Technology Conglomeration, Strategic Alliances, and Corporate Innovation

This article was downloaded by: [129.100.58.
76] On: 19 March 2019, At: 20:30

Publisher: Institute for Operations Research and the Management Sciences (INFORMS)
INFORMS is located in Maryland, USA
Management Science
Publication details, including instructions for authors and subscription information:
http://pubsonline.informs.org
Technology Conglomeration, Strategic Alliances, and

Corporate Innovation
Kai Li, Jiaping Qiu, Jin Wang
To cite this article:

Kai Li, Jiaping Qiu, Jin Wang (2019) Technology Conglomeration, Strategic Alliances, and Corporate Innovation. Management
Science
Published online in Articles in Advance 19 Mar 2019
. https://doi.org/10.1287/mnsc.2018.3085
Full terms and conditions of use: https://pubsonline.informs.org/page/terms-and-conditions
This article may be used only for the purposes of research, teaching, and/or private study. Commercial use
or systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisher
approval, unless otherwise noted. For more information, contact [email protected].
The Publisher does not warrant or guarantee the article’s accuracy, completeness, merchantability, fitness
for a particular purpose, or non-infringement. Descriptions of, or references to, products or publications, or
inclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, or
support of claims made of that product, publication, or service.
Copyright © 2019, INFORMS
Please scroll down for article—it is on subsequent pages
INFORMS is the largest professional society in the world for professionals in the fields of operations research, management
science, and analytics.
For more information on INFORMS, its publications, membership, or meetings visit http://www.informs.org
MANAGEMENT SCIENCE
Articles in Advance, pp. 1–26
http://pubsonline.informs.org/journal/mnsc/ ISSN 0025-1909 (print), ISSN 1526-5501 (online)
Technology Conglomeration, Strategic Alliances, and

Corporate Innovation
Kai Li,a Jiaping Qiu,b Jin Wangc
a Sauder School of Business, University of British Columbia, Vancouver, British Columbia V6T 1Z2, Canada; b DeGroote School of Business,
McMaster University, Hamilton, Ontario L8S 4M4, Canada; c Lazaridis School of Business and Economics, Wilfrid Laurier University,
Waterloo, Ontario N2L 3C5, Canada
Contact: [email protected] (KL); [email protected] (JQ); [email protected], http://orcid.org/0000-0001-9317-1260 (JW)
Received: February 6, 2017 Abstract. We examine the organizational choice and innovative activity of technology
Revised: August 7, 2017; January 6, 2018 conglomerates—firms that explore different technology fields with heated inventive activ-
Accepted: March 2, 2018 ity. We develop a measure of firm-to-economy technological proximity to capture the
Published Online in Articles in Advance: extent of a firm’s technology conglomeration. We show that technology conglomerates
March 19, 2019 are more likely to form alliances and that these alliances lead to higher patent output.
https://doi.org/10.1287/mnsc.2018.3085 In terms of underlying mechanisms, we show that after alliance formation, there are sig-
nificant knowledge pooling and cross-fertilization between technology conglomerates and
Copyright: © 2019 INFORMS their alliance partners. Moreover, technology conglomerates produce more patents that
are novel and/or with greater impact. Our findings suggest that both synergy and tol-
erance for failure are important motives for technology conglomerates to use alliances to
accelerate corporate innovation.
History: Accepted by Gustavo Manso, finance.

Funding: The authors acknowledge financial support from the Social Sciences and Humanities
Research Council of Canada (project “Technology Innovations and Firm Boundaries”). The first
author acknowledges financial support from the undergraduate international student research
assistant program at UBC and the Sauder Exploratory Research Grant program.
Supplemental Material: The online appendix is available at https://doi.org/10.1287/mnsc.2018.3085.
Keywords: firm-to-economy technological proximity • technology conglomeration • innovation • synergy • strategic alliances • knowledge pooling
• cross-fertilization • tolerance for failure
1. Introduction Despite an increasing number of firms becoming

