustainability of the firm’s future earnings

as well as viability of its product portfolio (Hoberg et al. 2014).

6

Within the domain of product market threats, we distinguish new entry threats specifically from

competition from other existing competitors on two key dimensions. First, competition is

contemporaneous and is usually measured by market outcomes, such as market share/concentration

(Blundell et al. 1999) or Lerner Index (Aghion 2003), while new entry threats are forward-looking since

they imperil the “stability and sustainability of future earnings” (Brav et al. 2005). Second, while

competition describes equilibrium market outcomes, new entry threat is inherently about disequilibrium,

often associated with changes in competitive dynamics (Cockburn and MacGarvie 2011). The ability to

threaten future earnings, especially in the IT sector, is closely linked to technological changes and product

innovation (Coad and Rao 2008), thereby making prior work in industry lifecycles of particular relevance

(Abernathy and Utterback 1978). Industry lifecycles describe the processes that unfold in technologically

progressive industries as they evolve from birth to maturity (Klepper 1996). In general, when new

industries form, there is considerable new entry into the field, firms offer many varieties of similar

products, prices are typically high and there is considerable product innovation that occurs within the

nascent industry (Jovanovic and MacDonald 1994; Klepper 1996). Over time, despite steady market

growth, new entry into the field ceases, prices drop and there is a shakeout in the number of firms. At this

stage, a dominant design emerges in the industry (Jovanovic and MacDonald 1994) and incumbent firms

shift to process innovations that improve upon the dominant design, leading again to increased

competition in the industry. Thus, industry lifecycles are characterized by successive waves of product

and process innovations that introduce high volatility within the technology industry.

These dynamics are particularly salient in the tech sector for three reasons. First, prior work in

technology-based industries has indicated that in fast-moving contexts, there is likely little separation

between periods of product and process innovation (Klepper and Simons 1993). Effectively, product and

process innovations emerge concurrently from different players in the industry, thereby generating

considerable product market threats from new venture firms (Abernathy and Utterback 1978). Second, the

IT industry incorporates significant levels of heterogeneity in terms of firm capabilities, learning and the

ability to capitalize on innovation activities (Agarwal and Gort 2002; Reinganum 1983). Therefore, given

7

this heterogeneity, it is more likely that incumbents are bombarded with concurrent and successive waves

of uncertainty from ongoing process and product innovation at any point in time, as has been documented

clearly in the IT industry (Rai and Tang 2013). Finally, evidence indicates that the pace at which industry

lifecycles emerge and stabilize, even briefly, in the last two decades has increased in the IT industry

(Carrillo 2005; Fine 1998; Mendelson and Pillai 1998). As “clockspeed” quickens, there is thus

systematically greater threat of new entry for incumbents.

How may incumbents respond to an increased perception of new entry threats into their product

markets? A review of the empirical literature studying firm responses to new entry within their markets

provides some clear indications. One clear strategic response that incumbents adopt to fight new entry is

through pricing (Geroski 1995). Pricing allows incumbents to deter entry or retaliate after entry has

occurred. Theoretical analyses of new entry suggest several approaches to strategic pricing that

incumbents may adopt. One such strategy is limit pricing, whereby the incumbent sets prices low enough

so as to make new entry into the product market unprofitable (Bain 1956; Goolsbee and Syverson 2008).

Alternatively, predatory pricing may also be adopted, whereby the incumbent cuts prices after entry in

order to drive out the new entrants and develop a tough reputation that helps deter further entry (Milgrom

and Roberts 1982). Recent empirical work examining incumbents’ pricing responses have, however,

provided mixed results associated with these strategies (Simon 2005), suggesting that the successful use

of these pricing strategies depend on the extent to which the incumbent is able to respond aggressively

(Yamawaki 2002) and its incentives to respond (Simon 2005).

In differentiated product markets, firms are likely to use other responses to new entry since

demand tends to be inelastic to price. These include incumbent investments in advertising, increased

diversification, expanding product lines, retrenchment in order to conserve cash, or ceding the market to

the new entrant. Each of these options appears attractive depending on the specific context and the

technological sophistication of the new entrant. Thomas (1999) shows that in the ready-to-eat breakfast

cereal industry, incumbents tend to respond to the threat of new entry through increasing advertising, as a

way to limit or deter further new entry. Large sunk investments in advertising raise the fixed costs of

8

operating in an industry for new entrant, and allow firms to credibly expand output (Sutton 1991).

Similarly, Koski and Majumdar (2002) find that in the telecommunications industry, incumbent local

exchange carriers significantly raise advertising expenditures when faced with the threat of entry.

Alternatively, expanding into new areas or introducing a battery of new products that crowd the product

attribute space leaves little profitable room for new entrants to enter (Schmalensee 1978; Shankar 1999).

However, as discussed by Judd (1985), such a strategy may not always be viable for the incumbent; in the

case of multiproduct incumbents, it may be preferable to withdraw from specific product markets in order

to avoid intense post-entry competition. Finally, Hoberg et al. (2014) show that in the presence of product

market threats from other incumbents, firms are more likely to conserve cash and make fewer payouts in

the forms of dividends or repurchases. They argue that this indicates firms respond to impending entry if

and when it occurs by conserving critical resources.

In lieu of an exhaustive review of the literature on how incumbents respond to new entry, we

have condensed a representative set of empirical studies examining this issue in Table 1, listing out the

empirical context, the mode of entry being discussed, and the strategic responses taken by incumbents.

There are two central insights that emerge from this table that are particularly relevant to the IT industry,

and to our work in this paper. First, very few prior studies have investigated the threat of de novo entry –

entry by startup firms. The vast majority of the literature has focused on de alio entry – lateral,

diversifying entry by other existing competitors. In some cases, there are no distinctions drawn between

these two forms of new entry (Carroll et al. 1996; Khessina and Carroll 2008). In the IT industry de novo

entry is widespread, common, and are often the source of disruptive technologies that present significant

threats to existing firms. Indeed, prior work establishes that a majority of product innovations (i.e. new

designs and technologies, rather than process innovations) emerge from the entrepreneurial space

(Agarwal and Gort 2002; Prusa and Schmitz Jr 1991; Prusa and Schmitz Jr 1994), making the study of de

novo entry particularly apposite for the IT industry. Second, with the exception of analytical models

(Lukach et al. 2007) and the recent work by Aghion et al. (Aghion et al. 2009), most prior research has

examined incumbent responses in terms of pricing/capacity, product market expansion/diversification,

9

and advertising, while largely ignoring the extent to which the incumbents can respond through

innovation. If innovation remains a primary driver of competitive advantage in the IT industry, it is

noteworthy that the literature is largely silent on the role of innovation as a strategic variable in response

to new entry within the IT industry. In this study, we aim to bridge this specific gap in the literature. In

the next section, we discuss the potential theoretical mechanisms that may link the incumbent’s response

to new entry threats through innovation, before delving into empirical operationalizations.

[Insert Table 1 about here]

2.2 New Entry Threat and the Incumbent’s Investment in Innovation

How may an incumbent firm respond to new entry threats in its product markets through innovation

spending? Prior research offers four specific theoretical perspectives that inform the relationship between

new entry threat and the incumbents’ response through innovation investments. The first mechanism is

based on a strategy of pre-emption, specifically along the dimension of innovation itself. In contexts

where demand is increasing and innovation is critical to retain advantage, the incumbent may pre-empt

potential new entry by building capacity (i.e. innovative products or solutions) well before it is needed,

essentially pre-empting innovation investments that any entrant may choose to make (Eaton and Lipsey

1979). Pre-emptive market expansion into new product lines or market segments by the incumbent creates

out a crowding out effect of the market spectrum, thereby eliminating viable niches for entrants while also

signaling that the market available to future new entry is likely to be less profitable (Schmalensee 1978).

To the extent that new product introduction requires heavy investments in product innovation in the IT

industry, this mechanism would postulate a positive relationship between the perception of new entry

threats for the incumbent and its investments in innovation, either through increased R&D spending or

higher patenting rates.

Providing a counter-argument to those presented above, a series of papers suggest that major

innovations may not be most efficiently carried out by large incumbent corporations due to a series of

communication and incentive problems (Chandy and Tellis 2000; Hamberg 1963; Mueller and Tilton

10

1969). These scholars argue, for instance, that incumbents generally prefer R&D projects that promise

short-term payoffs rather than long-term, risky projects that may be potentially groundbreaking.

Incumbents also have a vested interest in existing technologies, and are reluctant in pursuing major

innovations that would replace the currently profitable products. Finally, large firms have difficulty

building organizational norms that reward and nurture highly creative engineers and innovators, thereby

facing serious issues of incentive alignment. Thus, it is not surprising that Henderson (Henderson 1993)

finds that the research efforts of incumbents to exploit radical innovation are significantly less productive

than those of new entrants. Thus, a perspective rooted in the presence of considerable heterogeneity in the

abilities of incumbents to respond in a nimble manner to new entry threat would predict a negative

relationship between new entry threat and innovative spending. Incumbents facing potential new entrants

are likely to seek alternative means to compete, leading to underinvestment in innovation.

A third perspective argues that strategic considerations may deter incumbents from investing

heavily in new innovation when they face high rate of entry, especially in the early stages when the

potential of new innovation is uncertain. Teece (Teece 1986) argues that when new innovations are easily

imitated, the appropriation of the innovation depends heavily on the ownership of certain complementary

assets, and the profits may not accrue to the developers of the intellectual property. Furthermore, the

active development of products based on new innovative output typically faces high levels of technical

and demand uncertainty. Since incumbents, faced with the risk of losing the outcomes of new investments

as well as reputational loss from prior investments, tend to be risk-averse, they are likely to delay entry

into new market niches based on new innovation (Mitchell 1989). Building on these ideas, Mitchell and

Singh (Mitchell and Singh 1992) find that when incumbents enter a new product area, they often take a

more cautious approach that limits their investment exposure, choosing instead to expand their knowledge

base through strategic alliances. This cautious approach suggests that incumbents are not likely to

significantly increase their innovation spending when faced with new entry. Indeed, they might even

reduce their innovation spending, choosing instead to invest in complementary assets or pre-entry

alliances (Mitchell 1989).

11

Finally, Hauser and Shugan’s (1983) “defender” model provides a useful analytical framework

that helps shed light on an incumbent’s response to competitive entry. The authors show that under a wide

range of conditions, the optimal, profit-maximizing response to new entry may indeed be to respond

negatively through, for instance, cutting back resource commitments or raising prices (Gatignon et al.

1989; Kuester et al. 1999). The underlying argument in this model is that entry generally decreases every

other competitors’ profits and makes the overall market less attractive. In addition, the enhanced

competitive pressure brought about by new entrants in the marketplace has important implications for

managerial incentives (Schmidt 1997). Specifically, if new entry forces the incumbent to reduce slack,

enhance efficiency, and focus on margins, it also helps to discipline managers by aligning their incentives

with the firm, rather than pursuing their own goals (Hart 1983). Thus, reductions in organizational slack

likely forces managers to focus on firm profitability and thereby invest less in risky activities such as

innovation that may not have immediate payoff, leading to a negative relationship between new entry

threats and innovation spending by incumbent firms.

[Insert Table 2 about here]

We present a summary of these arguments, and the theoretical perspectives within, in Table 2. As

evident, there are contrasting predictions for how incumbents in the IT industry may respond to new entry

threats in their product markets. Given these countervailing predictions, we believe it is more useful to let

the empirical analysis reveal the patterns that apply in the IT industry. As a first step, we first propose a

new text-based measure of new entry threats arising from de novo entry in the next section, and describe

the validation process for the measure.

3. A Text-Base Measure of New Entry Threats

An empirical measurement of new entry threat has remained elusive in the innovation literature, given the

forward-looking nature of this construct. Most existing work has used current competitive intensity or

observed (actual) entry to account for these effects (Aghion et al. 2009; Chen et al. 1992) but these are

approximations at best and do not address the construct as discussed in the managerial and theoretical

12

literature (Caves and Porter 1977; Porter 2008). When facing de novo entry, the challenge for the

incumbent is to spot the disruptive entrepreneurial firms that constitute such a threat early on and respond

adequately (Rigby et al. 2002). While this ability to spot the specific disruptor is imprecise and uncertain

(Markides 2006), broad movements within the entrepreneurial space into markets that intersect with the

incumbent’s product markets can still be recognized as significant threats. In other words, while an

incumbent may not observe a specific startup with a particular product innovation, the incumbent will

surely observe and respond to shifts across the larger entrepreneurial landscape into certain specific

markets or technologies. Prior work studying technology fads and cascades has discussed these broader

trends as being significant predictors of firm and individual behavior (Abrahamson 1991; Bikhchandani et

al. 1998). We argue that such large-scale new venture formation within a certain product market is a valid

representative of new entry threat and impending competition in the future. New analytical techniques

using text analysis provide methods that may be used to measure such threats in a novel manner. We

explore these options to construct measures for new entry threats emerging from startups, described in the

remainder of this section.

3.1 Methodology: From Words to New Entry Threats

The use of text analysis requires reasonably descriptive text corpora from firms that may be used to

construct appropriate measures. A considerable body of work that has used public filings provided by the

firms, specifically firm annual reports (10-Ks) in the US, to create measures of firm fundamentals such as

competitive intensity, industry classes and firm strategy (Hoberg and Phillips 2010; Tetlock 2011;

Tetlock et al. 2008). These documents are useful as sources of data for two reasons. First, public firm

product descriptions must be representative and significant as required by financial market regulations.

Thus, product descriptions of public firms contain timely information about their products, markets and

competitors that are consistent with the firm’s perceptions. Second, as firms evolve, these descriptions are

modified and updated to reflect the changing nature of their businesses, thereby providing longitudinal

variation. We therefore consider the use of text-based measurement schemes for capturing new entry

threats in the context of IT industry.

13

We construct our new measure using techniques available in text analytics. We use the

VentureXpert dataset and focus on startups that are backed by venture capital funding. Using

VentureXpert data allows us to focus on IT entrepreneurs that have received venture capital funding, and

therefore are of baseline quality and represent credible threats to incumbents. It is important to note that

new ventures included in VentureXpert are typically too small and early-stage to count as competitors or

threats to incumbent firms. Therefore, we refine our definition of new entry threat from the startup space

in two ways. First, we argue that the threat of new entry from new ventures does not appear from any

single entrepreneur but from broad, collective movements in the startup space, i.e. evidence of systematic

entrepreneurial movements into a specific area or sub-industry are more representative of new entry threat

for an incumbent. Following Hoberg et al. (2014), we identify new entry threats at the level of the

“industry”; for the purposes of our analysis, we treat the whole set of entrepreneurial ventures in

VentureXpert that receive funding as the relevant “industry”, since they represent collectively the new

entry threat that is faced by incumbents. Second, it is unlikely that all entrepreneurial firms represent

emerging, new entry threat to the incumbent. Therefore, we consider those entrepreneurs that receive

first-stage funding in a given year as posing new entry threat to incumbents in that year. If the

entrepreneurial ecosystem observes value in a specific industry subclass or technology space and

systematically invests in new ventures at the early funding stage, there is likely to be a groundswell of

new ventures associated with this industry subclass entering the VentureXpert dataset in a given year,

which could then potentially lead to significant realized entry in 2-3 years, thereby representing new entry

threat for the incumbent. In summary, the new entry threat measure here is based on (a) new ventures that

receive new first stage venture capital funding, and (b) collective body of all entrepreneurs who receive

first-stage funding, rather than individual entrepreneurs.

This collective body of startups will represent varying levels of new entry threats to incumbents,

depending on how closely the entrepreneurial ventures are related to the primary market of the incumbent.

We therefore require a measure of the similarity between the product portfolios observed in the VC-

funded startup space and the incumbent’s product market. We use the cosine text similarity approach to

14

capture this similarity (Sebastiani 2002). The primary building block to construct the text cosine

similarity metric is the set of unique words that firms use to describe their products in their business

descriptions. For publicly traded firms (incumbents), the source of business descriptions is retrieved from

Section 1 of their 10-K annual filings. For startups, we use their business description from VentureXpert

database. We extract all detailed business descriptions from start-ups that received first-stage funding and

aggregate these descriptions for each year t; these business descriptions are short, with the typical

description consisting of 4-5 sentences. Aggregating these for a given year provides a more representative

and useful document of entrepreneurial entries in a particular year. Cosine similarity between this

collective entrepreneurial document and an incumbent’s business description forms the basis for

measuring new entry threat, effectively by calculating their overlap in word usage.

Specifically, once the respective text documents are available, we parse semantics at a sentence

level with the Natural Language Processing Toolkit (Bird et al. 2009) and retain the nouns and proper

nouns, which are the most meaningful words elements in product descriptions. We remove commonly

used English stop words. We also omit geographical words such as country, state and city names, as well

as the words describing time periods such as months and dates, following Hoberg and Phillips (2015).

Our results are robust to the inclusion of these stop words. Figure 1 presents a histogram of frequencies of

the number of unique words used in the product descriptions of the incumbent firms, showing that the

typical firm uses roughly 700 unique words. Figure 2 displays the number of startups that received first-

round funding, which range from 522 to 2299; and the number of unique words used in the collective

startups’ product descriptions across the years, which range from 3020 to 6508 words per year.

[Insert Figure 1 and Figure 2 about here]

Next, we define all incumbents’ business descriptions and the aggregated start-up document as a

cumulative document corpus (or collection) for each year t (that is, the corpus includes n+1 documents in

total, with n being the number of incumbents in year t). Subsequently, we build document vectors for each

incumbent’s text and the aggregated start-up text in year t. Let Jt denote a scalar equal to the length of the

words dictionary, which includes all unique words used in document corpus of year t. Let Wit represent an

15

ordered vector of length Jt describing the pattern in which the Jt words are used in document i (i = 0

represents the aggregated start-up file) in year t. We use Term Frequency times Inverse Document

Frequency (TF-IDF) (Tata and Patel 2007) as the weight for each word in the document vector. In this

case, each element j in Wit captures the relative importance of word (or term) j in document i, given its

within-document frequency and cross-document frequency. Term Frequency (fit) is defined as the number

of occurrences of words j in document i. The normalized term frequency is defined as:

TFji

=f

ji

maxk

fki

(1)

That is, the term frequency of term j in document i is fji normalized by the maximum number of

occurrences of any term in document i. This normalization process helps correct for biases caused by the

length of a document (i.e. term frequency gets inflated in longer documents). Inverse Document

Frequency (IDFji) for a term is defined in following fashion. Suppose term j appears in nj of the N

documents in the collection, then IDFj = log(N/nj). Naturally, a term that appears in many documents,

such as “service”, gets a lower IDF weight (and therefore is treated as less important), while a term that

occurs in only a few documents, such as “encryption”, gets a higher IDF weight. The weighting score of

TF-IDF for term j in document i thus defined to be TFji × IDFj. Intuitively, words with high within-

document frequency obtain higher weighting and those with high cross-document frequency are weighted

less. Lastly, since our main interest lies in the similarity between the document representing an incumbent

and the aggregate document representing start-ups in that year, we operationalize the text-based measure

of new entry threats for incumbent firm i in year t as:

(2)

where Wit denotes incumbent firm i’s document vector in year t (i = 1,2,3…n) and W0t represents the

aggregated start-ups’ business descriptions document vector. By construction, the cosine TF-IDF-based

16

measure of new entry threat NET_TFIDFit is bounded between [0, 1]. Higher values represent greater

threat of new entry for the incumbent (since the two word vectors are closer in unit vector space).

There are several reasons that the cosine similarity score calculated by using TF-IDF weighted

words vector is a good measure of new entry threat in our study. First, the properties of TF-IDF are well

understood given its wide use in the studies of information processing and text analysis (Aizawa 2003;

Aral and Van Alstyne 2011; Hiemstra 2000). Second, the measure is intuitive given its consideration of

words frequency within and between documents. Third, the method is only moderately computationally

burdensome, making it scalable to replicate or extend to large datasets. Finally, the cosine similarity’s

normalization builds in a natural control for document length, since it measures the angle between two

word vectors on a unit sphere.

3.2 Validation of the Measure of New Entry Threat

On the basis of the methodology defined above, we calculate new entry threat (NET) for all

COMPUSTAT public firms in the high-tech industries (of which IT industries form a subset) from 1997

to 2014. The high-tech industry is formed using the 46 4-digit NAICS codes defined by Hecker (1999).

Since NET is a new measure based on text mining of production descriptions, in this section, we aim to

illustrate the validity of our measure through a series of tests. These include assessing whether our text-

based measure captures changing trends in the startup space over time; examining how NET is associated

with the changes in the competitive dynamics of the incumbent’s industry in subsequent years; comparing

the turbulence experienced by high-NET firms with those experienced by low-NET firms in subsequent

years; and examining how NET in a selected number of industries are influenced by industry-level

demand shocks caused by well-known and major exogenous events. We believe these tests provide

significant validation for the measure, and describe these in some detail below.

3.2.1 Capturing Changing Trends in the Entrepreneurial Space

As a first step to establish the validity of our NET measure, we examine whether our approach of using

text indeed accounts for shifting technology trends within the startup space as reflected in the “hottest”

17

words used in product descriptions of new entrant firms across years in our sample. In Table 3, we present

the list of 20 words with the highest TF-IDF weights in the collective startup product descriptions

document in three selected years – 2000, when the dot-com bubble was at its peak; 2003, when the stock

market reached the lowest point after the collapse of the dot-com bubble; and 2006, when the economy

had recovered from the dot-com bubble.

[Insert Table 3 about here]

We observe several interesting patterns. First, there is significant longitudinal variation in the

most influential words that venture-funded startups use to describe their products and services. Second,

the changes in the vocabulary reflect systematic shifts in technology trends that are consistent with

observations in high-tech. For example, in 2000, the VC-funded entrepreneurial space was dominated by

firms related to the Internet or software industries: words such as “online”, “Internet”, “software”, “web”,

“email”, “broadband”, “ecommerce” and “portal” were among the most frequently used in their product

descriptions. In fact, all but 2 among the top 20 words are related to information and communication

technology industries in 2000. However, in 2003, we observe a significant change in the vocabulary used

to describe funded startups. The use of Internet-related words dramatically reduced, to be replaced by

words such as “disease”, “patient”, “drug”, “treatment”, “therapy”, “protein”, “biotech”, and “antibody”.

These changes show that the VC-funded startup space had shifted systematically from Internet/software

to pharmaceutical and biotech industries after the dot-com bubble. Interestingly, in 2006, we see the word

list reflecting a balance between IT and biotech industries. While some software and Internet-related

terms resurface, they do so with a completely different emphasis. Terms such as “search”, “cloud”, “blog”

“advertising”, “video” and “game” become more influential, reflecting a trend toward cloud computing,

social media, online advertising and video games in the IT industry. Overall, Table 3 provides evidence

for the significant longitudinal variation in the words that are used to describe the entrepreneurial firms in

different years, and shows that our text-based measure of new entry threat captures and builds on these

underlying trends within the startup space with fidelity.

18

3.2.2 New Entry Threat and Competitive Dynamics

If our NET measure does indeed capture the extent to which incumbents face potential new entry, one

way to validate the measure is to examine how a firm’s NET is associated with the changes in its

competitive landscape in the following years. This would imply that firms with higher values of NET are

likely to face, in the subsequent 2-3 years, an increase in the number of direct competitors, all else being

equal. This is likely to happen through a number of mechanisms. To start with, some fraction of the new

startups may eventually go public and become a real rival for the incumbent. In other cases, existing

incumbents may form various types of alliances or joint ventures with the startups, or acquire technology

licensing from them, and invade the product space of the focal incumbent (Mitchell and Singh 1992). We

present in Figure 3 a scatter plot depicting the correlation between the numbers of rivals of an incumbent

defined by text-based network industries classification (TNIC)3 (Hoberg and Phillips 2010) in years t +2

and t+3 and the NET measure for the firm in year t. From the scatter plot, we clearly observe that a higher

level of NET is positive associated with number of rivals an incumbent faces in the next few years,

indicating that some of the entry threats do indeed materialize in the subsequent years. While these plots

show univariate relationships, we also estimated regressions with the number of competitors as the

dependent variable and NET as independent variables, including a full set of control variables. These

results, available upon request, show a significant positive relationship between NET and the number of

competitors in subsequent years for the incumbent firm.

[Insert Figure 3 and Figure 4 about here]

Relatedly, prior literature has shown that one of the direct consequences of rampant new entry is

the reduction of profitability in the industry as a whole, due to increased competitive intensity (Audretsch

and Mata 1995). Some view entry as a mechanism through which profits in excess of long-run

equilibrium are eroded (Mueller 1990). To test these expectations, we again show in Figure 4 a scatter

plot relating NET in year t to a commonly used profitability measure, Return on Assets (ROA), in years

3 Only text-based network industries classification provide us time-varying rival numbers in the same industry (Hoberg and

Phillips 2015). Therefore, we use this classification rather than the static SIC or NAICS codes to define industry classes.

19

t+2 and t+3. As a validity check, firms with higher NET values should experience deteriorating

profitability in the years ahead, as the threats materialize. As expected, the slope of the fitted line of the

scatter plot is negative, showing that higher levels of NET are correlated with decreased operating

performance of the firm in the subsequent years, adding further validity to the NET measure.