Since the late 1990s, many technology firms have sig- technology conglomerates and redrawing their bound-
nificantly expanded their innovative capabilities, facili- aries, little research has addressed the organizational
tated by their access to new technology through acqui- choice of technology conglomerates and its role in cor-
sitions or strategic alliances. High-profile examples in porate innovation. One exception is a recent study by
recent years include Facebook’s purchase of virtual Bena and Li (2014), who show that the overlap in tech-
reality startups Pebbles Interfaces and its alliance with nologies between a pair of innovative firms is a key
Deloitte Digital to develop digital businesses; Google’s driver of their decision to combine. Our paper extends
acquisition of Skybox’s satellite technology and its this line of inquiry and asks the following research
alliance with Intel to accelerate enterprises’ adoption questions: How is the extent to which an innovative
firm is a technology conglomerate measured? Do tech-
of the cloud; and the Alibaba Group’s investment in
nology conglomerates use mergers and acquisitions
various technology firms such as Tango (a messaging
(M&As) as well as strategic alliances to gain access to
app startup) and Kabam (a video game startup) and
new technology? What determines the choice between
its recent alliance with Ford Motors to jointly explore M&As and strategic alliances? Do strategic alliances
mobility services, connectivity, cloud computing, dig- facilitate technology conglomerates’ innovative activi-
ital marketing, and artificial intelligence. Clearly, the ties and, if so, how?1
strategic considerations of these technology conglom- To answer these questions, we propose a novel mea-
erates are different from those of traditional conglom- sure, firm-to-economy technological proximity, to cap-
erates, whose main goal is product diversification. ture the extent of technology conglomeration for inno-
Today’s technology conglomerates do not simply pur- vative firms. Our new construct is a cosine similarity
sue diversification across unrelated technologies, but measure between a firm’s own patent output, mea-
rather explore technology fields with heated inventive sured by the number of patents across different tech-
activity, that could potentially strengthen their core nology classes, and the patent output of all other public
competencies. firms in the economy. Our measure differs from the
1
Li, Qiu, and Wang: Technology Conglomeration, Alliances, and Firm Innovation
2 Management Science, Articles in Advance, pp. 1–26, © 2019 INFORMS
technology diversification measure (e.g., Hirshleifer experiment”—differences across U.S. states requiring
et al. 2014) in that it measures not only the degree of the combined reporting of corporate income (Mazerov
diversification across different technology fields, but 2009, Bodnaruk et al. 2013) to serve as an instrumental
also the extent to which a firm’s innovative activity variable for alliance formation to help establish a causal
overlaps with those in the rest of economy, in par- effect of alliance formation on innovation output. Many
ticular, technology fields with heated inventive activ- U.S. public firms are composed of a “parent” corpora-
ity. In essence, our new measure captures three basic tion and a number of “subsidiary” corporations owned
features of technology conglomerates: (a) producing by the parent. Combined reporting essentially treats
innovations; (b) producing innovations in multiple the parent and its subsidiaries as one corporation for
technology fields; and, more importantly, (c) exploring state income tax purposes, which makes it harder for
technology fields with heated inventive activity. a multistate corporation to evade taxes. As such, com-
With this new measure, we start by examining bined reporting reduces the opportunity costs of form-
whether technology conglomerates form alliances ing alliances to commit assets, and as a result, firms
more frequently. Consistent with our conjecture, we with more of their operations in states with combined
find that technology conglomerates are more likely reporting are more likely to form alliances. Using the
to form alliances. When increasing the measure of treatment regression with our instrumental variable
firm-to-economy technological proximity by one stan- that captures whether a firm is located in a state with
dard deviation, a technology conglomerate’s number combined reporting, we find that after alliance forma-
of alliances formed per year increases by 5.4%. The tion, the innovation output of alliance firms is signifi-
results suggest that strategic alliance is an important cantly improved, especially for technology conglomer-
organizational form for technology conglomerates to ates. In contrast, we find no positive effect from a firm’s
access new technology. product market threats on its postalliance innovation
Technology conglomerates also face product mar- output.
ket competition that might incentivize them to form The positive effect of alliances on the innovation
alliances as well. To shed light on this possibility, we output of technology conglomerates is consistent with
control for product market threats as captured by the three potential motives behind alliance formation:
Hoberg et al. (2014) product market fluidity measure licensing, synergy, and tolerance for failure. Under the
in explaining alliance formation. Consistent with our licensing motive, forming alliances is simply a business
conjecture, we find that both our firm-to-economy tech- model for technology conglomerates to generate profit
nological proximity and their product market fluid- from licensing with no clear prediction on how postal-
ity measures are positively and significantly associ- liance innovation output is related to alliance partners’
ated with the number of alliance deals, suggesting that knowledge base. Under the synergy motive, forming
exploration in technology fields and threats in product alliances is to benefit both partners through knowl-
markets are distinctly different driving forces behind edge pooling and cross-fertilization. Under the toler-
firms redrawing their boundaries. ance for failure motive, forming alliances provides a
We then examine whether technology conglomer- more long-term view and greater tolerance for exper-
ates also engage in M&As more frequently and what imentation and failure than other forms of financing,
determines the choice between alliances and M&As for and in turn induces more exploration in innovation.
technology conglomerates. We first show that technol- Note that these motives are not mutually exclusive; evi-
ogy conglomerates also do more M&A deals, consis- dence in support of one does not necessarily invalidate
tent with anecdotal evidence provided earlier in the the other. With that caveat in mind, we explore the
introduction and findings in Phillips and Zhdanov channels through which alliances by technology con-
(2013) and Bena and Li (2014), which show that M&As glomerates increase innovation output.
play a significant role in facilitating technological inno- We find evidence of knowledge pooling, whereby
vation. We further find that, conditional on either after alliance formation, more patents of a technology
forming alliances or making acquisitions, technolog- conglomerate make backward citations to its partner’s
ical proximity between a firm pair leads the two to prealliance patents or backward citations made by
choose acquisitions over alliances, supporting the the- these patents, suggesting that the increase in technol-
ory of complementarity-driven M&As (Rhodes-Kropf ogy conglomerates’ postalliance patent output indeed
and Robinson 2008, Bena and Li 2014). This analy- comes from them “borrowing” their alliance partners’
sis suggests that technology conglomerates access new knowledge. We also find evidence of cross-fertilization,
technology through either alliances or M&As and that whereby more patents of a technology conglomerate
closer technological proximity between a firm pair is are in the technology classes of its partner’s patents,
more likely to lead to full integration through M&As. suggesting that the increased patent output really
Next, we examine postalliance innovation output comes from learning from each other in an alliance.
for technology conglomerates. We use a “natural Moreover, at the inventor level, we show that after
Management Science, Articles in Advance, pp. 1–26, © 2019 INFORMS 3
alliance formation, inventors whose prealliance patents proverbial next big thing, whereas traditional con-
are in their partners’ technology classes and/or who glomerates aim to gain competitive advantage and
are geographically closer produce more patents than increase profits by diversifying into different (and often
other inventors, and that inventors whose prealliance unrelated) sectors. There is a large finance literature
patents are in their partners’ technology classes pro- on the impact of sector diversification on corporate
duce more patents in these classes than other inven- financing/investment policies and valuation.2 Yet, to
tors, again suggesting knowledge spillovers and cross- our knowledge, this paper is the first to examine the
learning. Overall, our results support the synergy innovative activity of technology conglomerates. To
motive of alliance formation by technology conglom- this end, we develop a new measure, firm-to-economy
erates. Our evidence on knowledge spillover between technological proximity, to capture the extent to which
alliance partners also suggests that licensing is unlikely an innovative firm explores different technology fields
to be the sole motive for technology conglomerates to with heated inventive activity. Given the crucial role of
form alliances. technology in the knowledge-based economy and the
We also find evidence of tolerance for failure, where- rising importance of technology conglomerates, our
by after alliance formation, technology conglomer- new measure is a novel addition to the literature on
ates produce more exploratory and impactful patents. corporate innovation, and could help explain other cor-
Exploratory innovation requires new knowledge and porate policies.
its payoffs take longer to realize and are of higher Second, our paper adds to the literature on strate-
uncertainty (Balsmeier et al. 2017, Gao et al. 2018). We gic alliances. Although the importance of strategic
show that the number of exploratory patents (80% or alliances in facilitating knowledge transfer and techno-
more of these patents’ citations are not citing a firm’s logical innovation has been well recognized (Mowery
existing patents or the citations made by those patents), et al. 1996, Chan et al. 1997, Holmström and Roberts
the number of patents with the top 10% most cita- 1998, Gomes-Casseres et al. 2006, Fulghieri and Sevilir
tions, and the number of patents in new technology 2009), there is little large-sample evidence on the role
of technological factors in the formation of alliances.
classes increase significantly for technology conglom-
Robinson (2008) finds that the risk of alliance activi-
erates after alliance formation. The results are con-
ties outside a firm is greater than the risk of activi-
sistent with the notion that tolerance for failure is
ties conducted inside the firm. Bodnaruk et al. (2013)
an important motive for technology conglomerates to
show that firms with a higher quality of governance are
form alliances, which results in more exploratory and
more likely to form alliances and are also better able
impactful patents. Taken together, our results suggest
to reap their benefits. Chemmanur et al. (2017) exam-
that both synergy and tolerance for failure are impor-
ine the role of common equity blockholders in foster-
tant considerations for technology conglomerates to
ing alliance formation and establish a positive causal
form alliances. effect of alliances on corporate innovation. Our study
Our study is not without its limitations. In particu- sheds light on the sources of value creation in strate-
lar, our analysis is silent on answering the important gic alliances. Our paper extends Coase’s (1937) orig-
question of why some innovative firms become tech- inal insight that organizational forms have important
nology conglomerates while others remain technolog- implications for investment performance by focusing
ically focused. We believe answering this fundamental on the relationship between alliances formed by tech-
question will be an important and fruitful direction for nology conglomerates and corporate innovation. Our
future research. With this limitation in mind, our anal- results suggest that alliances are an effective organi-
ysis takes an important first step toward understand- zational form for firms to succeed in their innovation
ing the organizational choice and innovative activity of efforts, especially for technology conglomerates active
technology conglomerates. in explorations across a wide spectrum of technology
Our paper makes a number of contributions. First, fields.
our paper provides a systematic analysis of the inno- Third and finally, our paper adds to the growing liter-
vative activity of technology conglomerates. “Technol- ature on corporate innovation (see, for example, a recent
ogy conglomerate” is a new phenomenon that has survey by He and Tian 2018). One strand of that lit-
emerged over the last two decades (see, for exam- erature relates firm organizational structure to corpo-
ple, Amazon). A recent article in The New York Times rate innovative activities. Kortum and Lerner (2000) is
notes that acquisitions by technology giants such as one of the first papers documenting a positive associ-
Facebook and Alphabet might herald the return of ation between venture capital (VC) investing and port-
a time when conglomerates were both powerful and folio firms’ patenting rates. Motivated by the work of
popular (Solomon 2014). Technology conglomerates, Manso (2011) arguing the necessity of tolerance for fail-
however, are different from traditional conglomerates. ure to innovate, Tian and Wang (2014) show that ini-
Technology conglomerates focus on acquiring revo- tial public offering (IPO) firms backed by more failure-
lutionary technology that could potentially be the tolerant VCs tend to generate more patents and patents
with more citations. Chemmanur et al. (2014) find that Second, technology conglomerates can be more ef-
corporate VC-backed firms are more innovative than fective and efficient in managing different technolo-
those backed by independent VC because of techno- gies and research labs and in creating synergies from
logical fit between corporate VCs’ parent firms and pooling different knowledge sources. Since technol-
the startup companies backed by them and corporate ogy conglomerates are developing cutting-edge tech-
VCs’ greater tolerance toward failure than indepen- nologies and more versatile than other firms in the
dent VCs. Chan et al. (1997), Branstetter and Sakakibara economy whereby partners with new technologies and
(2002), Sampson (2005), Gomes-Casseres et al. (2006), production capabilities could generate greater alliance
and Peeters and van Pottelsberghe de la Potterie (2006) benefits (Podolny 1994, Stuart 1998), technology con-
show that alliances are associated with improved patent glomerates are more valuable to potential alliance part-
output in a large number of different settings. Outside ners.3 Indeed, Stuart et al. (1999) show that U.S. biotech
firm boundaries, Seru (2014) shows that target firms in startups with more prominent alliance partners go to
diversifying mergers pursue fewer innovative invest- IPO faster and are valued higher at IPO. Technolog-
ments and generate fewer patents or citations than ical conglomerates are also more likely to form mul-
failed targets do. Bena and Li (2014) find that the tech- tiple alliances that could reinforce the synergy bene-
nological synergies between the target and the acquirer fit. For instance, Ahuja (2000) shows that the number
have important implications for both the takeover out- of ties a firm maintains can affect its innovative out-
comes and the future innovative capacity of the merged put positively by providing greater knowledge sharing,
firm. Liu et al. (2016) further point out that acquiring bringing in complementary skills, and taking advan-
tage of scale economies. Lerner et al. (2003) find that
innovation might be an important motive for M&As.
big pharma firms often use alliances with small biotech
Our paper contributes to this strand of corporate inno-
firms to gain access to new drug opportunities. This
vation literature by highlighting the important interac-
“synergy” motive suggests that technology conglom-
tion between technology conglomeration and alliances
erates are more likely to see opportunities and benefit
in enhancing corporate innovative activities.
from forming strategic alliances.
Third, alliances as a network organization provide
2. Hypothesis Development participant firms with organizational flexibility, facili-
Jensen and Meckling (1992) refer to alliances as net- tating experimentation with new ideas and new com-
work organizations. They argue that such an organi- bination of participants in the pursuit of new technolo-
zational form can add value to participant firms by gies (Mody 1993, Chan et al. 1997). Alliances share a
aligning decision authority with decision knowledge. number of similarities to corporate VCs in promoting
In an alliance, such alignment is achieved when each corporate innovation (Chemmanur et al. 2014). In par-
participant has specific decision responsibility allo- ticular, both organizational structures share a long-
cated according to its expertise and business objective. term view and have more tolerance for failure. Manso
The benefits of forming an alliance are especially high (2011) argues that tolerance for failure is an important
for innovative firms, because innovation requires both ingredient to the optimal incentive that motivates inno-
highly specialized knowledge and decision authority vation. Baum et al. (2000) show that alliances allow
allocated to experts equipped with such knowledge. startups to access social, technical, and commercial
competitive resources that normally are only available
Further, a network organization provides participant
to mature and successful firms, and buffer them from
firms with organizational flexibility, facilitating exper-
hazards typically faced by new firms. Tian and Wang
imentation with new ideas and new combinations of
(2014) find that IPO firms backed by more failure-
participants in the pursuit of new technologies (Mody
tolerant VCs are significantly more innovative. This
1993, Chan et al. 1997) and allowing firms to divest
“tolerance for failure” motive suggests that technology
ex post failed investments at relatively low costs. Not
conglomerates, by having greater resources and more
surprisingly, Chan et al. (1997), Robinson and Stuart
diverse sets of innovative activities, are likely to have
(2007), and Robinson (2008) note that alliance forma- a greater tolerance for failure and benefit more from
tion is a common practice for technology firms and strategic alliances to innovate.
works as an effective strategy for firms to grow, net- The previous discussions lead to the following hy-
work, outsource, and commercialize. potheses:
Technology conglomerates can form alliances for a
number of reasons. First, technology conglomerates Hypothesis 1 (H1). Firm-to-economy technological prox-
with large diverse patent portfolios can make money imity (i.e., technology conglomeration) increases a firm’s
by licensing their patents via alliances. This “licens- likelihood of forming alliances.
ing” motive suggests that forming alliances is simply a Hypothesis 2 (H2). Alliances increase innovation output,
business model for technology conglomerates to profit especially for firms with greater firm-to-economy technolog-
from licensing. ical proximity (i.e., technology conglomerates).
In our empirical investigation, we test the afore- by the SDC. We require that (1) the alliance involves
mentioned hypotheses and also try to identify the at least one U.S. public firm as covered by Compus-
underlying mechanism(s) as all three motives sug- tat; and (2) the U.S. public firm involved is not from
gest that technology conglomerates can benefit from the financial sector (Standard Industrial Classification
strategic alliances but for different reasons. The licens- (SIC) 6000-6999). These filters yield 20,136 alliances for
ing motive predicts that alliances will have a positive our estimation period 1995–2007 because we need 10
impact on technology conglomerates’ innovative activ- years of data for our fixed effect estimator.7
ities because alliances increase the demand for their
innovation output, which gives them greater incen- 3.2. Measuring Firm-to-Economy
tive to develop more patents for licensing. Given that Technological Proximity
licensing is a one-way transfer of knowledge from We develop a measure of firm-to-economy technolog-
technology conglomerates to their partners, we would ical proximity to capture the extent of an innovative
not expect significant synergy or greater risk taking firm’s technology conglomeration, that is, its explo-
in innovative activities. The synergy motive also pre- ration of different technology fields with heated inven-
dicts a positive relationship between alliance formation tive activity. Innovative firms are public firms with at
least one awarded patent over the period 1950–2009,
and postalliance innovation output. Such an increase
where 1950 is the starting year of Compustat and 2009
in innovation output comes from knowledge sharing
is the ending year of the KPSS database.8
and synergy creation between conglomerates and their
To construct the variable, we first measure the scope
partners. We would expect to see evidence of knowl-
of innovative activity through patent output of firm i
edge pooling and cross-fertilization in new patents
using the technology vector Si, t (s i, 1, t , . . . , s i, K, t ), and
from technology conglomerates after alliance forma-
the scope of innovative activity through patent out-
tion. The tolerance for failure motive, similar to the two
put of all other public firms in the economy using the
other motives, also predicts that alliances will have a
aggregate technology vector S−i, t (s −i, 1, t , . . . , s−i, K, t ).
positive impact on technology conglomerates’ innova-
The subscript k ∈ (1, K) is the technology class index.9
tive activities. However, such a positive impact comes
The scalar s i, k, t (s−i, k, t ) is the ratio of the number of
from a more long-term view and greater tolerance for
awarded patents to firm i (all other public firms in the
experimentation and failure. We would expect to see economy excluding firm i) in technology class k with
more patents that are novel and/or with greater impact application years from t − 2 to t (application year t) to
after alliance formation. the total number of awarded patents to firm i (all other
In the next section, we describe our sample, define public firms in the economy) applied for over the same
our new measure of firm-to-economy technological three-year period (in the same year t). Our measure is
proximity, and present a sample overview. then computed as
Firm-to-Economy
3. Sample Formation and Variable
Sit S−it
Constructions Technological Proximity , . (1)
3.1. Our Sample kSit k kS−it k
Our alliance sample comes from the Thomson Finan- The higher the value of this cosine measure, the more
cial’s SDC database on Joint Ventures and Strategic intensively an innovative firm is exploring different
Alliances.4 This database has been used in recent stud- technology fields with heated inventive activity, and
ies (e.g., Allen and Phillips 2000, Fee et al. 2006, hence achieving technology conglomeration. We call
Lindsey 2008, Robinson 2008, Boone and Ivanov 2012, our measure firm-to-economy technological proximity.
Bodnaruk et al. 2013) given its comprehensive cover- It differs from the technology diversification measure
age.5 We obtain data on Compustat firms’ patenting (e.g., Miller 2006, Hirshleifer et al. 2014) in that it mea-
activity from the Kogan et al. (2017) Patent and Citation sures not only the degree of diversification across dif-
File (hereafter, the KPSS database).6 ferent technology fields, but also the extent to which
Our sample period starts in 1985, when data cover- a firm’s innovative activity overlaps with that in the
age on alliances started. Our sample period ends in 2007 rest of economy, in particular, technology fields with
because 2009 was the last year in which patenting infor- heated inventive activity. In essence, our new measure
mation from the KPSS database was available. Because captures three basic features of technology conglomer-
of the well-known patent approval lag between appli- ates: (a) producing innovations; (b) producing innova-
cation and award, data coverage of patents in 2007–2009 tions in multiple technology fields; and, more impor-
was poor; hence, we use patent data ending in 2006 tantly, (c) exploring technology fields with heated
(to predict alliances formed in 2007). inventive activity.
Our alliance sample includes all deals in which the Figure 1 plots the time series of our technological
form of the deal is coded as a “strategic alliance” conglomeration measure averaged across innovative
Figure 1. (Color online) Technology Conglomeration Over Time