3.2.3 Turbulence Experienced by High NET vs. Low NET Firms

One of the defining characteristics of new entry threat is that it threatens the sustainability of the firm’s

future earnings and the viability of its product portfolio (Hoberg et al. 2014). As a result, if a fraction of

the threats indeed materialize, firms with high levels of NET are more likely to suffer from deteriorating

operational performance due to heightened competition, and therefore more likely to experience turbulent

events such as liquidation or downsizing during the difficult time period that follows. We test these

eventualities in this section. We contrast firms with high NET values and those with low NET values to

investigate the likelihood of experiencing turbulent events, such as filing bankruptcy, announcing

significant downsizing or layoffs, or being acquired or merged by other companies in the five years

following the measurement of NET

Specifically, for each year in our NET data time frame (1997 – 2014), we select the ten firms with

the highest NET scores and the ten firms with the lowest NET scores, and put them into high group and

low group respectively. This results in a total of 103 unique firms in the low group and 82 unique firms in

the high group during the sample frame (the number is less than 10 times number of years, because a firm

may appear in a given group in multiple years). For each observation in the high group and the low group,

we conduct a search in the Lexis/Nexis database to identify news releases that are related to the turbulent

events associated with the firm in the subsequent 5 years, using a Boolean query that combines the

company name and keywords such as “bankruptcy”, “liquidation”, “layoff”, “cut jobs”, “merger”, and

“acquisition”. We limit the source to be any of the four types: Newspapers, Business and Industry News,

U.S. Newspapers, Web news. Through this exercise, we identify turbulent events associated with the firm

in the subsequent five years, and collect information such as the exact time of the events, other companies

involved in the event of a merger, and so on. We also search through news releases from the firm to

20

confirm the dates and details of the events thus identified. For the purposes of validation, we report the

frequency of such events across the two sets of firms, representing high and low rates of new entry

threats. Figure 5 reports the comparison of these groups in terms of the rates of incidents for three critical

events – bankruptcies, layoffs, and acquisitions/mergers4. As we expected, companies facing high level

of NET have much greater likelihood of experiencing turbulent events in the next few years. For

companies in the high group, 8.5% (7 companies) of 82 companies filed bankruptcy, 40.2% (33

companies) announced significant layoffs, and 29.3% (24 companies) were acquired by or merged with

other firms in the subsequent 5 years. In comparison, among the companies in the low group, only 4.8%

(5 companies) of 103 companies filed bankruptcy, 7.8% (8 companies) announced layoffs and 15.5% (16

companies) were acquired or merged. In addition, the two-sided t-tests show that the probability of

“Layoffs” and “M&A” are significantly different between the high NET group and the low NET group

(p<0.05 and p<0.01, respectively). On average, companies with low new entry threats fared much better

than companies with high new entry threats in subsequent years. In summary, this provides evidence that

the NET measure is forward-looking and does indeed correlate with firms’ future competitive dynamics.

[Insert Figure 5 about here]

3.2.4 New Entry Threat and Industry-level Exogenous Demand Shocks

A further source of validation for NET arises from being able to observe higher values in NET when an

exogenous event creates the potential for new entry into that industry; in this section, we assess these

effects. The underlying argument here is that a sudden increase (or decrease) in industry-level demand

due to exogenous event disrupts the relationship between supply and demand, and therefore encourages

(or discourage) entry. As a result, incumbents in the industry will likely face increasing (or decreasing if

demand shock is negative) levels of NET following the events. We consider 4 selected high-tech

industries and their associated exogenous events as exemplars. The first is the military armored vehicle

4 Beyond these comparisons, we also delve deeper into the two sets of firms for a given year. This qualitative information is

provided in Appendix 1 for the year 2009.

21

manufacturing industry5 following the terrorist attacks on September 11, 2001. The second event we

consider is changes in the Internet services provider and web search portal industry following the bursting

of the dot-com bubble in March 2000. Third, we consider changes in the software publishing industry

after Apple’s announcement of a major and critical SDK release for iOS, which drove hundreds of app

developers into the mobile apps market. Finally, we trace changes in the biotechnology industry which

experienced a demand boost following the complete sequencing of the human genome. We choose these

events due to their importance in shaping the trajectory of these high-tech industries, thereby potentially

offering new opportunities for entrepreneurial firms.

[Insert Figure 6 about here]

Military Armored Vehicle Manufacture (NAICS 336992). Historical data suggests that the U.S. military

spending reached a peak of nearly 6% of GDP during the Reagan defense buildup, and declined in the

1990s, bottoming out at 3.5% of GDP in 2001.6 But the terrorist attacks on September 11, 2001 reversed

that trend sharply, and defense spending began a substantial increase in the years following the attack,

eventually reaching 4.6% of GDP by 2005 during the Iraq invasion. Spending stabilized after 2005, but

increased further to 5.7% of GDP in 2010 and 2011 with the stepped up effort in Afghanistan.7 We expect

the 9/11 attacks to dramatically boost demand for the military and defense industries, which should lead

to more new entry by entrepreneurial firms in the following years. Figure 6a plots average NET faced by

incumbents in the military armored vehicle manufacturing industry over our sample period (the vertical

line represents the exogenous event of 9/11). We find strong support for our conjecture – we observe a

sudden surge in NET in the industry after 2001, indicating that our measure is responsive to industry-level

exogenous events that provide demand shocks. Furthermore, we find that NET in this industry peaked in

2010, the year in which military spending reaching a new high in recent years. These dynamics provide

some support for the thesis that new entry follows exogenous demand shocks.

5 Military Armored Vehicle Manufacturing industry is the only close related with military goods and services industry in high-

tech sector (Hecker 1999). 6 http://www.usgovernmentspending.com/defense_spending

7 http://www.usgovernmentspending.com/spending_chart_1990_2020USp_XXs2li011tcn_30f_Recent_Defense_Spending

22

Internet Related Industries (NAICS 5181, 5161, and 5182). We next consider the dot-com collapse and

examine its influence on NET values faced by Internet-related industries, which include Internet Service

providers and Web Search Portal (NAICS 5181), Internet Publishing (NAICS 5161), and Data Processing

(NAICS 5182). In contrast to the previous case, this event generated a significant negative demand shock.

The stock market collapsed in March 2000, causing plunges in stock prices of large e-commence and

Internet companies, including Amazon, eBay, and Cisco, while making investors reconsider the valuation

of Internet service companies (Wheale and Amin 2003). As a result, we expect VCs to shy away from

investments in Internet-related firms, thereby reducing new entry threat faced by firms in this industry.

Figure 6b provides evidence in support of this expectation. The average NET of Internet related firms (the

solid line) reached its peak in 2000, and dropped significantly in 2001. The negative trend continued until

year 2003 when the market started to recover from the dot-com bust.

Software Publishers (NAICS 5112). In Figure 6b we also present the average NET experienced by firms

in the software industry (in dashed line). We find that the level of NET for software publishers also

dropped following the dot-com bust. However, the decline in NET is smaller for software publishers than

for Internet related firms, suggesting that software publishers are less severely influenced by this

particular demand shock. In addition, we observe a strong increase in NET values for incumbent software

publishers more recently, starting from 2009. We partly attribute this to Apple’s launch of SDK3

(Software Development Kit 3) for iOS, which ignited exponential growth in its app ecosystem. Although

Apple’s AppStore opened in July 2008, the functionalities of earlier versions of SDK were too restrictive

and prohibited real innovation.8 With the introduction of SDK3, Apple offered over 100 new features to

the framework and more than 1000 new APIs. Some of these features, such as Map API, Payment API,

and push notifications, were critical for developers to operate in this ecosystem (Ghazawneh and

Henfridsson 2010). Thus, we see a corresponding increase in entrepreneurial software developers joining

the mobile app ecosystem, and a concomitant increase in NET values for the industry.

8 http://whydoeseverythingsuck.com/2008/03/apples-iphone-sdk-prohibits-real-mobile.html

23

Biotechnology Research and Development (NAICS 5417). Lastly, we turn to a milestone event in the

biotech industry– the completion of the human genome sequence – and examine how it shaped new entry

threat in the industry. Due to widespread international cooperation and advances in the field of genomics

(especially in sequence analysis), as well as major advances in computing technology, an initial

sequencing of the human genome -- the genetic blueprint for human beings -- was finished in 2000,

announced jointly by President Bill Clinton and Prime Minister Tony Blair on June 26, 2000. 9 In

addition, in 2001, the sequence of the human genome was published in Science and Nature,10

making it

possible for researchers across the world to begin developing treatments. These two events led to the

entry of a large number of biotechnology startups. As seen in Figure 6c, we see the average level of NET

for incumbents in the biotechnology R&D services industry increased dramatically in 2001, a trend that

continues to 2002, thus again verifying that our NET does track closely with external demand shocks that

affect new entry threats. In summary, we believe that our NET measures are valid, and represent a new

method to measure de novo new entry threats using text analysis. We now consider econometric models

that relate to our core question: how do incumbents respond to new entry threats in terms of innovation?

4. Empirical Analyses

We present our empirical analyses in this section in stages. We first start with a description of the sample

and variables we use in our empirical models. As mentioned above, R&D intensity is our primary

measure of innovation investments. We then address the endogeneity of new entry threats by using

dynamic panel models, and by estimating simultaneous equation models that incorporate the role of entry

barriers. We show that our results are robust to the use of alternative innovation measures, such as

number of patent applications and citation-weighted patents. Lastly, we examine how product diversity

within the incumbent firm moderates the relationship between NET and innovation strategy.

9 http://web.ornl.gov/sci/techresources/Human_Genome/project/clinton1.shtml 10 https://www.genome.gov/10002192

24

4.1 Data and Variables

We restrict our analyses to the IT industries, using the 24 4-digit NAICS industry codes that include IT

software, hardware, and services industries for the years 1997-2013 (Kim et al. 2016). Financial data and

other firm characteristics are obtained from Compustat. Our primary dataset consists of 2101 publicly

traded firms over the period 1997-2013 with 14,410 firm-year observations, representing an unbalanced

panel. The sample period includes years when there was considerable turbulence in the IT industry (e.g.,

during the Internet boom and bubble burst) as well as the less volatile years. The long panel also includes

the period of the global financial crisis in 2008 and the period of recovery afterwards, which significantly

affected IT-related venture capital funding and entrepreneurial activities in general. Together, the dataset

provides considerable longitudinal variation in our measure of new entry threats that allows us to use

firm-level fixed effect models to control for many unobserved firm heterogeneities. We describe the other

variables in our main analyses below.

Innovation: We employ two commonly used proxies for innovation of a firm: R&D intensity and patent

applications. R&D has been viewed as a key determinant and indicator of technological innovation of

firms (Cohen and Klepper 1996) and also measures the degree to which firms pursue product and process

innovation, and has been shown to drive firm growth in the high-tech sector (Coad and Rao 2008).

Following prior literature, we define R&D intensity as R&D expenditures over firm’s total asset

(Blonigen and Taylor 2000; Hall 1988). Unlike patents, which are usually observed with a lag, R&D

expenditures reflect contemporaneous managerial decisions that are more closely associated with a firm’s

innovation strategy. We also use patent applications and citation information as our alternative innovation

measure in robustness tests, based on data obtained from the NBER patent database (Hall et al. 2001). We

use the patent application year in our analysis, since the application year is more close to the time of the

actual innovation (Griliches et al. 1988); and use number of patents applications that were eventually

granted, consistent with prior work (Crépon et al. 1998; Griliches et al. 1988; Hall et al. 1986). To the

extent that patents vary in quality and importance (Griliches et al. 1988), we also construct citation-

25

weighted patent counts (Hall et al. 2001). Following Hall et al. (2001), we correct for the truncation bias

in patent citations using the “weight multiplier” that can be applied to the number of citations from

USPTO through 2006 received by the patent. Since the distributions of patent counts and weighted

citations are right skewed, we use the natural logarithm of patent counts and weighted citations.11

Control Variables: Following the innovation literature (Atanassov 2013; Becker-Blease 2011; Kaplan

and Zingales 1997), we control for a vector of firm characteristics that may affect a firm’s innovation

investments, including firm size, firm age, profitability, asset tangibility, leverage, capital expenditure,

product market competition, growth opportunity, and financial constraints. We control for

contemporaneous competition for the incumbent firm by Herfindahl-Hirschman index based on the

increasingly popular Text-based Network Industry Classification (TNIC) scheme created by Hoberg and

Phillips (2010), where industry classification is based on product market vocabularies. These

classifications are updated every year as firms file 10-K reports, allowing for a more accurate measure of

contemporaneous competition. Detailed variable definitions are described in Table 4, and summary

statistics and correlations are shown in Table 5.

[Insert Table 4 and Table 5 about here]

4.2 Baseline Analysis of NET on Incumbent Innovation

Our baseline model estimates the effect of new entry threats (NET) on the incumbent firm’s innovation

investment, using a two-way fixed effects panel data specification below:

Innovationit

=ht+l

i+ b

1´ NET

it+ X

itg + m

it (3)

where i indexes firms, and t indexes time periods. In the baseline model we use R&D intensity as the

dependent variable Innovationit. The variable NETit is from our new text-based measure of new entry

threats. Xit is a set of firm characteristics that affect a firms’ innovation investment. We control for time-

invariant unobservable firm characteristics by including firm fixed effects λi . We include year fixed

effects ηt to control for economy wide shocks. Standard errors are clustered at the firm-level to control

11 To avoid losing firm-year observations with zero patents or weighted citations, we add one to the actual value before taking the

natural logarithm.

26

for serial correlation (Wooldridge 2010). μit represents the idiosyncratic errors.

[Insert Table 6 about here]

We report the results from fixed effect model in Column (1) of Table 6. For comparison, we also

present a random effects panel data model in column (2). In random effects models, the unobserved

individual heterogeneity λi is assumed to be uncorrelated with the included regressors (Greene 2003). We

find the coefficient of new entry threat (NET) is negative in the fixed effects estimate and significant at

5% level, indicating that greater new entry threats are associated with lower level innovation investments,

all else being equal. The result from a Hausman’s test comparing the fixed effects and random effects

estimates rejects the orthogonality of the random effects and the regressors at p<0.01. Therefore, without

the strong assumption that the firm heterogeneity is uncorrelated with the regressors, random effect

estimates are biased. We therefore interpret our results using the fixed effects estimates.

To illustrate the magnitude of effect of NET on innovation, consider a one standard deviation

(s.d.) increase in new entry threat (before standardization, NET has a mean of 0.07 and s.d. of 0.05). The

coefficient of NET implies a 0.35 percentage point reduction in R&D intensity, which translates to a

reduction of $10.75 million in R&D expenditure based on the mean level of asset in the sample. While

extant literature is divided on the direct effect of the threat of new entry on the incumbent firm’s

innovation strategy, our analysis indicates that in the IT industry, the average firm responds to such

threats by reducing its innovation investments.

While a fixed effects model controls for many unobserved firm heterogeneities, a particular

concern here is that the presence of pre-emptive innovation might deter entry or other unobserved

industry-wide shocks (such as technological opportunities) may influence both entrepreneurial entry, VC

funding decisions and incumbent innovation, causing the endogeneity of new entry threats. Therefore, we

relax the assumption that NET is strictly exogenous and use the generalized method of moments (GMM)-

based dynamic panel data models estimator (Arellano and Bond 1991; Blundell and Bond 1998). Taking

advantage of our long panel and large number of firms in our dataset, we construct internal instruments

within the data consistent with dynamic panel data models. More specifically, we use the differences

27

GMM estimator, employing lag term of our endogenous variables, NET and L.R&D Intensity, and all

difference of other exogenous variables including year dummies as our instrument variables for

differenced equation. We use the second lag and onward of endogenous variables for the difference GMM

specifications12

. We checked the validity of the moment conditions required by the differences GMM

estimator using the Hansen test, which does not reject the assumption that our instruments are exogenous

(Arellano and Bond 1991; Roodman 2009). We also test the validity of generalized method of moments

(GMM) assumptions in our model. The test results are reported at the bottom of Table 6, which indicate

that our model specification shows no significant serial correlation in the first-differenced disturbances.

We report the results from Arellano-Bond estimator of dynamic panel data models, treating NET

as exogenous first in Column (3) and then as endogenous in Column (4) of Table 6. We observe that the

coefficient estimate of NET in Column (3) is similar to that of the fixed effects model, consistent with our

main finding that new entry threats reduce innovation. Moreover, the coefficient of estimate of NET in

column 4 (-1.337) is significantly higher than that from the fixed effects model. The larger estimate in the

dynamic panel data model, after accounting for the endogeneity of NET, suggests that the presence of

endogeneity causes a downward bias in the FE model, while the FE model generates more conservative

estimates of the effect of NET. Overall, the Arellano-Bond estimates provide strong support for the main

finding that threat of future competition through de novo entry reduces the inclination of incumbents to

invest in risky innovation investments. Our results are consistent with existing work showing that firms

facing market turbulence are more likely to respond conservatively (Brav et al. 2005; Hoberg et al. 2014).

This strategy of conservativeness appears to extend to innovation-related investments as well, given the

high risk of failure in such activities and the uncertainty of appropriating the returns from such activities

(Lukach et al. 2007, Fontana et al. 2009).

4.3 Simultaneous Equations Models

Beyond the dynamic panel model to account for the endogeneity of NET, we use an alternative

12 In unreported analysis, we try varying levels of lags and their combinations; our results are fully robust to the selection of lag

structures.

28

identification strategy that involves modeling the endogenous variable through a separate equation. We

thus propose and estimate jointly a system of equations to account for the endogeneity of NET

(Wooldridge 2002, p.209). We use the method of Simultaneous Equations Models (SEMs) to address the

reverse causality issues – i.e. incumbent’s R&D investments may lead to entry barrier (Mueller and Tilton

1969). As shown by Orr (1974) and Cockburn and MacGarvie (2011), new entry in an industry is likely

influenced by industry level innovation investments. For example, in Orr (1974)’s model, entry is

assumed to be a function of the incentives to enter relative to the level of entry barrier. Such entry

incentives and barriers may include factors such as industry-level R&D intensity, the growth rate in the

industry, the level of profitability observed in the industry, industry-level uncertainty and industry

concentration. In order to model how startups may make the decision to enter a certain industry (and

thereby endogenize NET), we build on his model to derive the following equation, which is then

estimated as part of a simultaneous equations model along with equation (3):

NETit

=at+t

i+q

1×Ind _ R & DIntensity

i ,t-1

+q2×Ind _ SaleGrowthRate

i ,t-1

+q3×Ind _Concentration

i ,t-1

+q4×Ind _ ProfitExp

i ,t-1

+q5×Ind _Uncertainty

i ,t-1+u

it

(4)

where t indexes time periods, and i indexes incumbents. We define industry using the TNIC (Hoberg and

Phillips 2015), which allows each firm to have its own unique set of competitors identified by the

similarity of product descriptions in their annual reports. NETit is a proxy for the level of entry into

incumbent i’s industry at year t, and is modeled as a function of lagged value of average R&D intensity,

average sale growth rate, expectation for future profit13

, industry concentration, and market uncertainty of

the incumbent’s industry. Detailed variable constructions are described in Table 4. We control for time-

invariant unobservable firm characteristics by including fixed effects, τi, and use year fixed effects αt to

13 A commonly used proxy for the expectation of future profit is Tobin’s Q (Jankowski 1998).

29

control for economy wide shocks. Standard errors are clustered at the firm-level to control for serial

correlation (Wooldridge 2010). υit represents the idiosyncratic error. We estimate this equation together

with equation (3) – which models incumbent firm’s innovation investments – as a simultaneous equations

system to establish the causal inference of new entry threats on incumbent’s innovation strategy. We use

three-stage least squares (3SLS) estimation procedures. The 3SLS estimator assumes that the errors are

correlated across equations and is more efficient than estimation of each equation by two-stage least

squares (Cameron and Trivedi 2005).

[Insert Table 7 about here]

The results of 3SLS estimation are shown in Table 7. Qualitatively, the results from the

Incumbent Equation. (i.e. equation (3)) are similar to the FE specification in Table 6: the coefficient of

NET is negative and statistically significant, providing strong evidence for the causal relationship between

new entry threats and incumbent innovation. We note that the coefficient estimate of NET is larger than

that from the fixed effects model in Table 6, which again suggests that the endogeneity of NET due to

reverse causality likely biases the coefficient estimate downwards (as suggested by the Arellano-Bond

models). The FE model thus generates a more conservative estimate of the effect of NET on innovation.

Interestingly, the estimation of the Entry Equation provides yet another validation test for our

text-based measure of new entry threats. The results from this equation are highly consistent with prior

work on entry incentives and entry barriers (Bresnahan and Reiss 1991; Cockburn and MacGarvie 2011;

Orr 1974). The coefficients for Industry Sale Growth Rate and Expectation for Future Profit are both

positive and significantly associated with new entry threats (p<0.05 and p<0.01, respectively) since the

past industry growth rate and profit in future is a strong incentive to enter (Orr 1974; Siegfried 1994).

Industry R&D intensity is negative and significantly associated with new entry (p<0.01), as R&D

investments generate intellectual property rights that deter entry (Cockburn and MacGarvie 2011).

Industry Concentration also deters entry (p<0.01), as predicted by prior literature. In highly concentrated

industries, the incumbents have greater advantage of economies of scale over potential entrants (Hurdle et

al. 1989), and are likely to use strategic entry deterrence more often (Bunch and Smiley 1992). Thus, the

30

results from the first stage equation provide further support for the validity of the NET measure as well.

4.4 Alternative Innovation Outcome Measures

In this section, we use the number of patent and patent citation from the NBER database (Hall et al. 2001)

as alternative measures of innovation to further corroborate the relationship between new entry threats

and incumbent innovation we discover. As discussed before, we use the patent application year to

construct our innovation variables. In order to address the truncation issue with patent citation, we take

two measures. First, we use the “weight multiplier”, which can be applied to the number of citations from

USPTO through 2006 received by the patent, to correct for the right hand side truncation bias in patent

citations (Hall et al. 2001). Second, we limit our sample to the period of 1997-2002 to avoid the left hand

side truncation problem of NBER patent data close to 2006. This is because a typical patent is granted

about two years after application (Jaffe et al. 1993). 14

Our final data therefore includes 7,824

observations associated with 2,089 firms. We test the same Two-Way Fixed Effect specification in

equation (3) by replacing the dependent variable Innovationit with one of the following two measures: the

natural log of patent application counts, Ln(1+Patentt+2), or the natural log of the citation-weighted patent

application counts, Ln(1+WeightedCitationt+2). We impose a two-year lag on the observed new entry

threat variables as well as other firm level controls relative to innovation outcome since the actual

investments in these innovations are likely to have been made earlier (Chemmanur and Tian 2012). The

two-year lag also allows for sufficient time on the part of the firm to react to new entry threats as well as

reduces the chances of reverse causality in the regression.

[Insert Table 8 about here]

Table 8 reports the results of the tests with patent as innovation output. Panel A reports results

with the full IT industry sample. Here again, we find that the coefficient of NET is negative in both

column 1 and column 2 but only significant for citation-weighted patents at the 1% level, indicating that

greater new entry threats are associated with fewer high-quality patent applications from incumbents.

14

Therefore most patent applications filed in 2005 and 2006 do not appear in the NBER patent database. We thank an

anonymous reviewer for suggesting this.

31

Within the IT industry in particular, patent protection tends to be weaker in the service sectors than the

hardware/software product sectors, which leads to a lower propensity for patenting. Therefore patent

output may not be a perfect measure for inventive activities in IT services. In Panel B of Table 8 we

report the results after excluding IT service sectors. 15

We find that for this subsample, the estimates of

NET are negative and significant for both the number of patents as well as citation-weighted patents.

Comparing the results in Panel B to those in Panel A, it is clear that the coefficient estimates are of higher

magnitude, suggesting that new entry threats have a greater impact on patenting in the hardware/software

industries than in IT service industries. By our estimate from column 3 and column 4, a one s.d. increase

in new entry threat is associated with a 2.7% reduction in patent applications and a 12.4% reduction in

citation-weighted patents. Overall, our analyses with patent data confirm that our central finding is robust.

4.5 Reaction to New Entry Threat by Diversified vs. Specialized Firms

Prior research has found that lateral entry from other incumbents triggers a stronger reaction from more

focused, specialized firms than diversified firms in terms of market expansion/retrenchment (Khanna and

Tice 2000). In this section, we examine whether diversified firms respond differently to de novo entry

threats than specialized firms. We argue there are at least two reasons why a specialized firm may react to

new entry threats more strongly than a diversified firm. First, specialized firms may be better at detecting

emerging product market threats from startups, due to their focus. Managers in specialized firms closely

monitor the competitive landscape in their narrow product fields. In contrast, managers in diversified

firms have to divide their attention across a variety of product markets, and may not be able to detect

entry threats in every product market due to their limited information processing capacities. Second, by

virtue of a diversified product portfolio, diversified firms are able to diversify their risks across different

product lines. When they face entry threats in a particular market, they may find that other lines of

business provide a hedge against these threats. Diversified firms are better at weathering the turbulence

caused by new entry, and therefore have less incentive to react as dramatically.

15 IT product subsample is defined with full sample excluding IT services sector, identified by 4 digital NAICS code:

5182,5413,5415,5416, and 5417.