Firm-to-economy technological proximity
0.10
0.09
0.08
0.07
0.06
0.05
0.04
0.03
0.02
0.01
0
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
Notes. This figure shows the average level of technology conglomeration over the period 1980–2006. The sample includes all innovative U.S.
public firms covered by Compustat. Innovative firms are those that are awarded at least one patent during 1950–2009 by the U.S. Patent and
Trademark Office.
firms. We observe a gradual rise in the level of tech- peaks in the late 1990s and declines after the burst of
nological conglomeration until the burst of the Inter- the Internet bubble in 2000. The evidence that alliances
net bubble, and a gradual decline till 2007, the end of are most active around the Internet bubble is consis-
our sample period. Table 1 lists the top 10 firms hav- tent with the view that technological innovation is a
ing the highest values of firm-to-economy technolog- key driver. Column (2) presents the sample used in
ical proximity in 1980, 1990, and 2000. We find that our multivariate analysis after imposing the require-
over time, technology conglomerates shift from manu- ment that alliance firms have at least one patent over
facturing and resources to IT and computers. Overall, the period 1950–2009. This filter is applied because our
the results suggest that our new measure captures both focus is on alliances related to technological innova-
the heterogeneity and the dynamics in technological tion. Comparing between the two columns, we show
agglomeration. that about three-quarters of alliances involve innova-
tive U.S. public firms.10
3.3. Summary Statistics To capture innovation output as well as the strength
Table 2 presents the temporal distribution of alliances of intellectual property rights, we use patent count,
over our sample period 1995–2007. Column (1) pre- constructed as the year- and technology-class adjusted
sents the sample without imposing the requirement number of patents over the three-year period pre-
that alliance firms have at least one patent in the KPSS ceding the formation of an alliance. Patents grant
database. We show that the number of alliance deals assignees property rights and hence clearly delineate
Table 1. List of the Top 10 Firms with the Highest Values of Firm-to-Economy Technological Proximity in 1980, 1990, and 2000
1980 1990 2000
General Electric Co 0.633 Du Pont (E I) De Nemours 0.666 IBM 0.715

Du Pont (E I) De Nemours 0.631 Dow Chemical 0.614 Intel Corp 0.686
Dow Chemical 0.627 Pharmacia Corp 0.573 LSI Corp 0.670
Union Carbide Corp 0.588 3M Co 0.569 Texas Instruments Inc 0.668
3M Co 0.575 Rohm and Haas Co 0.538 Sun Microsystems Inc 0.602
Shell Oil Co 0.568 Rhone-Poulenc Rorer 0.538 Lucent Technologies Inc 0.584
Akzona 0.558 General Electric Co 0.534 National Semiconductor 0.573
Rhone-Poulenc Rorer 0.557 Grace (W R) & Co 0.530 Agilent Technologies Inc 0.565
Rockwell Automation 0.533 Honeywell International 0.503 Honeywell International 0.562
Grace (W R) and Co 0.530 American Cyanamid Co 0.488 Motorola Solutions Inc 0.545
Table 2. Strategic Alliances Over Time, 1995–2007 mean (median) value of firm-to-economy technolog-
ical proximity is 0.079 (0.029), which is the correla-
Alliances involving at least one public firm tion between the distribution of an average (a median)
All firms Innovative firms firm’s patents across different patent classes and the
Year (1) (3) distribution of the rest of the economy’s patents.
1995 1,887 1,457
The mean (median) value of product market fluid-
1996 1,356 1,037 ity is 0.062 (0.056). The mean (median) value of tech-
1997 1,950 1,445 nology spillovers is $18 million ($6.3 million). The
1998 2,287 1,605 mean (median) patent count in the three-year period
1999 2,641 1,827 preceding alliance formation is 35.5 (1.3). The mean
2000 2,015 1,359
(median) value of patent diversification is 0.329 (0).11
2001 1,292 928
2002 961 707 The remaining firm characteristics are typical of Com-
2003 1,358 1,047 pustat firms.
2004 1,138 864 In panel B, we present pairwise correlation coeffi-
2005 1,216 911 cients. We first show a positive and significant correla-
2006 1,063 791
tion between firm-to-economy technological proximity
2007 972 700
and the number of alliances, as well as that between
Total 20,136 14,678
product market fluidity and the number of alliances.
Notes. This table reports the number of strategic alliances involving It is worth noting that the correlation between firm-to-
U.S. public firms by the year of alliance deal announcement over the
period 1995–2007. In column (1), a deal enters the sample if at least
economy technological proximity and the number of
one of the alliance partner firms or their parent firms is covered by alliances is twice the size of the correlation between
Compustat and alliance firms are not financial firms (SIC 6000-6999). product market fluidity and the number of alliances.
In column (2), we further require that alliance firms be innovative. We then show strong positive correlations between
Innovative firms are firms with at least one awarded patent over the
period 1950–2009 by the USPTO.
firm-to-economy technological proximity and technol-
ogy spillovers, patent count, and patent diversification,
suggesting that we need to control for those three vari-
their contractual rights. Gans et al. (2002) argue that the ables in our multivariate analysis. More general exami-
presence of patents reduces transaction costs associ- nation of the correlation matrix suggests little problem
ated with collaborative arrangements such as alliances. of multicollinearity.12
Frésard et al. (2015) highlight the distinction between
unrealized innovation through R&D and realized inno- 4. Technology Conglomeration and
vation through patents in firms’ decisions to be ver- Alliance Formation
tically integrated. To capture the diversity of a firm’s Our empirical investigation in this section helps an-
patent portfolio, following Hirshleifer et al. (2014), we swer the following questions: Do technology conglom-
use patent diversification, computed as one minus the erates use strategic alliances to gain access to new tech-
Herfindahl index of a firm’s patent portfolio. To cap- nology? What drives the choice between alliances and
ture the competitive threats faced by a firm in its M&As?
product market space, we use the fluidity measure of
Hoberg et al. (2014) based on changes in rival firms’ 4.1. Panel Data Evidence
products relative to the firm’s products. We also con- To test our first hypothesis relating technology con-
sider a measure of technological spillovers developed glomeration to alliance formation, we run the follow-
by Bloom et al. (2013) to capture the opportunity of ing Poisson regression using the panel data sample:
technological spillovers among firms. Detailed variable
# of alliancesi, t
definitions and constructions can be found in Table A.1
in the appendix. α + β 1 Firm-to-Economy Tech Proximityi, t−1
Table 3 presents summary statistics for the panel + β2 Patent Counti, t−1 + β 3 Patent Diversificationi, t−1
data sample that consists of innovative nonfinancial + β4 R&D i, t−1 + β5 Other Firm Characteristicsi, t−1
firms covered by Compustat over the period 1994–2006.
+ Firm FEs + Year FEs + e i, t . (2)
For a firm to be included in the sample, we require that
it have at least four consecutive observations following The dependent variable is the number of alliances a
Bloom et al. (2013). All continuous variables are win- firm joins in year t. The set of firm innovation charac-
sorized at the 1st and 99th percentiles. All dollar values teristics includes firm-to-economy technological prox-
are measured in 2007 dollars. imity, patent count (in logarithms), patent diversifi-
In panel A, we show that on average an innovative cation, and R&D spending. Other firm characteris-
U.S. public firm forms 0.47 alliances every year. The tics that explain alliance formation include firm size
Table 3. Descriptive Statistics for the Panel Data Sample
Panel A: Summary statistics for the panel data sample
Mean S.D. Median
No. of alliances 0.470 2.033 0.000

Firm-to-economy technological proximity 0.079 0.112 0.029
Product market fluidity 0.062 0.033 0.056
Technology spillovers 17,664 25,398 6,292
Patent count 35.521 207.603 1.333
Patent diversification 0.329 0.365 0.000
R&D 0.062 0.113 0.019
Total assets (2007 $ million) 5,798 36,163 364
Leverage 0.228 0.230 0.197
ROA 0.066 0.263 0.117
Cash holdings 0.161 0.200 0.075
Tobin’s Q 2.192 2.313 1.528
Sales growth 0.149 0.522 0.078
Capex 0.054 0.050 0.040
Number of observations 20,301
Panel B: Correlations
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13)
(1) No. of alliances