32

We analyze these differences between diversified firms and specialized firms by constructing a

firm product diversification variable using Jacquemin and Berry (1979) entropy measure of sale shares in

different lines of business. This measure has been used widely in prior literature (Wiersema and Bowen

2008). Entropy is computed using data on firm sales in each 6-digit NAICS businesses as reported by the

COMPUSTAT Segment database. Mathematically, Entropy for firm i in year t is defined as following:

Entropyit

= Psit

s=1

n

å ×ln1

Psit

(5)

where n denotes the total number of 6-digit NAICS business sectors reported by firm i in year t, and Psit is

the share of sales from business sector s over the total sales of firm i in year t. Because not all

COMPUSTAT firms report sales by lines of business, we are able to calculate this variable for 954 firms

in our sample. We then define a firm’s product Diversityi as the within-firm mean of Entropyit over the

sample period.

[Insert Table 9 about here]

In Table 9 we report the regression results with the firm diversification measure. In Column (1),

we report the estimates from a fixed effects model that incorporates the interaction term of NET and

Diversity. The interaction term is positive and statistically significant (p<0.01), showing that consistent

with our conjecture, firms with higher level of product diversification response less dramatically in

reducing innovation investments. We also present split sample analyses comparing diversified firms to

specialized firms (Column (2) and Column (3)), using the sample mean of Diversity to split the sub-

samples. Interesting, we find that only specialized firms adjust their innovation investments according to

new entry threats, while diversified firms seem to be much more inertial, thus confirming the insight from

the literature that diversified firms are less likely to respond aggressively to new entry threats through

innovation spending.

33

5. Discussion and Conclusion

It is widely acknowledged that innovation is one of the central engines of competitive advantage in the IT

industry (Giarratana 2004). It is equally well established that the IT industry tends to be volatile, with

quick clockspeeds, rapid change, and hypercompetition (McAfee and Brynjolfsson 2008). Juxtaposing

these two observations begs the question – how do incumbent IT firms respond to increased threats of

new entry in terms of their innovation strategies? This question remains, surprisingly, understudied in

extant IS research. In this study, we examine this question by studying how incumbents respond to

emerging threats from new, entrepreneurial firms. As the means to answering this question, we first

develop a text-based measure of new entry threats by analyzing the product descriptions of both

incumbent firms and startups. We conduct a series of validation tests and show that the NET measure

indeed captures impending threats from the startup space. Subsequently, using data on a panel of 2101

firms in the IT industry over the period of 1997-2013, we find that increasing level of new entry threats is

associated with a reduction in a firm’s innovation investments, measured by R&D intensity. Our finding

is robust to the use of alternative measures of innovation outputs, including the count of patent

applications and citation-weighted patent applications, as well as alternative regression specifications

addressing the endogeneity of NET. Across all these tests, we show highly consistent results – that IT

firms tend to reduce their innovation investments when faced with new entry threats from startups.

There are several potential explanations for the relationship we uncover between new entry threat

and innovation investments. As suggested by theory, facing greater threats of product market entry,

managers may choose strategic alliances to gain a window into technological changes (Cui and O'Connor

2012; Mitchell and Singh 1992) rather than directly invest in R&D. Alternatively, the incumbent firms,

realizing that their research efforts in exploiting radical innovation are not as efficient as new entrants

(Henderson 1993), may invest in corporate venture capital to fund external startups directly (Kim et al.

2016). Another mechanism may involve managers deciding to improve efficiency or cut slack within the

firm as a response to the incipient instability emerging from threats of new entry (Schmidt 1997); this

would manifest as a reduction in innovation but an increase in cash holdings as a buffer (Hoberg et al.

34

2014), which may then be used in acquiring and strengthening complementary assets essential in the

commercialization of new innovations (Teece 1986). While a clear identification of these mechanisms is

beyond the scope of this work, we conduct some exploratory analyses to understand how firms may be

responding to new entry threats, and present details on one of these analyses: cash holdings by the

incumbents – in Appendix 2. We find that on average, incumbent firms also react to NET by increasing

their cash holdings, consistent with Hoberg et al’s (2014) finding that firms are more financially

conservative when their product markets become more volatile. Interestingly, here again we find that

specialized firms react more strongly (hold more cash) than diversified firms, which echoes our earlier

findings with regard to R&D intensity, and is consistent with prior literature (Khanna and Tice 2000).

Our work makes several useful contributions to the literature on innovation within the IT

industry. First, although prior literature has investigated many forms of strategic responses to entry,

empirical studies in this field have largely focused on diversifying entry instead of de novo entry (Carroll

et al. 1996), partly due to the difficulty of measuring such entries. In addition, while there have been

plenty of papers investigating incumbent responses in terms of pricing, excessive capacity, product

market mix, or advertising, very few of them (Aghion et al. 2009) have examined how incumbents may

respond through innovation. More to the point, while many studies have examined the effects of entry

after entry has occurred, relatively few have addressed the question of how firms respond before entry

materializes; some rare exceptions include Seamans (2013) and Goolsbee and Syverson (2008). Our work

here pushes the frontier on both dimensions. We provide an empirical answer to the question of how new

entry threat affects innovation investments in the IT industry, and also provide a new measure that

addresses ex ante threat of entry from the startup ecosystem, a vital source of disruption in the IT

industry. Our work thus fills these two vital gaps in the literature on entry and innovation.

Second, existing theories are divided in their predictions of the relationship between new entry

threats and innovation. As we highlighted earlier, while the pre-emptive market expansion theory (Eaton

and Lipsey 1979) predicts that incumbent firms should increase the level of innovation in the face of new

entry, the theories of heterogeneous research capabilities (Chandy and Tellis 2000; Hamberg 1963;

35

Mueller and Tilton 1969), strategic interaction (Mitchell and Singh 1992), and post-entry defense models

(Gatignon et al. 1989; Kuester et al. 1999) all make the opposite prediction. By presenting empirical

evidence in the U.S. IT industry, we show that an average incumbent firm reacts to the challenge of new

entry more conservatively, instead of competing with startups head-on. There may be good reasons for

this strategic behavior; returning to our example of Intel, while the firm has reacted aggressively to the

threats brought by ARM’s low power, RISC-based processes through its own Atom mobile processors,

the success of this strategy is questionable. Five years after the introduction of the Atom processor, it is

still a distant second to ARM chips, while Intel’s mobile group has reported a loss of $7 Billion over the

period 2013-2014 alone. 16

Clearly, we cannot draw causal conclusions here, but this example provides

some context for why incumbents may choose to draw down their innovation spending when faced with

new entry threats from startups.

In addition to theoretical contributions, we make a methodological contribution by creating and

validating a new measurement of new entry threat from the startup ecosystem. Our text-mining approach,

in contrast to earlier measurement of market threats based on industry classifications or market shares, not

only captures forward-looking threats in a firm’s competitive environment, but also changes over time as

firms enter and exit certain product markets, as shown in the case of the TNICs by Hoberg and Phillips

(2015). It is our intention to provide these measures to the broader IS community so as to allow the

community to further test and verify the value of these measures. There are several empirical and

theoretical contexts where new entry threats faced by incumbents play a central role; the availability of a

standard and accepted measure will help by allowing for comparability across models and theories. We

hope that our work here will encourage the use of this measure, as well as the development of similar

text-based approaches to studying the important topic of competitive dynamics in the IT industry.

16 http://appleinsider.com/articles/14/11/16/after-losing-apples-ipad-business-intel-has-bled-7-billion-while-heavily-subsidizing-

cheap-x86-atom-android-tablets

36

References

Abernathy, W. J., and Utterback, J. M. 1978. "Patterns of Industrial Innovation," Technology Review

(80:7), pp. 40-47.

Abrahamson, E. 1991. "Managerial Fads and Fashions: The Diffusion and Rejection of Innovations,"

Academy of Management Review (16:3), pp. 586-612.

Agarwal, R., and Gort, M. 2002. "Firm and Product Life Cycles and Firm Survival," American Economic

Review (92:2), pp. 184-190.

Aghion, P. 2003. "Empirical Estimates of the Relationship between Product Market Competition and

Innovation," in Institutions, Innovation and Growth: Selected Economic Papers. Edward Elgar

Publishing, pp. 142-169.

Aghion, P., Blundell, R., Griffith, R., Howitt, P., and Prantl, S. 2009. "The Effects of Entry on Incumbent

Innovation and Productivity," The Review of Economics and Statistics (91:1), pp. 20-32.

Aghion, P., and Howitt, P. 1992. "A Model of Growth through Creative Destruction," Econometrica

(60:2), pp. 323-351.

Aizawa, A. 2003. "An Information-Theoretic Perspective of Tf–Idf Measures," Information Processing &

Management (39:1), pp. 45-65.

Aral, S., and Van Alstyne, M. 2011. "The Diversity-Bandwidth Trade-Off," American Journal of

Sociology (117:1), pp. 90-171.

Arellano, M., and Bond, S. 1991. "Some Tests of Specification for Panel Data: Monte Carlo Evidence and

an Application to Employment Equations," The Review of Economic Studies (58:2), pp. 277-297.

Atanassov, J. 2013. "Do Hostile Takeovers Stifle Innovation? Evidence from Antitakeover Legislation

and Corporate Patenting," The Journal of Finance (68:3), pp. 1097-1131.

Audretsch, D. B., and Mata, J. 1995. "The Post-Entry Performance of Firms: Introduction," International

Journal of Industrial Organization (13:4), pp. 413-419.

Bain, J. S. 1956. Barriers to New Competition: Their Character and Consequences in Manufacturing

Industries. Harvard University Press Cambridge, MA.

Becker-Blease, J. R. 2011. "Governance and Innovation," Journal of Corporate Finance (17:4), pp. 947-

958.

Bikhchandani, S., Hirshleifer, D., and Welch, I. 1998. "Learning from the Behavior of Others:

Conformity, Fads, and Informational Cascades," The Journal of Economic Perspectives (2:3), pp.

151-170.

Bird, S., Klein, E., and Loper, E. 2009. Natural Language Processing with Python. O'Reilly Media, Inc.

Blonigen, B. A., and Taylor, C. T. 2000. "R&D Intensity and Acquisitions in High-Technology

Industries: Evidence from the Us Electronic and Electrical Equipment Industries," The Journal of

Industrial Economics (48:1), pp. 47-70.

Blundell, R., and Bond, S. 1998. "Initial Conditions and Moment Restrictions in Dynamic Panel Data

Models," Journal of Econometrics (87:1), pp. 115-143.

Blundell, R., Griffith, R., and Van Reenen, J. 1999. "Market Share, Market Value and Innovation in a

Panel of British Manufacturing Firms," The Review of Economic Studies (66:3), pp. 529-554.

Brav, A., Graham, J. R., Harvey, C. R., and Michaely, R. 2005. "Payout Policy in the 21st Century,"

Journal of Financial Economics (77:3), pp. 483-527.

Bresnahan, T. F., and Reiss, P. C. 1991. "Entry and Competition in Concentrated Markets," Journal of

Political Economy (99:5), pp. 977-1009.

Brown, L. D., and Caylor, M. L. 2006. "Corporate Governance and Firm Valuation," Journal of

Accounting and Public Policy (25:4), pp. 409-434.

Brynjolfsson, E., and McAfee, A. 2011. Race against the Machine: How the Digital Revolution Is

Accelerating Innovation, Driving Productivity, and Irreversibly Transforming Employment and

the Economy. Lexington, MA: Digital Frontier Press.

Bunch, D. S., and Smiley, R. 1992. "Who Deters Entry? Evidence on the Use of Strategic Entry

Deterrents," The Review of Economics and Statistics (74:3), pp. 509-521.

37

Cameron, A. C., and Trivedi, P. K. 2005. Microeconometrics: Methods and Applications. Cambridge

university press.

Carrillo, J. E. 2005. "Industry Clockspeed and the Pace of New Product Development," Production and

Operations Management (14:2), pp. 125-141.

Carroll, G. R., Bigelow, L. S., Seidel, M. D. L., and Tsai, L. B. 1996. "The Fates of De Novo and De Alio

Producers in the American Automobile Industry 1885–1981," Strategic Management Journal

(17:S1), pp. 117-137.

Caves, R. E., and Porter, M. E. 1977. "From Entry Barriers to Mobility Barriers: Conjectural Decisions

and Contrived Deterrence to New Competition," The Quarterly Journal of Economics (91:2), pp.

241-261.

Chandy, R. K., and Tellis, G. J. 2000. "The Incumbent’s Curse? Incumbency, Size, and Radical Product

Innovation," Journal of marketing (64:3), pp. 1-17.

Chemmanur, T., and Tian, X. 2012. "Do Anti-Takeover Provisions Spur Corporate Innovation?," AFA

2012 Chicago Meetings Paper), Available at SSRN: http://ssrn.com/abstract=1572219.

Chen, M.-J., Smith, K. G., and Grimm, C. M. 1992. "Action Characteristics as Predictors of Competitive

Responses," Management Science (38:3), pp. 439-455.