(2) Firm-to-economy 0.362
technological proximity (0.00)
(3) Product market fluidity 0.142 0.288
(0.00) (0.00)
(4) Log(1 + technology 0.151 0.652 0.156
spillovers) (0.00) (0.00) (0.00)
(5) Log(1 + patent count) 0.341 0.667 0.078 0.701
(0.00) (0.00) (0.00) (0.00)
(6) Patent diversification 0.201 0.642 0.073 0.747 0.835
(0.00) (0.00) (0.00) (0.00) (0.00)
(7) R&D 0.063 0.284 0.377 0.189 0.081 0.102
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
(8) Log(total assets) 0.222 0.209 0.043 0.183 0.434 0.334 −0.381
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
(9) Leverage −0.047 −0.127 −0.006 −0.108 −0.053 −0.065 −0.073 0.138
(0.00) (0.00) (0.49) (0.00) (0.00) (0.00) (0.00) (0.00)
(10) ROA 0.042 −0.050 −0.266 0.004 0.106 0.073 −0.651 0.392 −0.155
(0.00) (0.00) (0.00) (0.54) (0.00) (0.00) (0.00) (0.00) (0.00)
(11) Cash holdings 0.051 0.280 0.379 0.179 0.055 0.071 0.495 −0.296 −0.318 −0.316
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
(12) Tobin’s Q 0.084 0.152 0.220 0.108 0.076 0.058 0.448 −0.204 −0.003 −0.371 0.355
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.67) (0.00) (0.00)
(13) Sales growth 0.020 0.048 0.139 0.040 −0.006 0.003 0.113 −0.037 −0.031 −0.010 0.105 0.153
(0.00) (0.00) (0.00) (0.00) (0.41) (0.70) (0.00) (0.00) (0.00) (0.16) (0.00) (0.00)
(14) Capex 0.046 0.036 0.025 0.011 0.035 0.028 −0.034 0.094 0.022 0.092 −0.151 0.042 0.047
(0.00) (0.00) (0.00) (0.13) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
Notes. This table reports descriptive statistics for the panel data sample. The panel data sample includes nonfinancial innovative firms covered
by Compustat over the period 1994–2006. Panel A provides summary statistics. Panel B provides Pearson correlations (with p-values in
parentheses). Firm characteristics are measured at the fiscal year end before the alliance announcement year. Definitions of the variables are
provided in Table A.1 in the appendix.
(the logarithm of total assets), leverage, ROA, cash likely to have permanently higher numbers of alliances
holdings, Tobin’s Q, sales growth, and capital expen- than others because of omitted firm-specific effects.
ditures (e.g., Gomes-Casseres et al. 2006, Boone and One interpretation of the firm fixed effect in Equa-
Ivanov 2012, Bodnaruk et al. 2013). Even after con- tion (2) is that it reflects the “entry level of alliance
trolling for known firm characteristics, some firms are formation” of each firm, that is, the number of past
Table 4. Forming Alliances: The Panel Data Sample
(1) (2) (3) (4) (5) (6)
Firm-to-economy 2.641∗∗∗ 1.150∗∗∗ 0.891∗∗∗ 0.848∗∗∗ 0.818∗∗∗

technological proximity (0.407) (0.341) (0.326) (0.328) (0.076)
Product market fluidity 6.517∗∗∗ 6.436∗∗∗
(1.214) (1.227)
Log(1 + technology spillovers) 0.012
(0.012)
Log(1 + patent count) 0.344∗∗∗ 0.213∗∗∗ 0.116∗∗∗ 0.135∗∗∗ 0.132∗∗∗ 0.085∗∗∗
(0.069) (0.045) (0.037) (0.039) (0.039) (0.010)
Patent diversification −0.580∗∗ −0.707∗∗∗ −0.066 −0.095 −0.168 −0.298∗∗∗
(0.271) (0.210) (0.141) (0.153) (0.153) (0.036)
R&D 3.416∗∗∗ 2.909∗∗∗ 3.162∗∗∗ 1.563∗∗∗ 1.548∗∗∗ 0.392∗∗∗
(0.396) (0.380) (0.341) (0.367) (0.366) (0.037)
Log(total assets) 0.418∗∗∗ 0.395∗∗∗ 0.388∗∗∗ 0.337∗∗∗ 0.340∗∗∗ 0.040∗∗∗
(0.031) (0.029) (0.028) (0.028) (0.027) (0.003)
Leverage −1.425∗∗∗ −1.161∗∗∗ −1.261∗∗∗ −1.041∗∗∗ −1.051∗∗∗ −0.061∗∗∗
(0.297) (0.266) (0.241) (0.214) (0.213) (0.011)
ROA 0.378∗ 0.372∗ 0.565∗∗∗ 0.203 0.202 0.082∗∗∗
(0.211) (0.200) (0.179) (0.189) (0.189) (0.015)
Cash holdings 1.309∗∗∗ 0.858∗∗∗ 1.039∗∗∗ 0.733∗∗∗ 0.720∗∗∗ 0.191∗∗∗
(0.181) (0.190) (0.178) (0.196) (0.196) (0.018)
Tobin’s Q 0.056∗∗∗ 0.063∗∗∗ 0.069∗∗∗ 0.076∗∗∗ 0.076∗∗∗ 0.018∗∗∗
(0.012) (0.011) (0.010) (0.007) (0.007) (0.002)
Sales growth 0.050∗ 0.052∗ 0.053∗ −0.001 −0.002 0.023∗∗∗
(0.030) (0.029) (0.028) (0.035) (0.035) (0.003)
Capex 1.501∗∗ 0.424 1.290∗∗ 0.994 1.029 0.129∗∗
(0.725) (0.771) (0.621) (0.651) (0.650) (0.051)
Intercept −4.162∗∗∗ −3.893∗∗∗ −3.922∗∗∗ −3.897∗∗∗ −3.950∗∗∗ −0.315∗∗∗
(0.198) (0.202) (0.188) (0.189) (0.192) (0.017)
Firm fixed effects No No Yes Yes Yes No
Year fixed effects Yes Yes Yes Yes Yes Yes
Number of observations 20,301 20,301 20,301 16,252 16,252 54,146
Notes. This table examines the determinants of alliance formation. Columns (1)–(5) use the Poisson regression model in Equation (2) where
the dependent variable is the number of alliance deals. The panel data sample includes nonfinancial innovative firms covered by Compustat
over the period 1995–2007. Column (6) uses the OLS regression model where the dependent variable is the logarithm of one plus the number
of alliance deals. The panel data sample includes nonfinancial innovative firms covered by Compustat over the period 1985–2007. Firm
characteristics are measured at the fiscal year end before the alliance announcement year. Definitions of the variables are provided in Table A.1
in the appendix. Robust standard errors that allow for clustering at the firm level are reported in parentheses.
∗ ∗∗
, , and ∗∗∗ denote significance at the 10%, 5%, and 1% level, respectively.
alliances at the beginning of the panel data sample. fixed effects. We find that technology conglomerates
We control for firm fixed effects following the “pre- form more alliances, consistent with our first Hypothe-
sample mean scaling” method of Blundell et al. (1999), sis (H1). In terms of economic significance, column (3)
Aghion et al. (2013), and Bloom et al. (2013) specific to shows that when the measure of firm-to-economy tech-
regressions involving count data.13 Table 4 presents the nological proximity increases by one standard devia-
results. tion, the number of alliances formed per year increases
In column (1) we first show that innovative firms by 5.4%; and when the amount of R&D spending
with more patents are more likely to form alliances, increases by one standard deviation, the number of
consistent with Gans et al. (2002), who argue that the alliances formed per year increases by 15%. The effects
strength of intellectual property rights is important of these technological factors on alliance formation are
for alliances. We further show that firms with more economically significant. Column (3) also shows that
R&D spending are more likely to form alliances, con- large firms with low leverage, high ROA, high cash
sistent with Robinson and Stuart (2007), who find that holdings, high Tobin’s Q, fast sales growth, and high
alliances tend to cluster in risky, high-R&D settings. capital expenditures form more alliances.
In column (2) we include firm-to-economy techno- A technology conglomerate is also likely to face
logical proximity, and in column (3) we include both product market competition; such competitive threats
firm-to-economy technological proximity and firm might incentivize it to form alliances as well. To shed
light on this possibility, we include Hoberg et al. (2014) firm; (3) its total assets in year t − 1 is between 50%
product market fluidity measure and report the results and 150% of the alliance firm’s total assets; and (4) it is
in column (4). The coefficient on product market flu- not an alliance firm in the seven-year period from year
idity is significant and positive, suggesting that some t − 3 to year t + 3. As noted by Bena and Li (2014), such
alliances are formed to address a firm’s competitive matching creates a pool of potential alliance firms that
threats in product markets.14 It is worth noting that captures clustering not only in time, but also by indus-
including product market fluidity has no big impact try. Further, industry-matching controls for product
on the magnitude and significance of the coefficient market competition and size-matching partially con-
on firm-to-economy technological proximity, support- trols for the life cycles of corporate innovation. In addi-
ing the notion that exploration in technology fields and tion to using industry matching and size matching to
threats in product markets are distinctly different driv- obtain control firms for alliance firms, we also obtain
ing forces behind firms redrawing their boundaries. control firms for each alliance firm by randomly draw-
As discussed in the hypothesis development sec- ing five firms that are not alliance firms in the seven-
tion, technology conglomerates are more versatile in year period centered at the alliance deal announcement
different technologies than other firms in the econ- year t.
omy, and hence present opportunities for knowledge Table 5 presents summary statistics for alliance
spillovers. To shed light on this possibility, in col- firms and their control firms. We find that alliance
umn (5) we further control for Bloom et al. (2013) firms are in significantly closer technological proxim-
knowledge spillovers in explaining alliance formation. ity to all other firms in the economy, face significantly
We find that there is no significant association between greater product market threats, and have significantly
knowledge spillovers and alliance formation. After larger technology spillovers than their control firms. In
controlling for knowledge spillovers, we find that the terms of innovative activities, alliance firms have more
positive effect of firm-to-economy technological prox- patents, greater patent diversification, and higher R&D
imity on alliance formation remains significant with spending compared to their partners. Finally, alliance
a small change in the magnitude of coefficient (from firms are larger, have lower leverage, are less prof-
0.891 to 0.848), suggesting that the effect of opportu- itable, and have higher cash holdings, higher Tobin’s Q,
nities for knowledge spillovers (as measured by the faster sales growth, and higher capital expenditures
Bloom et al. (2013) measure) absorb little explanatory than their control firms. The evidence strongly sup-
power of firm-to-economy technological proximity. ports the view that technological factors are important
Finally, as a robustness check on the Poisson regres- considerations for firms to form alliances.
sion model with count data and firm fixed effects, in We run the following conditional logit regression
column (6) we employ a simple OLS regression with- using a cross-sectional sample of event and control
out firm fixed effects where the dependent variable is firms:15
the logarithm of one plus the number of alliance deals,
and firm-year observations over the presample period Event Firmim, t
1985–1994 are added back to the panel. It is worth not- α + β 1 Firm-to-Economy Tech Proximityim, t−1
ing that our main findings remain unchanged.
+ β 2 Patent Countim, t−1 + β 3 Patent Diversificationim, t−1
4.2. Matching Firm Evidence + β 4 R&Dim, t−1 + β 5 Other Firm Characteristicsim, t−1
The advantage of using the panel data sample is that it + Deal FEm + e im, t . (3)
provides large sample evidence on the importance of
technology conglomeration in alliance formation. The The dependent variable, Event Firmim, t , takes the value
downside is that many firms included in the analysis of one if firm i is in an alliance deal m, and zero other-
are totally different from alliance participants, which wise. For each deal m, there is one observation for the
makes rejecting our hypothesis less likely. alliance firm, and multiple observations for the control
To further examine whether and how technology firms. Deal FEm is the fixed effect for each alliance firm
conglomeration affects a firm’s likelihood of joining an and its control firms in deal m. Table 6 presents the
alliance, we follow Bena and Li’s (2014) methodology results.
by employing matched control firms for alliance firms. Columns (1)–(2) report the results when we use
For each firm in an alliance deal announced in year t, industry- and size-matched control firms. We find that
we find up to five control firms matched by industry a firm’s technological proximity to other firms in the
and size. We move up to the one-digit SIC industry economy is positively associated with the likelihood of
if we cannot find three control firms in the two-digit that firm joining an alliance (column (1)). Similarly, a
SIC industry. We require a control firm to satisfy the firm’s product market threat is also positively associ-
following conditions: (1) it is an innovative firm; (2) ated with its likelihood of joining an alliance. We fur-
it shares at least one-digit SIC code with the alliance ther find that firms with more patents and high R&D
Table 5. Descriptive Statistics for the Matching Firm Sample
Alliance firms Control firms
Mean S.D. Median Mean S.D. Median t-test Wilcoxon
Firm-to-economy technological proximity 0.147 0.131 0.133 0.057 0.087 0.015 (0.00) (0.00)
Product fluidity 0.092 0.035 0.091 0.066 0.035 0.062 (0.00) (0.00)
Technology spillovers 33,761 32,853 28,238 13,841 22,786 3,047 (0.00) (0.00)
Patent count 30.398 87.281 3.167 6.111 21.811 0.500 (0.00) (0.00)
Patent diversification 0.426 0.355 0.500 0.239 0.325 0.000 (0.00) (0.00)
R&D 0.139 0.147 0.104 0.053 0.099 0.014 (0.00) (0.00)
Total assets (2004 $ million) 2,288 6,789 327 1,680 5,399 266 (0.00) (0.00)
Leverage 0.146 0.199 0.065 0.220 0.229 0.169 (0.00) (0.00)
ROA −0.003 0.308 0.083 0.083 0.211 0.122 (0.00) (0.00)
Cash holdings 0.358 0.266 0.329 0.191 0.233 0.086 (0.00) (0.00)
Tobin’s Q 4.171 4.912 2.559 2.163 1.887 1.545 (0.00) (0.00)
Sales growth 0.543 1.151 0.169 0.226 0.664 0.100 (0.00) (0.00)
Capex 0.056 0.050 0.042 0.052 0.050 0.037 (0.00) (0.00)
Number of observations 9,549 43,385
Notes. This table reports descriptive statistics for the matching firm sample over the period 1995–2007. The sample contains innovative alliance
firms and their control firms. For each alliance firm, we identify up to five control firms that satisfy the following conditions: (1) they are
innovative; (2) they share at least one-digit SIC code with the alliance firm; (3) their total assets in year t falls between 50% and 150% of the
alliance firm’s total assets; and (4) they are not part of an alliance in the seven-year period from year t − 3 to year t + 3 centered at the alliance
deal announcement year t. Definitions of the variables are provided in Table A.1 in the appendix. The p-values (in parentheses) of the t-test
and Wilcoxon test are reported in the last two columns.
spending are more likely to join alliances, whereas on M&A deals from the SDC Mergers and Acquisitions
firms with greater patent diversification are less likely database.16 Table 7 presents the results.
to do so. On the one hand, firms with more diver- Panel A presents the panel data regression results
sified patent portfolios might have a greater need to based on the specification in Equation (2) where the
pool resources with other firms, as they are engaged in dependent variable is the number of M&A deals. We
many fronts, and are likely to have more synergies with first show that technology conglomerates enter into
other firms. On the other hand, firms with more diver- more M&A deals than other firms. These firms’ explo-
sified patent portfolios might have a reduced need to ration in various technology fields with heated inven-
form alliances in order to focus on developing in-house tive activity prompts them to engage in M&As to gain
technologies. Our findings are consistent with the lat- access to new technology, much as alliance forma-
ter interpretation. tion does. We then show that innovative firms with
Columns (3)–(4) report the results when we use low R&D spending do more deals, consistent with
randomly drawn control firms. We again find that a Bena and Li (2014), who find that acquirers tend to
firm’s proximity to other firms in technological devel- spend less on R&D. Moreover, we find that large firms
opment is positively associated with the likelihood of with low leverage, high ROA, low cash holdings, high
that firm joining an alliance, as is its proximity to other Tobin’s Q, fast sales growth, and low capital expen-
firms in product development. ditures do more deals. In terms of economic signifi-
In summary, the firm-level cross-sectional evidence cance, when the measure of firm-to-economy techno-
in Table 6 is largely consistent with the panel data logical proximity increases by one standard deviation,
evidence and provides strong support for our first the number of M&A deals per year increases by 10.8%;
Hypothesis (H1) that firm-to-economy technological when the amount of R&D spending decreases by one
proximity, that is, technology conglomeration, is an standard deviation, the number of M&A deals per year
important factor in firms’ decisions to form alliances. increases by 8.6%. The effects of these technological
factors on M&As are economically significant.
4.3. Alliances versus M&As Panel B examines, conditional on changing the
Given the beneficial role of alliances in helping firms boundaries of the firm, the choice between alliance and
gain access to new technology, we naturally wonder M&A partners using a logit model where the depen-
why more alliances are not observed in practice (see the dent variable takes the value of one if it is an alliance
discussion in Section 2). In this subsection, we examine deal, and zero otherwise (i.e., if it is an M&A deal).17
the relationship between technology conglomeration In this analysis of pair formation (i.e., alliance pairs
and another popular organizational form—M&As— versus M&A pairs), we include three pairwise mea-
and answer the following question: What drives the sures. Firm-to-Firm Technological Proximity from Jaffe
choice between alliances and M&As? We obtain data (1986) measures the correlation of partner firms’ patent
Table 6. Forming Alliances: The Matching Firm Sample
Alliance firm 1, control firm 0
Industry match and size match Random match
(1) (2) (3) (4)
Firm-to-economy 4.000∗∗∗ 3.372∗∗∗ 4.967∗∗∗ 4.415∗∗∗