Coad, A., and Rao, R. 2008. "Innovation and Firm Growth in High-Tech Sectors: A Quantile Regression

Approach," Research Policy (37:4), pp. 633-648.

Cockburn, I. M., and MacGarvie, M. J. 2011. "Entry and Patenting in the Software Industry,"

Management Science (57:5), pp. 915-933.

Cohen, W. M., and Klepper, S. 1996. "Firm Size and the Nature of Innovation within Industries: The

Case of Process and Product R&D," The Review of Economics and Statistics (78:2), pp. 232-243.

Crépon, B., Duguet, E., and Mairessec, J. 1998. "Research, Innovation and Productivity: An Econometric

Analysis at the Firm Level," Economics of Innovation and New Technology (7:2), pp. 115-158.

Cui, A. S., and O'Connor, G. 2012. "Alliance Portfolio Resource Diversity and Firm Innovation," Journal

of Marketing (76:4), pp. 24-43.

Eaton, B. C., and Lipsey, R. G. 1979. "The Theory of Market Pre-Emption: The Persistence of Excess

Capacity and Monopoly in Growing Spatial Markets," Economica (46:182), pp. 149-158.

Fine, C. H. 1998. Clockspeed: Winning Industry Control in the Age of Temporary Advantage. Reading,

MA: Perseus Books.

Fontana, R., and Nesta, L. 2009. "Product Innovation and Survival in a High-Tech Industry," Review of

Industrial Organization (34:4), pp. 287-306.

Gatignon, H., Anderson, E., and Helsen, K. 1989. "Competitive Reactions to Market Entry: Explaining

Interfirm Differences," Journal of Marketing Research (26:1), p. 44.

Geroski, P. A. 1995. "What Do We Know About Entry?," International Journal of Industrial

Organization (13:4), pp. 421-440.

Ghazawneh, A., and Henfridsson, O. 2010. "Governing Third-Party Development through Platform

Boundary Resources," the International Conference on Information Systems (ICIS): AIS

Electronic Library (AISeL), pp. 1-18.

Giarratana, M. S. 2004. "The Birth of a New Industry: Entry by Start-Ups and the Drivers of Firm

Growth: The Case of Encryption Software," Research Policy (33:5), pp. 787-806.

Gilbert, R. J., and Newbery, D. M. 1982. "Preemptive Patenting and the Persistence of Monopoly,"

American Economic Review (72:3), pp. 514-526.

Gompers, P., and Lerner, J. 2001. "The Venture Capital Revolution," Journal of Economic Perspectives

(15:2), pp. 145-168.

Goolsbee, A., and Syverson, C. 2008. "How Do Incumbents Respond to the Threat of Entry? Evidence

from the Major Airlines," The Quarterly Journal of Economics (123:4), pp. 1611-1633.

Greene, W. H. 2003. Econometric Analysis. Pearson Education India.

Griliches, Z., Pakes, A., and Hall, B. H. 1988. "The Value of Patents as Indicators of Inventive Activity."

NBER Working Paper No. 2083. Available at: http://www.nber.org/papers/w2083.

38

Hall, B. H. 1988. "The Effect of Takeover Activity on Corporate Research and Development," in

Corporate Takeovers: Causes and Consequences. University of Chicago Press, pp. 69-100.

Hall, B. H., Griliches, Z., and Hausman, J. A. 1986. "Patents and R&D: Is There a Lag?," International

Economic Review (27:2), pp. 265-284.

Hall, B. H., Jaffe, A. B., and Trajtenberg, M. 2001. "The Nber Patent Citation Data File: Lessons, Insights

and Methodological Tools," in Patents, Citations and Innovations. Cambridge, MA.

Hamberg, D. 1963. "Invention in the Industrial Research Laboratory," Journal of Political Economy

(71:2), pp. 95-115.

Hart, O. D. 1983. "The Market Mechanism as an Incentive Scheme," Bell Journal of Economics (14:2),

pp. 366-382.

Hauser, J. R., and Shugan, S. M. 1983. "Defensive Marketing Strategies," Marketing Science (2:4), pp.

319-360.

Hecker, D. 1999. "High-Technology Employment: A Broader View," Monthly Lab. Rev. (122:6), pp. 18-

28.

Henderson, R. 1993. "Underinvestment and Incompetence as Responses to Radical Innovation: Evidence

from the Photolithographic Alignment Equipment Industry," The Rand Journal of Economics

(24:2), p. 248.

Hiemstra, D. 2000. "A Probabilistic Justification for Using Tf× Idf Term Weighting in Information

Retrieval," International Journal on Digital Libraries (3:2), pp. 131-139.

Hoberg, G., and Phillips, G. 2010. "Product Market Synergies and Competition in Mergers and

Acquisitions: A Text-Based Analysis," Review of Financial Studies (23:10), pp. 3773-3811.

Hoberg, G., Phillips, G., and Prabhala, N. 2014. "Product Market Threats, Payouts, and Financial

Flexibility," The Journal of Finance (69:1), pp. 293-324.

Hoberg, G., and Phillips, G. M. 2015. "Text-Based Network Industries and Endogenous Product

Differentiation," Journal of Political Economy (Forthcoming).

Hurdle, G. J., Johnson, R. L., Joskow, A. S., Werden, G. J., and Williams, M. A. 1989. "Concentration,

Potential Entry, and Performance in the Airline Industry," The Journal of Industrial Economics

(38:2), pp. 119-139.

Jacquemin, A. P., and Berry, C. H. 1979. "Entropy Measure of Diversification and Corporate Growth,"

The Journal of Industrial Economics (27:4), pp. 359-369.

Jaffe, A. B., Trajtenberg, M., and Henderson, R. 1993. "Geographic Localization of Knowledge

Spillovers as Evidenced by Patent Citations," The Quarterly Journal of Economics (108:3), pp.

577-598.

Jankowski, R. 1998. Profits, Taxes, and the State. Greenwood Publishing Group.

Jorgenson, D. W., and Stiroh, K. J. 1999. "Information Technology and Growth," American Economic

Review (89:2), pp. 109-115.

Jorgenson, D. W., Stiroh, K. J., Gordon, R. J., and Sichel, D. E. 2000. "Raising the Speed Limit: U. S.

Economic Growth in the Information Age.," Brookings Papers on Economic Activity (31:1), pp.

125-235.

Jovanovic, B., and MacDonald, G. 1994. "The Life-Cycle of a Competitive Industry," Journal of Political

Economy (102:2), pp. 322-347.

Judd, K. L. 1985. "Credible Spatial Preemption," The RAND Journal of Economics (16:2), pp. 153-166.

Kaplan, S. N., and Zingales, L. 1997. "Do Investment-Cash Flow Sensitivities Provide Useful Measures

of Financing Constraints?," The Quarterly Journal of Economics (112:1), pp. 169-215.

Khanna, N., and Tice, S. 2000. "Strategic Responses of Incumbents to New Entry: The Effect of

Ownership Structure, Capital Structure, and Focus," The Review of Financial Studies (13:3), pp.

749-779.

Khessina, O. M., and Carroll, G. R. 2008. "Product Demography of De Novo and De Alio Firms in the

Optical Disk Drive Industry, 1983-1999," Organization Science (19:1), pp. 25-38.

39

Kim, K., Gopal, A., and Hoberg, G. 2016. "Product Market Competition and Corporate Venture Capital

Investment in the It Industry: An Empirical Analysis," Information Systems Research

(Forthcoming).

Kleis, L., Chwelos, P., Ramirez, R. V., and Cockburn, I. 2012. "Information Technology and Intangible

Output: The Impact of It Investment on Innovation Productivity," Information Systems Research

(23:1), pp. 42-59.

Klepper, S. 1996. "Entry, Exit, Growth, and Innovation over the Product Life Cycle," The American

Economic Review (86:3), pp. 562-583.

Klepper, S., and Simons, K. L. 1993. "Industry Shakeouts and Technological Change," International

Journal of Industrial Organization. (23:1), pp. 23-43.

Klepper, S., and Simons, K. L. 1997. "Technological Extinctions of Industrial Firms: An Inquiry into

Their Nature and Causes," Industrial and Corporate Change (6:2), pp. 379-460.

Koski, H. A., and Majumdar, S. K. 2002. "Paragons of Virtue? Competitor Entry and the Strategies of

Incumbents in the U.S. Local Telecommunications Industry," Information Economics and Policy

(14:4), pp. 453-480.

Kuester, S., Homburg, C., and Robertson, T. S. 1999. "Retaliatory Behavior to New Product Entry,"

Journal of Marketing (63:4), pp. 90-106.

Lieberman, M. B. 1987. "Postentry Investment and Market Structure in the Chemical Processing

Industries," The RAND Journal of Economics (18:4), pp. 533-549.

Lukach, R., Kort, P. M., and Plasmans, J. 2007. "Optimal R&D Investment Strategies under the Threat of

New Technology Entry," International Journal of Industrial Organization (25:1), pp. 103-119.

Markides, C. 2006. "Disruptive Innovation: In Need of Better Theory," Journal of Product Innovation

Management (23:1), pp. 19-25.

McAfee, A., and Brynjolfsson, E. 2008. "Investing in the It That Makes a Competitive Difference,"

Harvard Business Review (86:7/8), p. 98.

McCann, B. T., and Vroom, G. 2010. "Pricing Response to Entry and Agglomeration Effects," Strategic

Management Journal (31:3), pp. 284-305.

Mendelson, H., and Pillai, R. R. 1998. "Clockspeed and Informational Response: Evidence from the

Information Technology Industry," Information Systems Research (9:4), pp. 415-433.

Mendelson, H., and Pillai, R. R. 1999. "Industry Clockspeed: Measurement and Operational

Implications," Manufacturing & Service Operations Management (1:1), pp. 1-20.

Milgrom, P., and Roberts, J. 1982. "Predation, Reputation, and Entry Deterrence," Journal of economic

theory (27:2), pp. 280-312.

Mitchell, W. 1989. "Whether or When? Probability and Timing of Incumbents' Entry into Emerging

Industrial Subfields," Administrative Science Quarterly (34:2), p. 208.

Mitchell, W., and Singh, K. 1992. "Incumbents' Use of Pre-Entry Alliances before Expansion into New

Technical Subfields of an Industry," Journal of Economic Behavior & Organization (18:3), pp.

347-372.

Mueller, D. C. 1990. "Profits and the Process of Competition," in The Dynamics of Company Profits: An

International Comparison. Cambridge University Press, pp. 1-14.

Mueller, D. C., and Tilton, J. E. 1969. "Research and Development Costs as a Barrier to Entry," The

Canadian Journal of Economics/Revue Canadienne d'Economique (2:4), pp. 570-579.

O’Connor, M., and Rafferty, M. 2012. "Corporate Governance and Innovation," Journal of Financial and

Quantitative Analysis (47:02), pp. 397-413.

Orr, D. 1974. "The Determinants of Entry: A Study of the Canadian Manufacturing Industries," The

Review of Economics and Statistics (55:1), pp. 58-66.

Porter, M. E. 2008. "The Five Competitive Forces That Shape Strategy," Harvard Business Review

(86:1), pp. 25-40.

Prusa, T. J., and Schmitz Jr, J. A. 1991. "Are New Firms an Important Source of Innovation?: Evidence

from the Pc Software Industry," Economics Letters (35:3), pp. 339-342.

40

Prusa, T. J., and Schmitz Jr, J. A. 1994. "Can Companies Maintain Their Initial Innovative Thrust? A

Study of the Pc Software Industry," The Review of Economics and Statistics (76:3), pp. 523-540.

Rai, A., and Tang, X. 2013. "Research Commentary-Information Technology-Enabled Business Models:

A Conceptual Framework and a Coevolution Perspective for Future Research," Information

Systems Research (25:1), pp. 1-14.

Reinganum, M. R. 1983. "The Anomalous Stock Market Behavior of Small Firms in January: Empirical

Tests for Tax-Loss Selling Effects," Journal of Financial Economics (12:1), pp. 89-104.

Rigby, D. K., Christensen, C. M., and Johnson, M. 2002. "Foundations for Growth: How to Identify and

Build Disruptive New Businesses," MIT Sloan Management Review (43:3), pp. 22-32.

Roodman, D. 2009. "How to Do Xtabond2: An Introduction to Difference and System Gmm in Stata,"

Stata Journal (9:1), pp. 86-136.

Schmalensee, R. 1978. "Entry Deterrence in the Ready-to-Eat Breakfast Cereal Industry," The Bell

Journal of Economics (9:2), pp. 305-327.

Schmidt, K. M. 1997. "Managerial Incentives and Product Market Competition," The Review of Economic

Studies (64:2), pp. 191-213.

Seamans, R. C. 2013. "Threat of Entry, Asymmetric Information, and Pricing," Strategic Management

Journal (34:4), pp. 426-444.

Sebastiani, F. 2002. "Machine Learning in Automated Text Categorization," ACM Computing Surveys

(CSUR) (34:1), pp. 1-47.

Shankar, V. 1999. "New Product Introduction and Incumbent Response Strategies: Their Interrelationship

and the Role of Multimarket Contact," Journal of Marketing Research (36:3), pp. 327-344.

Siegfried, J. J. 1994. "Empirical Studies of Entry and Exit: A Survey of the Evidence," Review of

Industrial Organization (9:2), p. 121.

Simon, D. 2005. "Incumbent Pricing Responses to Entry," Strategic Management Journal (26:13), pp.

1229-1248.

Spence, A. M. 1977. "Entry, Capacity, Investment and Oligopolistic Pricing," The Bell Journal of

Economics (8:2), pp. 534-544.

Sutton, J. 1991. Sunk Costs and Market Structure: Price Competition, Advertising, and the Evolution of

Concentration. MIT press.

Tambe, P., Hitt, L. M., and Brynjolfsson, E. 2012. "The Extroverted Firm: How External Information

Practices Affect Innovation and Productivity," Management Science (58:5), pp. 843-859.

Tata, S., and Patel, J. M. 2007. "Estimating the Selectivity of Tf-Idf Based Cosine Similarity Predicates,"

ACM SIGMOD Record (36:2), pp. 7-12.

Teece, D. J. 1986. "Profiting from Technological Innovation: Implications for Integration, Collaboration,

Licensing and Public Policy," Research Policy (15:6), pp. 285-305.

Tetlock, P. C. 2011. "All the News That's Fit to Reprint: Do Investors React to Stale Information?,"

Review of Financial Studies (24:5), pp. 1481-1512.

Tetlock, P. C., Saar-tsechansky, M., and Macskassy, S. 2008. "More Than Words: Quantifying Language

to Measure Firms' Fundamentals," The Journal of Finance (63:3), pp. 1437-1467.

Thomas, L. A. 1999. "Incumbent Firms' Response to Entry: Price, Advertising, and New Product

Introduction," International Journal of Industrial Organization (17:4), pp. 527-555.

Wheale, P. R., and Amin, L. H. 2003. "Bursting the Dot. Com" Bubble': A Case Study in Investor

Behaviour," Technology Analysis & Strategic Management (15:1), pp. 117-136.

Wiersema, M. F., and Bowen, H. P. 2008. "Corporate Diversification: The Impact of Foreign

Competition, Industry Globalization, and Product Diversification," Strategic Management

Journal (29:2), pp. 115-132.

Wiggins, R. R., and Ruefli, T. W. 2005. "Schumpeter's Ghost: Is Hypercompetition Making the Best of

Times Shorter?," Strategic Management Journal (26:10), pp. 887-911.

Wooldridge, J. M. 2010. Econometric Analysis of Cross Section and Panel Data. Cambridge, MA: MIT

press.

41

Yamawaki, H. 2002. "Price Reactions to New Competition: A Study of Us Luxury Car Market, 1986–

1997," International Journal of Industrial Organization (20:1), pp. 19-39.

42

Figure 1: Number of Unique Words Used in 10-K Product Descriptions (Incumbent)

Figure 2: Number of Startups that Received First-round Funding, and Number of Unique Words

Used in the Collection of Startups’ Product Descriptions

43

Figure 3: New Entry Threats and Ex Post Number of Competitive Rivals for Incumbent (TNIC)

Figure 4: New Entry Threats and Ex Post Operating Performance (Return on Asset) for Incumbent

44

Figure 5: Comparing Turbulent Events between Top 10 Highest-NET Companies

and Top 10 Lowest-NET Companies

Note: Two-sided t-test of group means: Bankruptcy (p=0.329), Layoffs (p=0.000), M&A (p=0.028)

Figure 6a: Average NET Over Time, Military Armored Vehicle Manufacturing

45

Figure 6b: Average NET Over Time, Internet Related Firms and Software Publishers

Figure 6c: Average NET over Time, Biotechnology R&D Services

46

Table 1: Literature on Entry and Incumbent’s Strategic Responses

Paper/Journal Type of Entry

Industrial Setting Strategic variable of

incumbent’s response

Khanna and Tice 2000 /

The Review of Financial

Studies

Diversifying entry Department store Market expansion/retrenchment

Thomas 1999 /

International Journal of

Industrial Organization

Diversifying entry Ready-to-eat cereal industry Pricing, advertising, and

diversification

Yamawaki 2002/

International Journal of

Industrial Organization

Diversifying/Foreign entry Luxury cars Pricing

Simon 2005 / Strategic

Management Journal

Mixed Magazine subscription Pricing

McCann and Vroom 2010 /

Strategic Management

Journal

Mixed Hotel Pricing

Seamans 2013 / Strategic

Management Journal

Not specified Cable TV Pricing

Kuester et al. 1999 /Journal

of Marketing

Not specified Consumer and industrial

goods manufacturers

Product mix and price

Gatignon et al. 1989

/Journal of Marketing

Research

Mixed OTC gynecological product

and airline

Advertising expenditure

Aghion et al. 2009 /The

Review of Economics and

Statistics

Diversifying/Foreign entry Manufacturing Innovation

Koski and Majumdar 2002

/Information Economics

and Policy

Not specified Telecom Pricing, Advertising and

Diversification

Lieberman 1987 / The

RAND Journal of

Economics

Diversifying entry Chemical processing Excess capacity

Goolsbee and Syverson

2008 /The Quarterly

Journal of Economics

Diversifying entry US airline industry Pricing

47

Table 2: Theoretical Perspectives on the Relationship between New Entry Threat and the

Incumbent’s Innovation Response

Theoretical

Perspective

Optimal Strategic

Reaction

Implication for

Incumbent Innovation

Supporting Literature

Pre-emptive market

expansion

Enter into new product

markets to pre-empt entry Increase innovation Eaton and Lipsey 1979;

Gilbert and Newbery 1982;

Schmalensee 1978

Heterogeneous Research

Capability

Underinvest in radically new

and commercially important

inventions, and focus on

“improvement” inventions

Reduce innovation Chandy and Tellis 2000;

Hamberg 1963; Henderson

1993; Mueller and Tilton

1969

Strategic interaction Acquire complementary

assets required for the

commercialization of

innovation; form

interorganizational strategic

alliances with new entrants

Reduce innovation Mitchell 1989; Mitchell and

Singh 1992; Teece 1986

Post-entry defense Cut slack; improve

efficiency; align managerial

incentives

Reduce innovation Gatignon et al. 1989; Hart

1983; Hauser and Shugan

1983; Kuester et al. 1999

Table 3: Top 20 Words with Highest TF-IDF Weights in Entrepreneurial Firms’ Documents in the

Years 2000, 2003, 2006

Year Words List

2000 onlin, internet, softwar, wireless, softwareprovid,

servicesprovid, solut, web, content, broadband, ecommerc,

softwaredevelop, servicesdevelop, platform, media, drug, email,

network, enterpris, patient, portal

2003 diseas, patient, drug, therapeut, cancer, softwar, therapi,

treatment, protein, wireless, discord, video, molecul, inhibitor,

tissu, healthcar, biotechnolog, antibodi, pharmaceut, cell

2006 onlin, diseas, patient, cancer, drug, therapeut, search, publish,

media, video, therapi, cloud, blog, antibodi, tissu, treatment,

game, softwar, advertis, platform

Note: All words are stemmed and pre-processed

48

Table 4: Variable Definitions and Data Sources

Innovation Variables (data source: Compustat & NBER Patent Database)

R&D_intensity𝑖𝑡 Research and Development expenditure to total assets ratio of firm i in year t

Patents𝑖𝑡 Number of patents firm i applied for in year t

WCitations𝑖𝑡 The number of citation-weighted patent applications of firm i in year t

New Entry Threat Variable (data source: VentureXpert & 10K files)

NET𝑖𝑡 A text-based measure of threat from new entry by term frequency–inverse document

frequency weighted cosine similarity between business description of startups and

established firms.

Firm Characteristics (data source: Compustat & TNIC)

Sale𝑖𝑡 Total Sale of firm i in year t (in $ Billion)

Age𝑖𝑡 Number of years since listing of firm i in year t

ROA𝑖𝑡 Operating income before depreciation to total assets ratio of firm i in year t

Asset_tangibility𝑖𝑡 Net property, plants and equipments to total assets ratio of firm i in year t

Leverage𝑖𝑡 Total debt of firm i in year t divided by its total assets

CapExp/Assets𝑖𝑡 Capital expenditure to total assets ratio of firm i in year t

Tobin′s Q𝑖𝑡 Market to book ratio of firm i in year t as defined in Brown and Caylor (2006)

KZ Index𝑖𝑡 Kaplan-Zingales Index (Kaplan and Zingales 1997) is a relative measurement of

reliance on external financing. Companies with higher KZ-Index scores are more

likely to experience difficulties when financial conditions tighten since they may have

difficulty financing their ongoing operations.17

.

TNIC_HHI𝑖𝑡 Herfindahl-Hirschman Index of firm i in year t based on Text-based Network

Industry Classifications (TNIC) (Hoberg et al. 2014).

Industry Level Variables (data source: Compustat & TNIC)

Ind_R&DIntensity𝑖𝑡 Mean of R&D Intensity of all companies in company i’s industry at year t

Ind_SaleGrowthRate𝑖𝑡 Mean of sale growth over last year of all companies in company i’s industry at year t

Ind_Concentration𝑖𝑡 Herfindahl-Hirschman Index of market share by sale in company i’s industry at year t

Ind_ProfitExp𝑖𝑡 Mean of Tobin’s q of all companies in company i’s industry at year t

Ind_Uncertainty𝑖𝑡 Standard deviation of Profit Rate of all companies in company i’s industry at year t

17 Following Chemmanur and Tian (2012), we use the regression coefficients from Kaplan and Zingales (1997) to compute the

KZ index as : 1.002* 39.368* 1.315* +0.28* +3.18*Cash flow Dividends Cash flow Q Leverage

49

Table 5: Summary Statistics and Correlation Coefficients

This table reports the summary statistics for primary variables constructed based on the sample of U.S. public firms in the IT Industries18

from 1997 to 2013.

Please see Table 4 for the description of the variables. Pearson Correlation Coefficients are reported for our sample of 14,410 firm year observations.

Variable

Mean

Std.

dev.

R&D

Intensity NET Sale Age ROA

Asset

Tangibility Leverage

CapExp/

Assets

Tobin

Q

KZ

Index

Correlation Coefficients

R&D Intensity (%) 12.87 15.52

NET 0.07 0.05 0.092*

Sale (Billion) 1.36 7.03 -0.085* 0.040*

Age 15.51 12.37 -0.195* -0.338* 0.254*

ROA (%) -1.26 35.99 -0.645* -0.065* 0.093* 0.178*

Asset Tangibility (%) 13.33 11.91 -0.057* -0.368* 0.113* 0.408* 0.114*

Leverage (%) 9.34 17.21 -0.015 -0.149* 0.057* 0.132* 0.019* 0.341*

CapExp/Assets (%) 3.92 4.27 0.050* -0.048* 0.038* 0.019* -0.017* 0.488* 0.138*

Tobin’s Q 2.48 2.96 0.249* 0.045* -0.031* -0.125* -0.302* -0.072* 0.001 0.052*

KZ Index 0.25 2.69 -0.009 0.009 -0.001 -0.012 0.018* 0.045* 0.166* 0.042* 0.221*

TINC-HHI 0.23 0.21 -0.119* -0.148* -0.065* 0.006 -0.024* -0.137* -0.022* -0.088* -0.031* -0.037*

18 We include all IT Industries, such as hardware, software and IT services industries, which is defined by 4 digital NAICS code: 2211, 3332, 3333, 3336, 3339, 3341, 3342, 3343,

3344, 3345, 3346, 5112, 5161, 5171, 5172, 5173, 5174, 5179, 5112, 5181, 5182, 5413, 5415, 5416 and 5417.

50

Table 6: New Entry Threats and Innovation

This table reports the estimates for R&D Intensity as dependent variables. The sample constructed based on the

sample of U.S. public firms in the IT Industries from 1997 to 2013.