technological proximity (0.176) (0.234) (0.176) (0.224)
(0.654) (0.595)
Log(1 + technology spillovers) −0.004 −0.012∗
(0.006) (0.006)
Log(1 + patent count) 0.340∗∗∗ 0.378∗∗∗ 0.321∗∗∗ 0.369∗∗∗
(0.019) (0.023) (0.016) (0.020)
Patent diversification −0.421∗∗∗ −0.273∗∗∗ −0.851∗∗∗ −0.808∗∗∗
(0.073) (0.091) (0.066) (0.085)
R&D 4.987∗∗∗ 3.475∗∗∗ 4.810∗∗∗ 3.248∗∗∗
(0.202) (0.235) (0.157) (0.200)
Log(total assets) 0.557∗∗∗ 0.632∗∗∗ 0.642∗∗∗ 0.583∗∗∗
(0.026) (0.036) (0.010) (0.013)
Leverage −0.519∗∗∗ −0.569∗∗∗ −0.926∗∗∗ −0.877∗∗∗
(0.083) (0.097) (0.088) (0.106)
ROA −0.016 −0.054 0.373∗∗∗ 0.347∗∗∗
(0.095) (0.106) (0.066) (0.083)
Cash holdings 1.434∗∗∗ 0.670∗∗∗ 1.912∗∗∗ 0.986∗∗∗
(0.082) (0.096) (0.074) (0.093)
Tobin’s Q 0.123∗∗∗ 0.115∗∗∗ 0.119∗∗∗ 0.118∗∗∗
(0.007) (0.007) (0.005) (0.006)
Sales growth 0.214∗∗∗ 0.099∗∗∗ 0.215∗∗∗ 0.106∗∗∗
(0.018) (0.020) (0.017) (0.021)
Capex 1.123∗∗∗ 1.506∗∗∗ 0.743∗∗∗ 0.412
(0.324) (0.353) (0.287) (0.330)
Deal fixed effects Yes Yes Yes Yes
Number of observations 52,934 35,032 104,148 104,148
Pseudo R-squared 0.34 0.35 0.60 0.59
Notes. This table examines the determinants of alliance formation using the conditional logit model
in Equation (3). The dependent variable is an indicator variable that takes the value of one if a firm
is in an alliance, and zero otherwise. Columns (1) and (2) use the industry- and size-matched control
firms. Column (3) and (4) use randomly drawn control firms. In columns (3) and (4), we repeat the
regression in Equation (3) 500 times by drawing with replacement at the deal cluster level to obtain
standard errors of the coefficient estimates. Definitions of the variables are provided in Table A.1 in the
appendix. Robust standard errors that allow for clustering at the deal level are reported in parentheses.
∗ ∗∗
portfolios. Same Industry is an indicator variable that change their boundaries (through either alliances or
takes the value of one if partner firms operate in the M&As) to gain access to new technology. The choice
same industry, and zero otherwise. Same State is an between full integration (via M&As) and alliances,
indicator variable that takes the value of one if part- however, depends on the degree of pairwise technolog-
ner firms are headquartered in the same state, and ical proximity between partner firms, with closer pairs
zero otherwise. Since this analysis requires informa- in technological development choosing to do M&As.
tion on both partners in a deal, we restrict the sample The results are consistent with the property rights the-
to include only bilateral (alliance or M&A) deals in ory of the firm (Grossman and Hart 1986, Hart and
which both partners are U.S. public firms. The coeffi- Moore 1990, Hart 1995), which posits that significantly
cient on Firm-to-Firm Technological Proximity is negative complementary assets should be placed under com-
and significant at the 1% level, indicating that pairs in mon ownership. Panel B provides evidence in support
M&As have closer technological proximity than pairs of the theory and also explains why we do not observe
in alliances. The finding is interesting in its own right, more alliances in practice because the optimal organi-
as it suggests that exploration in different technology zational form depends on the intensity of firm-to-firm
fields with heated inventive activity prompts firms to technological proximity.
Table 7. Choosing Between Alliances and M&As 5. Technology Conglomerates, Alliances,

Panel A: Determinants of doing M&A deals
and Postalliance Innovation Output
In this section we answer the following question: Do
# M&A deals strategic alliances facilitate technology conglomerates’
Firm-to-economy technological proximity 1.342∗∗∗
innovative activities and, if so, how? Because of the
(0.282) patent approval lag noted earlier and the require-
Log(1 + patent count) 0.082∗∗∗ ment for measuring postalliance innovation output
(0.026) over a three-year period, for this investigation we limit
Patent diversification −0.299∗∗ alliances to those formed before 2004.
(0.124)
R&D −1.066∗∗ 5.1. The Instrumental Variable
(0.452) According to our second Hypothesis (H2), firm-to-
Log(total assets) 0.158∗∗∗ economy technological proximity strengthens the pos-
(0.015) itive effect of alliances on postalliance patent output.
Leverage −1.000∗∗∗ However, it is also possible that firms expecting to
(0.171)
improve patent output may choose to join alliances.
ROA 1.738∗∗∗ For identification, we employ the treatment regression
(0.227)
framework with an instrumental variable that clearly
Cash holdings −0.427∗∗
(0.174)
affects alliance decisions but has nothing to do with
Tobin’s Q 0.053∗∗∗
patent output other than through the channel of join-
(0.012) ing alliances.18
Sales growth 0.206∗∗∗ For this purpose, we use a “natural experiment”—
(0.035) differences across U.S. states requiring the combined
Capex −3.905∗∗∗ reporting of corporate income (Mazerov 2009) to help
(0.585) identify the direction of causality (Bodnaruk et al.
Intercept −1.994∗∗∗ 2013). We consider situations where the opportunity
(0.121) costs of forming alliances differ because of exogenous
Firm fixed effects Yes reasons that are not firm specific, and examine how the
Year fixed effects Yes
differential reaction to this variation is related to patent
Number of observations 20,301
output.
Panel B: Determinants of choosing between Many U.S. public firms are composed of a “par-
alliance and M&A partners ent” corporation and a number of “subsidiary” cor-
Alliance 1, M&A 0 porations owned by the parent. Combined reporting
essentially treats the parent and its subsidiaries as
Firm-to-firm technological proximity −1.062∗∗∗
one corporation for state income tax purposes. Their
(0.241)
nationwide profits are combined and the state then
Same industry −1.006∗∗∗
(0.131)
taxes a share of that combined income. The share is cal-
Same state −0.667∗∗∗
culated by a formula—a combination of property, pay-
(0.139) roll, and/or sales factors—that takes into account the
Other firm characteristics Yes corporate group’s level of activity in the state as com-
Number of observations 2,346 pared to its activity in other states (Huddleston and
Pseudo R-squared 0.25 Sicilian 2008, Mazerov 2009). According to Huddleston
Notes. This table examines the determinants of doing M&As and the and Sicilian (2008), combined reporting is particularly
choice between alliances and M&As. Panel A examines the deter- effective in eliminating income-shifting transactions
minants of doing M&As using the Poisson regression model in within corporate groups, thereby reducing the oppor-
Equation (2) where the dependent variable is the number of M&A
tunity cost of “ring-fencing” assets via alliances.
deals. The panel data sample includes nonfinancial innovative firms
covered by Compustat over the period 1995–2007. Robust standard Sixteen states mandated combined reporting before
errors that allow for clustering at the firm level are reported in paren- 1985. In 2004, Vermont became the first state in more
theses. Panel B examines the formation of an alliance pair versus than 20 years to adopt combined reporting, effective
an M&A pair using a logit model where the dependent variable is
an indicator variable that takes the value of one if the firm pair is
in 2006 (see Figure 1, Mazerov 2009). Given that our
in an alliance deal, and zero if the firm pair is in an M&A deal. sample period is from 1995–2004, we rely on only the
The sample includes bilateral alliance and M&A deals announced cross-sectional variation in states’ adoption of com-
over the period 1995–2007 and in which both parties are U.S. public bined reporting as our instrumental variable.
firms. Definitions of the variables are provided in Table A.1 in the
appendix.
Our identification assumption is that combined re-
∗ ∗∗
, , and ∗∗∗ denote significance at the 10%, 5%, and 1% level, porting reduces the opportunity costs of forming al-
respectively. liances to commit assets; we expect that firms with
more of their operations in states with combined re- patent output by all inventors in a given state. Panel B
porting will be more likely to form alliances. presents the results. We do not find any significant dif-
Our instrumental variable is a firm’s combined references in R&D spending or patent output between
porting index, with a higher value of the index indi- entities in states without combined reporting and those
cating that more of the firm’s operations are located in states with combined reporting. The findings help
in states that require combined reporting. To con- mitigate the concern that states with combined report-
struct the index, we need location information on the ing might have more innovation opportunities.
firm’s subsidiaries (including its headquarters) and Second, it is worth noting that a gap of approxi-
whether the state of each of its subsidiaries requires mately 10 years exists between states adopting com-
combined reporting or not. The index is the num- bined reporting and our sample period. State-level
ber of its subsidiaries in states with combined report- innovation opportunities available more than 10 years
ing scaled by the total number of its subsidiaries. ago would arguably have no effect on firms’ alliance
Bodnaruk et al. (2013, Appendix D) provide detailed decisions today.
information on the construction of the index. We have Third and finally, the argument that a firm may chase
data on the firm-level combined reporting index for innovation opportunities and relocate its operations
1998, 2000, 2002, and 2004. We use the 1998 data for is based on the assumption that the benefits of mov-
alliances formed before 1998; the 2000 data for alliances ing to locations with more innovation opportunities
formed between 1999–2000; the 2002 data for alliances are greater than the costs of moving to those loca-
formed between 2001–2002; and finally the 2004 data tions from otherwise optimal location choices. Given
for alliances formed between 2003–2004. that firms with multiple subsidiaries are likely to face
One potential concern about identification is that greater relocation costs (noting that our instrumental
firm location is endogenous. In particular, states with variable is based on the locations of all subsidiaries),
different innovation opportunities may also differ the concern about endogenous location choices should
in combined-corporate-income-reporting regulations. be mitigated for firms with multiple locations of opera-
We address this concern in several ways. tions. Table IA3 in the online appendix shows that our
First, we compare annual state-level R&D spend- results on alliance firms’ postalliance innovation out-
ing (state-level R&D scaled by state-level total assets) put are qualitatively similar if only firms with multiple
and patent output (state-level patent count scaled by locations of operations are included in the analysis.
state-level total assets) between the 16 states with com-
bined reporting and those without over the 10-year 5.2. The Treatment Regression
period ending in 1985. Table IA2 in the online appendix To test our second hypothesis, we run the following
presents results from this analysis. In panel A, we treatment regression:
limit the comparison to Compustat firms, given that Log(1 + Patent Countit+1 )
our main analyses are based on public firms. We do
not find any significant differences in patent output α + β 1 Samplei × Firm-to-Economy Tech Proximityit
between these two groups of states. Moreover, we find + β 2 Samplei + β3 Firm-to-Economy Tech Proximityit
that average R&D spending is significantly higher in + β 4 Innovation Characteristicsit
states without combined reporting than in states with + β 5 Other Firm Characteristicsit + Year FEs + e it . (4)
combined reporting, while there is no significant dif-
ference in median values. Both findings help mitigate The dependent variable is the logarithm of one plus
the concern that states with combined reporting might the alliance firm’s patent count from year t + 1 to year
have more innovation opportunities as we do not find t + 3. Samplei takes the value of one if firm i is an
any support for it.19 Given that state policies, includ- alliance deal, and zero otherwise. Samplei × Firm-to-
ing combined reporting, could be influenced by any Economy Technological Proximityit captures the differen-
entities headquartered in that state, for a more thor- tial effect of firm-to-economy technological proximity
ough investigation of the validity of our instrumental on postalliance patent output of alliance firms vis-à-vis
variable we implement the same test using the state- their control firms. The cross-sectional sample consists
level patent output as well as R&D expenditures by all of alliance firms and their industry- and size-matched
entities in that state that engage in innovative activities control firms. All explanatory variables are measured
(i.e., including public and private firms, as well as not- at the fiscal year end before the alliance deal announce-
for-profit organizations). To do so, we obtain data on ment. Table 8 presents the results.
state-level R&D expenditures from the National Cen- Panel A presents results from the first-stage instru-
ter for Science and Engineering Statistics that has been mental variable regressions where the dependent vari-
available since 1987. We further obtain data on state- able is an indicator variable that takes the value of one
level patent output from the Harvard Business School if a firm is in an alliance, and zero otherwise. The vari-
U.S. Patent Inventor Database (Li et al. 2014) that covers able of interest is a firm’s combined reporting index.
Table 8. Instrumental Variable Regressions
Panel A: The first-stage regressions
(1) (2) (3)
Combined reporting index 0.171∗∗∗ 0.169∗∗∗ 0.156∗∗∗

(0.019) (0.025) (0.025)
Firm-to-economy technological proximity 2.073∗∗∗ 1.714∗∗∗ 1.846∗∗∗
(0.103) (0.133) (0.150)
(0.407) (0.407)
Log(1 + technology spillovers) −0.007∗
(0.004)
Log(1 + patent count) 0.210∗∗∗ 0.259∗∗∗ 0.266∗∗∗
(0.012) (0.015) (0.016)
Patent diversification −0.489∗∗∗ −0.494∗∗∗ −0.481∗∗∗
(0.046) (0.059) (0.060)
R&D 2.193∗∗∗ 1.182∗∗∗ 1.186∗∗∗
(0.096) (0.118) (0.118)
Other firm characteristics Yes Yes Yes
Year fixed effects Yes Yes Yes
Number of observations 39,075 23,263 23,263
F-test for weak instrument (p-value) (0.00) (0.00) (0.00)
Panel B: The second-stage regressions
Log(1 + patent countt+1, t+3 )
(1) (2) (3)