Fixed Effect Random Effect Dynamic panel model

NET as exogenous NET as endogenous

(1) (2) (3) (4)

NET -0.351** -0.074 -0.335* -1.337***

(0.144) (0.114) (0.174) (0.335)

L.R&D Intensity - - 0.119*** 0.122***

(0.040) (0.039)

Ln (Sales) 0.152 0.011 0.863* 1.011**

(0.365) (0.113) (0.515) (0.504)

Ln(Age) 1.079* 0.222 1.291* 0.867

(0.553) (0.302) (0.766) (0.767)

ROA -26.460*** -27.022*** -28.657*** -28.730***

(2.422) (0.294) (2.957) (2.953)

PPE / Assets 16.151*** 9.525*** 14.862*** 14.836***

(2.426) (1.285) (3.321) (3.328)

Leverage -2.049 -2.901*** -3.052 -2.993

(1.324) (0.559) (1.905) (1.908)

Capx / Assets 8.180* 8.071***

13.016* 13.105*

(4.558) (2.422) (7.008) (7.042)

Tobin’s Q 0.178** 0.155*** 0.194 0.194

(0.087) (0.029) (0.121) (0.121)

KZ Index 0.096

0.093*** 0.005 0.003

(0.086) (0.029) (0.057) (0.057)

TNIC HHI 0.087 -1.513*** -0.535 -0.786

(0.530) (0.474) (0.555) (0.564)

Firm Dummies Yes -

Year Dummies Yes Yes Yes Yes

Observations 14,410 14,410 10,490 10,490

# of Firms 2101 2101 1,557 1,557

Adjusted R2 0.761 - -

Note: Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1

New entry threats are standardized with mean of zero and standard deviation of one.

The dynamic model in column 3 treats L.R&D Intensity as endogenous variable. The model in column 4 assumes both NET

and L.R&D Intensity are endogenous variables.

Instruments for differenced equation: GMM-type: L(2/.).NET, L(2/.).L.R&D Intensity, i.e., all available lags

from lag2 onward; and all difference of exogenous variables, including year dummies.

Arellano-Bond test results for zero autocorrelation in the first differences errors

Model with NET as exogenous Model with NET as endogenous

Order z Pr > z Order z Pr > z

1 -3.68 0.000 1 -3.79 0.000

2 -1.85 0.064 2 -1.90 0.058

51

Table 7: 3SLS Simultaneous Equations System Estimation

This table reports the 3SLS estimates for R&D Intensity and NET as endogenous variables. The sample is

constructed based on the sample of U.S. public firms in the IT Industries from 1997 to 2013.

Incumbent Eqn. Entry Eqn.

Dependent Variable: R&D Intensity (%) NET

NET -2.845***

(0.672)

Ln (Sales) -0.120

(0.159)

Ln(Age) 2.802***

(0.379)

ROA -26.263***

(0.302)

PPE / Assets 12.392***

(1.347)

Leverage -1.071**

(0.524)

Capx / Assets 9.105***

(2.442)

Tobin’s Q 0.218***

(0.028)

KZ Index -0.028

(0.025)

TNIC HHI -0.290

(0.526)

L.Ind_R&DIntensity (%) -0.012***

(0.002)

L.Ind_SaleGrowthRate 0.000**

(0.000)

L.Ind_Concentration -0.299***

(0.035)

L.Ind_ProfitExp19

0.199***

(0.009)

L.Ind_Uncentainty 0.017

(0.044)

Firm/Industry Dummies Yes Yes

Year Dummies Yes Yes

R2 0.391 0.044

Observations 13,760

Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1

Note: new entry threats are standardized with mean of zero and standard deviation of one.

19 Tobin's q is often used as a proxy for profit expectation (see e.g., Jankowsky 1998, p34).

52

Table 8: New Entry Threats and Patents

This table reports the estimates for Number of Patents and Weighted Citation as dependent variables. The sample

constructed based on the sample of U.S. public firms in the IT sector and IT product sector from 1997 to 2002.

Dependent

Variable

Panel A

Full IT Sample

Panel B

IT Subsample Excluding Service20

Number of

Patents

Weighted Citations

Number of

Patents

Weighted Citations

(1) (2) (3) (4)

NET -0.019 -0.091*** -0.028** -0.133***

(0.012) (0.030) (0.014) (0.034)

Ln (Sales) 0.003 -0.043 -0.009 -0.073*

(0.014) (0.036) (0.017) (0.041)

Ln(Age) 0.033 0.098 0.017 0.009

(0.044) (0.123) (0.050) (0.138)

ROA 0.007 0.054 0.011 0.077

(0.024) (0.070) (0.027) (0.079)

PPE / Assets 0.102 0.600** 0.067 0.613**

(0.106) (0.285) (0.114) (0.306)

Leverage -0.074 -0.166 -0.056 -0.187

(0.054) (0.133) (0.062) (0.152)

Capx / Assets -0.114 -0.031 -0.030 0.280

(0.169) (0.475) (0.194) (0.537)

Tobin’s Q 0.007** 0.011 0.007* 0.011

(0.003) (0.007) (0.004) (0.008)

KZ Index -0.005*** -0.015*** -0.006*** -0.014**

(0.002) (0.006) (0.002) (0.006)

TNIC HHI 0.034 0.208* -0.022 0.055

(0.044) (0.118) (0.054) (0.141)

Firm FE Yes Yes Yes Yes

Year FE Yes Yes Yes Yes

Observations 7,824 7,824 6,539 6,539

# of Firms 2089 2089 1738 1738

Adjusted R2 0.890 0.733 0.888 0.737

Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1

Note: new entry threats are standardized with mean of zero and standard deviation of one.

20 IT product industries includes IT software, Hardware and Telecom. We exclude IT services industry defined by 4 digital

NAICS code: 5182,5413,5415,5416,and 5417.

2( )tLn Patent 2( )tLn WeightedCitation 2( )tLn Patent 2( )tLn WeightedCitation

53

Table 9: Diversified vs. Specialized Firms: The Effects of NET on Innovation

This table reports the estimates for R&D Intensity as dependent variables. The sample constructed based on the

sample of U.S. public firms in the IT Industries from 1997 to 2013.

Dependent Variables

R&D Intensity

Full Sample Specialized Firms Diversified Firms

(1) (2) (3)

NET -0.821*** -0.466** 0.089

(0.307) (0.223) (0.137)

NET × Diversity 0.573*** - -

(0.221) - -

Ln (Sales) -0.192 0.113 -0.685**

(0.475) (0.647) (0.280)

Ln(Age) 0.946 0.830 0.048

(0.598) (0.983) (0.550)

ROA -27.380*** -27.994*** -22.534***

(3.698) (4.097) (2.831)

PPE / Assets 14.432*** 20.432*** 6.135***

(2.782) (4.405) (1.974)

Leverage -3.353* -2.746 -3.547***

(1.951) (2.928) (0.854)

Capx / Assets 7.763 4.177 11.691***

(6.052) (9.404) (4.160)

Tobin’s Q 0.146 0.091 0.289***

(0.109) (0.127) (0.110)

KZ Index 0.154 0.227 -0.121

(0.129) (0.161) (0.094)

TNIC HHI -0.141 0.240 -0.782*

(0.566) (1.014) (0.428)

Year & Firm Fixed

Effects

Yes Yes Yes

Observations 9,571 5,508 4,063

# of Firms 954 594 360

Adjusted R2 0.736 0.720 0.742

Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1

Note: new entry threats are standardized with mean of zero and standard deviation of one.

54

Appendix 1: NET and Turbulent Events - A Qualitative Examination of Year 2009

We select companies with high and low NET values in 2009 to qualitatively examine the industry

composition of the firms as well as their turbulent events in the following five years. There are two

reasons for why we pick firms in year 2009. First, we try to avoid most recent year in our sample, since

we allow a 5-year period to observe the subsequent turbulent events. Second, selecting year of 2009

allows us to minimize the influence from dot-com bubble in 2000 and finance crisis in 2007-2008. We

summarize the industry sectors (4 digital NACIS) of all companies and present this information in Table

A1. As can be seen from Table A1, the majority of low NET companies belong to manufacturing industry

including pharmaceutical and medicine manufacturing, communications equipment manufacturing, and

computer and peripheral equipment manufacturing. These industries represent particularly stable

ecosystem within the high-tech sector. The probable reasons for this stability include: manufacturing in

general has high capital requirements that are natural barriers to entry, markets for manufacturing in the

US are relatively mature and highly concentrated, standards and protocols within the manufacturing

industry are relatively stable, leading to relatively fewer process and product innovations in 2009. By

contrast, the high NET group includes a majority of software publishers, where firms experience high

margins, high growth rates, access to skilled human capital, and relatively low entry barriers. These

factors make this industry particularly attractive to entrepreneurs, and correspondingly incumbents

experience high new entry threat. Table A1 also shows the turbulent events of these companies in the

following 5 years (2010 – 2014). We observe that the incident rates of bankruptcies, layoffs, and M&A

are much greater for high NET firm.

Appendix 2: Analyses of Cash Holdings

We study how an incumbent firm adjusts its cash holdings in response to NET. We compute a firm’s cash

holding as the natural log of firm cash and cash equivalents, and run a fixed effects panel data model with

using cash holding as the dependent variable (Hoberg et al. 2014). We report the results in Table A2. The

results from Column (1) show that on average, firms withhold more cash in year t as new entry threats

55

increase, consistent with Hoberg et al’s (2014) finding that firms are more financially conservative when

their product markets are volatile. Interestingly, in Column (2), we also find a significant moderating

effect of firm product diversification in that diversified firms tend to reserve less cash holdings in the face

of new entry. In the subsample analyses, presented in Column (3) and (4), the estimates of NET further

confirm that a specialized firm reacts to entry more dramatically so that it holds more cash. 21

21

We also conducted a robustness test where we used cash ratio (which is the ratio of cash holdings over assets) as the dependent

variable, and all the findings are consistent.

56

Table A1: Industry Composition and Turbulent Events of High NET vs. Low NET companies, 2009

Company Name 4 Digit NAICS Industry Sector Events in 2010 - 2014

Companies with Low NET Bankruptcy M&A Layoff

China Botanic Pharmaceutical

Inc.

Pharmaceutical and medicine manufacturing

FirstEnergy Corp. Electric power generation, transmission, and distribution

Intellicheck Mobilisa, Inc. Computer and peripheral equipment manufacturing

Intermec, Inc. Computer and peripheral equipment manufacturing Yes

Plantronics, Inc. Communications equipment manufacturing

The Singing Machine, Inc. Audio and video equipment manufacturing

Strasbaugh, Inc. Industrial machinery manufacturing Yes

Wayside Technology Group, Inc.

Professional and commercial equipment and supplies,

merchant wholesalers

World Surveillance Group Inc. Aerospace product and parts manufacturing

WorldGate Communications, Inc. Communications equipment manufacturing Yes

Companies with High NET

Limelight Networks, Inc Wired telecommunications carriers

Health Grades, Inc. Wired telecommunications carriers Yes

Salesforce.com Inc. Software publishers Yes

Convera Corporation Software publishers Yes Yes

THQ Inc. Software publishers Yes Yes

Take-Two Interactive Software,

Inc.

Software publishers

Electronic Arts, Inc. Software publishers Yes

Vocus Inc. Software publishers Yes Yes

Microsoft Corp. Software publishers Yes

Digital River, Inc.

Professional and commercial equipment and supplies,

merchant wholesalers

Yes

57

Table A2: New Entry Threats and Cash Holdings

This table reports the estimates for Log of Cash and Cash Equivalents as dependent variables. The sample

constructed based on the sample of U.S. public firms in the IT industries from 1997 to 2013.

Dependent Variables

Ln(Cash Holdings)

Full Sample Specialized Firms Diversified Firms

(1) (2) (3) (4)

NET 0.100*** 0.187*** 0.136*** 0.030

(0.011) (0.029) (0.017) (0.019)

NET × Diversity - -0.096*** - -

- (0.025) - -

Ln (Sales) 0.509*** 0.533*** 0.471*** 0.661***

(0.018) (0.021) (0.026) (0.036)

Ln(Age) -0.559*** -0.460*** -0.466*** -0.348***

(0.043) (0.047) (0.067) (0.071)

ROA 0.140*** 0.083 0.150*** -0.323**

(0.037) (0.052) (0.053) (0.162)

PPE / Assets -3.065*** -2.837*** -2.664*** -2.968***

(0.163) (0.203) (0.271) (0.296)

Leverage -0.274*** -0.258*** -0.270*** -0.222*

(0.069) (0.080) (0.105) (0.128)

Capx / Assets 0.343 0.714** 0.792* 0.727

(0.241) (0.338) (0.447) (0.468)

Tobin’s Q 0.002 -0.002 -0.004 0.016*

(0.003) (0.004) (0.005) (0.009)

KZ Index -0.009** -0.011** -0.009* -0.025

(0.004) (0.006) (0.006) (0.017)

TNIC HHI -0.051 -0.045 -0.107 0.029

(0.046) (0.052) (0.074) (0.073)

Year & Firm Fixed Effects Yes Yes Yes Yes

Observations 14,410 9,571 5,508 4,063

# of Firms 2100 954 594 360

Adjusted R2 0.889 0.903 0.885 0.913

Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1

Note: new entry threats are standardized with mean of zero and standard deviation of one.