∗∗∗ ∗∗∗
Sample × Firm-to-economy technological proximity 1.907 1.496 0.379∗∗
(0.107) (0.137) (0.181)
Sample × Product market fluidity −3.736∗∗∗ −2.690∗∗∗
(0.461) (0.451)
Sample × Log(1 + technology spillovers) −0.000
(0.005)
Sample 0.120∗∗∗ 0.399∗∗∗ 0.335∗∗∗
(0.034) (0.058) (0.062)
Firm-to-economy technological proximity −1.242∗∗∗ −1.050∗∗∗ 0.710∗∗∗
(0.076) (0.096) (0.111)
(0.248) (0.242)
Log(1 + technology spillovers) −0.068∗∗∗
(0.002)
Log(1 + patent count) 0.772∗∗∗ 0.787∗∗∗ 0.885∗∗∗
(0.008) (0.010) (0.010)
(0.027) (0.034) (0.033)
R&D 0.308∗∗∗ −0.104 −0.087
(0.064) (0.075) (0.073)
Notes. This table reports postalliance patent output of alliance firms using the treatment regression in Equation (4) with an instrumental
variable. The sample includes alliance firms and their industry- and size-matched control firms. The sample period is 1995–2003. Panel A
presents the first-stage instrumental variable regressions where the dependent variable is an indicator variable that takes the value of one if a
firm is in an alliance, and zero otherwise. We use a firm’s combined reporting index as the instrumental variable in the first-stage regressions.
The construction of this instrumental variable follows Bodnaruk et al. (2013). The combined reporting index is constructed using location
information on a firm’s headquarters and its subsidiaries. p-values (in parentheses) of the F-test for weak instrument with standard errors
clustered at the state-year level are reported in the last row. Panel B presents the second-stage treatment regressions where the dependent
variable is the logarithm of one plus patent count over the three-year period (year t + 1 to t + 3, where year t is the alliance announcement year).
The estimation of the two-stage regressions using Stata command itreatreg (Brown and Mergoupis 2010) allows for consistent estimation of the
interaction terms involving endogenous variables. Explanatory variables are measured at the fiscal year end before the alliance announcement
year. Definitions of the variables are provided in Table A.1 in the appendix.
∗ ∗∗
In column (1) we show that indeed, when a firm has We show that technology conglomerates are more
a higher value of the combined reporting index, that likely to form R&D-related alliances, and that postal-
firm is more likely to join an alliance. The p-values of liance, their innovation output significantly improves.
the F-test for a weak instrument with standard errors
clustered at the state-year level indicate that we reject 5.3. Controlling for Product and
the null hypothesis that our instrumental variable is Geographic Diversification
weak at the 1% level. In column (2) we also include So far, we have shown that technology conglom-
product market fluidity. Consistent with earlier results erates—firms that explore different technology fields
that product market fluidity is positively associated with heated inventive activity—have been using strate-
with the number of alliances, such fluidity is also pos- gic alliances to leverage and further enhance their inno-
itively associated with the likelihood of a firm joining vative capabilities. Given that technology conglomer-
alliances. To ensure that our results are not driven by ates are likely to have diversified patent and product
the knowledge spillovers effect, we further control for portfolios that are licensed and sold worldwide, it is
this effect in column (3) and find qualitatively similar important to control for the influence of product and
results. geographic diversity in our main analysis.
Panel B presents results from the second-stage treat- To capture product diversity, we use the number
ment regressions in Equation (4). In column (1) we of product segments reported in Compustat historical
show that the coefficient on the interaction term segment file. To capture geographic diversity, we intro-
Sample × Firm-to-Economy Technological Proximity is pos- duce an indicator variable, multinational corporation
itive and significant at the 1% level, suggesting that (MNC), that takes the value of one if the revenues out-
technology conglomerates do generate significantly side the United States exceeds 30% of the total revenues
more patents than other firms, consistent with our sec- (i.e., a multinational company, MNC), and zero other-
ond Hypothesis (H2). In column (2) we also include wise. Tables IA5–7 in the online appendix present the
product market fluidity and its respective interac- results.
tion term with the sample indicator variable, and Table IA5 presents pairwise correlation coefficients
in column (3) we further add technology spillovers for the outcome variable—the number of alliances
and its respective interaction term with the sample formed and different measures of diversity and prod-
indicator variable, to the baseline model in Equa- uct market competition. We show a weak correlation
tion (4). We show that the coefficient on the interac- between our measure of firm-to-economy technolog-
tion term Sample × Product Market Fluidity is negative ical proximity and product diversity, and a positive
and significant at the 1% level, again suggesting that correlation between our measure of firm-to-economy
alliances formed to cope with product market threats technological proximity and the MNC indicator, sug-
might serve purposes other than accelerating inno- gesting that technology conglomerates are more likely
vation output. For example, in alliances triggered by to be MNCs.
product market threats, more resources might be allo- Table IA6 replicates Table 4 by additionally control-
cated to product development after alliance forma- ling for product and geographic diversity. We show
tion, reducing resources available for corporate innova- that firms with more diverse product offerings are
tion. Importantly, after controlling for product market more likely, whereas multinationals are less likely, to
threats, the coefficient on the interaction term Sample × form alliances. Importantly, after controlling for those
Firm-to-Economy Technological Proximity remains posi- two diversity measures, our main findings remain:
tive and significant. The results provide further evi- technology conglomerates are more likely to form
dence that exploration in technology fields and prod- alliances.
uct market threats are distinctly different driving forces Table IA7 replicates Table 8 by additional controlling
behind firms redrawing their boundaries. for product and geographic diversity. In panel B, we
Ideally, we would focus on alliances intended only show that firms with more diverse product offerings
for expanding a firm’s innovative capability. In real- are associated with lower innovation output, whereas
ity, a R&D-related alliance related to innovation could multinationals are associated with more innovation
serve purposes other than just corporate innovation output. After controlling for these two diversity mea-
(as shown in examples in Table IA1). For example, sures, our main findings remain: technology conglom-
an alliance for implementing an innovative project erates are associated with more innovation output after
could appear as a funding agreement or a licens- alliance formation. We conclude that our main findings
ing agreement instead of a R&D agreement. With the are robust to other measures of diversity.
previous observation in mind, we limit our analysis In summary, we find that alliance formation has a
to a sample of alliance deals where a R&D agree- positive impact on technology conglomerates’ innova-
ment is involved, following Chemmanur et al. (2017). tion output, supporting our second Hypothesis (H2).
Table IA4 in the online appendix presents the results. Next, we implement mechanism/channel tests to shed
light on why technology conglomerates form alliances alliance’s formation, alliance partners produce more
and how alliances by technology conglomerates in turn patents in each other’s technology fields (classes), espe-
enhance their innovation output. cially for technology conglomerates, than they would
had they not formed an alliance. For each participant
6. The Mechanisms in an alliance deal announced in year t, we compute
Thus far, we have shown that technology conglomerates the number of patents applied for over the three-year
reap benefits from strategic alliances. As for how these period after alliance formation (t + 1 to t + 3) that are
benefits arise, the licensing motive suggests a one-way in the same classes as any of its partner’s patents dur-
transfer of knowledge from technology conglomerates ing the same period and compare the resulting num-
to their partners, the synergy motive suggests that the ber with that of a matching firm. Column (2) in Table 9,
increase in innovation output comes from knowledge panel B presents the results.
sharing and the cross-fertilization of innovative ideas Consistent with the synergy motive, we find that
between conglomerates and their partners, while the postalliance, alliance firms produce more patents in
tolerance for failure motive predicts greater risk taking their partners’ technology classes than their match-
that leads to more novel patents and/or patents with ing firms, especially when alliance firms are technol-
greater impact. In this section, we implement a number ogy conglomerates. The coefficient on the interaction
of tests to help identify the mechanism(s). term Samplei × Firm-to-Economy Technological Proximityit
is positive and significant at the 1% level (column (2),
6.1. The Synergy Motive panel B). This set of results helps substantiates that the
To examine knowledge pooling, we start with the con- increased patent output of technology conglomerates
cept of a firm’s existing knowledge, as first introduced really comes from these conglomerates learning from
by Benner and Tushman (2002) and Katila and Ahuja
their partners.
(2002), and recently applied in finance by Brav et al.
The final piece of firm-level evidence in support
(2017) and Gao et al. (2018). A firm’s existing knowl-
of the synergy motive is presented in Table IA8 in
edge refers to its portfolio of patents over the past five
the online appendix. We find that after an alliance’s
years and citations made by these patents; we use a
formation, pairwise technological proximity between
five-year window to smooth out the lumpiness in a
alliance partners significantly increases. However, we
firm’s patent output. For each participant in an alliance
do not find any significant change in the pairwise prod-
deal announced in year t, the existing knowledge of a
uct similarity measure of Hoberg and Phillips (2010)
participant is based on its patents applied for over the
between alliance partners after alliance formation. This
period t −7 to t −3, as our prealliance period is t −2 to t.
If knowledge pooling takes place between alliance is not surprising, as our main hypothesis defines the
participants, we would expect that after an alliance’s effect of technology conglomerates’ alliances on their
formation, one alliance partner would cite more than innovation output as measured by patents; whether
its control firm would of its partner’s existing knowl- and when those patents become commercialized may
edge, especially for technology conglomerates. For be far in the future.
each participant in an alliance deal, we compute the After establishing knowledge pooling and cross-
number of patents applied for over the three-year fertilization at the firm level, it is natural to explore
period after alliance formation (t + 1 to t + 3) that cite its whether and how such knowledge transfer takes place
partner’s existing knowledge and compare the result- at the inventor level. Some subset of inventors within
ing number with that of a matching firm. Column (1) a technology conglomerate will likely be closer than
in Table 9, panel B presents the results. others to alliance partners in terms of their technol-
Consistent with the synergy motive, we find that ogy expertise and/or geographic proximity. After an
postalliance, alliance firms produce more patents citing alliance’s formation, these inventors’ patent output
their partners’ existing knowledge than their match- should become both greater and closer to their part-
ing firms, especially for technology conglomerates. The ners’ patents in comparison with other inventors. This
coefficient on the interaction term Samplei × Firm-to- evidence will again help substantiate the influence of
Economy Technological Proximityit is positive and sig- alliances on subsequent innovation performance.
nificant at the 1% level (column (1), panel B). Taken To capture whether an inventor of one alliance part-
together, we verify that the increase in patent output ner works in similar technology fields as the other
indeed comes from technology conglomerates “bor- partner firm, an indicator variable, Inventor Overlapping
rowing” their partners’ knowledge, supporting the Technology Classes, takes the value of one if at least one
synergy motive of alliance formation. of the inventor’s prealliance patents over the three-year
As a result of and complementing the evidence period (t − 3 to t − 1) is in the same class as their part-
on knowledge pooling, we examine whether cross- ner’s patents over the same period, and zero otherwise.
fertilization also takes place. As noted in the Introduc- To capture whether an inventor of one alliance partner
tion, cross-fertilization refers to cases whereby after an locates closer than other inventors to the other partner
Table 9. The Underlying Mechanism: Synergy
(1) (2)
Combined reporting index 0.181∗∗∗ 0.194∗∗∗

(0.033) (0.033)
Firm-to-economy technological proximity 2.105∗∗∗ 1.937∗∗∗
(0.188) (0.190)
Log(1 + # patents citing partner’s existing knowledge) 0.335∗∗∗
(0.032)
Log(1 + # patents in partner’s technology classes) 0.171∗∗∗
(0.018)
Log(1 + patent count) 0.158∗∗∗ 0.135∗∗∗
(0.022) (0.023)
Patent diversification −0.396∗∗∗ −0.378∗∗∗
(0.082) (0.082)
R&D 2.359∗∗∗ 2.285∗∗∗
(0.177) (0.177)
Other firm characteristics Yes Yes
Year fixed effects Yes Yes
F-test for weak instrument (p-value) (0.00) (0.00)
Log(1 + # patents citing partner’s Log(1 + # patents in partner’s

existing knowledget+1, t+3 ) technology classest+1, t+3 )
(1) (2)
Sample × Firm-to-economy technological proximity 1.046∗∗∗ 1.428∗∗∗

(0.094) (0.143)
Sample 0.134∗∗∗ 0.178∗∗∗
(0.025) (0.040)
Firm-to-economy technological proximity −0.400∗∗∗ −0.773∗∗∗
(0.071) (0.108)
Log(1 + citations to partner’s existing knowledge) 0.757∗∗∗
(0.010)
Log(1 + cross fertilization) 0.573∗∗∗
(0.009)
Log(1 + patent count) 0.079∗∗∗ 0.114∗∗∗
(0.007) (0.011)
Patent diversification −0.164∗∗∗ −0.277∗∗∗
(0.024) (0.036)
R&D 0.106∗ 0.337∗∗∗
(0.058) (0.088)
Other firm characteristics Yes Yes
Year fixed effects Yes Yes
Notes. This table reports the channel tests on how the improvement in patent output of alliance firms takes place, using the treatment
regression in Equation (4) with an instrumental variable. The sample includes alliance firms and their industry- and size-matched control
firms. The sample period is 1995–2003. Panel A presents the first-stage instrumental variable regressions where the dependent variable is an
indicator variable that takes the value of one if a firm is in an alliance, and zero otherwise. We use a firm’s combined reporting index as the
instrumental variable in the first-stage regressions. The construction of this instrumental variable follows Bodnaruk et al. (2013). The combined
reporting index is constructed using location information on a firm’s headquarters and its subsidiaries. p-values (in parentheses) of the F-test
for weak instrument with standard errors clustered at the state-year level are reported in the last row. Panel B presents the second-stage
treatment regressions where the dependent variable in column (1) is the logarithm of one plus the number of patents citing a partner’s existing
knowledge, and the dependent variable in column (2) is the logarithm of one plus the number of patents in the same technology classes
as a partner’s patents. The estimation of the two-stage regressions using Stata command itreatreg (Brown and Mergoupis 2010) allows for
consistent estimation of the interaction terms involving endogenous variables. Explanatory variables are measured at the fiscal year end before
the alliance announcement year. Definitions of the variables are provided in Table A.1 in the appendix.
∗ ∗∗
Table 10. Inventor Closeness to a Partner Firm and Patent Output
Panel A: Inventor patent output
Log(1 + No. of an inventor’s patentst+1, t+3 )
(1) (2) (3)
Inventor overlapping technology classes 0.042∗∗∗ 0.037∗∗∗

(0.007) (0.007)
Inventor geographic proximity 0.017∗∗ 0.017∗∗
(0.009) (0.009)
Log(1 + No. of an inventor’s patents) 0.454∗∗∗ 0.461∗∗∗ 0.458∗∗∗
(0.006) (0.006) (0.007)
Log(1 + firm’s patent count) 0.076∗∗∗ 0.080∗∗∗ 0.079∗∗∗
(0.004) (0.004) (0.004)
Firm’s patent diversification −0.117∗∗∗ −0.113∗∗∗ −0.108∗∗∗
(0.027) (0.027) (0.027)
Firm’s R&D −0.283∗∗∗ −0.308∗∗∗ −0.325∗∗∗
(0.048) (0.049) (0.049)
Adjusted R-squared 0.24 0.24 0.24
Panel B: Inventor patent output in the same technology classes of partners’ patents
Log(1 + No. of an inventor’s patents in overlapping technology classest+1, t+3 )
(1) (2) (3)
Inventor overlapping technology classes 0.752∗∗∗ 0.743∗∗∗

(0.007) (0.007)
Inventor geographic proximity −0.014 −0.011
(0.010) (0.009)
Log(1 + No. of an inventor’s patents) 0.153∗∗∗ 0.221∗∗∗ 0.157∗∗∗
(0.007) (0.008) (0.007)
Log(1 + firm’s patent count) 0.021∗∗∗ 0.047∗∗∗ 0.026∗∗∗
(0.004) (0.005) (0.004)
Firm’s patent diversification −0.209∗∗∗ −0.310∗∗∗ −0.213∗∗∗
(0.029) (0.032) (0.029)
Firm’s R&D 0.517∗∗∗ 0.814∗∗∗ 0.482∗∗∗
(0.053) (0.059) (0.054)
Adjusted R-squared 0.30 0.09 0.30
Notes. This table reports inventor-level patent output after alliance formation. The sample of inventors includes all “stayers” in alliance firms.
An inventor is deemed as a “stayer” in an alliance firm if the following criteria are satisfied: (1) she is the inventor of at least one patent with the
same firm as the assignee during the three-year period before the alliance announcement; (2) she is the inventor of at least one patent with
the same firm as the assignee during the three-year period after the alliance announcement; and (3) she is not the inventor of any patent with
other firms as the assignee during either the three-year period before or the three-year period after the alliance announcement. The sample
period is 1995–2003. In panel A, the dependent variable is the logarithm of one plus the number of an inventor’s patents over the three-year
period (year t + 1 to t + 3, where year t is the alliance announcement year). In panel B, the dependent variable is the logarithms of one plus the
number of an inventor’s patents over the three-year period (year t + 1 to t + 3) that are in the same technology classes as its partner’s patents
over the same period. Explanatory variables are measured at the fiscal year end before the alliance announcement year. Definitions of other
variables are provided in Table A.1 in the appendix.
∗ ∗∗
firm, an indicator variable, Inventor Geographic Proxim- In panel A we examine the relationship between
ity, takes the value of one if the inventor resides less inventor closeness to partners and their patent out-
than 200 km away from their partner’s headquarters, put. Consistent with our conjecture, we show that
and zero otherwise. Table 10 presents the results. after alliance formation, inventors whose prealliance
patents are in their partners’ technology classes and/or (Levinthal and March 1993, McGrath 2001, Benner and
who are geographically closer produce more patents Tushman 2002, Smith and Tushman 2005, Manso 2011).
than other inventors. In panel B we examine the rela- Following Balsmeier et al. (2017) and Gao et al.
tionship between inventor closeness to partners and (2018), we use three measures to capture exploratory
cross-fertilization as measured by the number of an innovation. A patent is categorized as “exploratory” if
inventor’s postalliance patents in the same technol- 80% or more of its citations are not citing a firm’s exist-
ogy classes as their partner’s patents. We show that, ing patents or the citations made by those patents. The
after alliance formation, inventors whose prealliance first measure, No. of Exploratory Patents applied for in
patents are in their partners’ technology classes pro- year t + 1 to year t + 3, captures the amount of a firm’s
duce more patents in these classes than other inven- exploratory innovation. The second measure, No. of
tors, providing further evidence in support of knowl- Patents with Top 10% Citations, gauges the impact of a
edge spillovers and cross-learning. However, we do firm’s innovation by its number of patents with the top
not find that inventors’ geographic proximity is related 10% most citations. The third measure, No. of Patents
to cross-fertilization, which is not surprising, as geo- in New Classes, captures the extent of a firm patent-
graphical closeness could facilitate sharing knowledge ing into new technology classes. If tolerance for failure
that is not limited to common knowledge (i.e., in the plays an important role for technology conglomerates
same technology classes). It is also worth noting that in strategic alliances, we would expect that technology
our evidence on knowledge spillover between alliance conglomerates generate significantly more exploratory
partners suggests that licensing is unlikely to be the and impactful patents after alliance formation.
main reason behind alliance formation by technology Table 11 reports the results. The dependent variables
conglomerates. in columns (1)–(3) are the logarithm of one plus the
number of exploratory patents, the number of patents
6.2. The Tolerance for Failure Motive with the top 10% most citations, and the number of
To explore the tolerance for failure motive, we exam- patents in new technology classes. In panel B, we show
ine whether technological conglomerates engage more that the coefficients on the interaction term Samplei ×
in exploratory innovation after alliance formation. Firm-to-Economy Technological Proximityit are positive
Exploratory innovation requires new knowledge or and significant across all columns. The results are con-
a departure from existing knowledge and its payoffs sistent with the tolerance for failure motive that tech-
take longer to realize and are of higher uncertainty nological conglomerates have a more long-term view
Table 11. The Underlying Mechanism: Tolerance for Failure
(1) (2) (3)
Combined reporting index 0.164∗∗∗ 0.134∗∗∗ 0.160∗∗∗

(0.019) (0.019) (0.019)
Firm-to-economy technological proximity 1.800∗∗∗ 1.844∗∗∗ 2.085∗∗∗
(0.109) (0.105) (0.104)
Log(1 + # explorative patents) 0.193∗∗∗
(0.025)
Log(1 + # patents with top 1% citations) 0.223∗∗∗
(0.018)
Log(1 + # patents in new classes) 0.012
(0.015)
Log(1 + patent count) 0.051∗∗ 0.062∗∗∗ 0.206∗∗∗
(0.024) (0.017) (0.013)
(0.046) (0.049) (0.047)
R&D 2.176∗∗∗ 2.190∗∗∗ 2.201∗∗∗
(0.096) (0.096) (0.096)
Firm fixed effects No Yes Yes
F-test for weak instrument (p-value) (0.00) (0.00) (0.00)
Table 11. (Continued)
Log(1 + # exploratory Log(1 + # patents with Log(1 + # patents in

patentst+1, t+3 ) top 10% citationst+1, t+3 ) new classest+1, t+3 )
(1) (2) (3)
Sample × Firm-to-economy technological proximity 2.064∗∗∗ 1.295∗∗∗ 0.408∗∗∗

(0.108) (0.072) (0.082)
Sample 0.102∗∗∗ 0.094∗∗∗ 0.052∗
(0.034) (0.020) (0.027)
Firm-to-economy technological proximity −1.171∗∗∗ −0.478∗∗∗ −1.151∗∗∗
(0.080) (0.051) (0.059)
Log(1 + No. of explorative patents) 0.377∗∗∗
(0.016)
Log(1 + No. of patent with top 1% citations) 0.522∗∗∗
(0.008)
Log(1 + No. of patent in new classes) 0.096∗∗∗
(0.007)
Log(1 + patent count) 0.287∗∗∗ 0.238∗∗∗ 0.169∗∗∗
(0.015) (0.007) (0.007)
Patent diversification −0.435∗∗∗ −0.356∗∗∗ −0.013
(0.028) (0.019) (0.021)
R&D 0.381∗∗∗ −0.063 0.341∗∗∗
(0.064) (0.042) (0.049)
Firm fixed effects No Yes Yes
Notes. This table reports postalliance patent output of alliance firms using the treatment regression in Equation (4) with an instrumental
variable. The sample includes alliance firms and their industry- and size-matched control firms. The sample period is 1995–2003. Panel A
presents the first-stage instrumental variable regressions where the dependent variable is an indicator variable that takes the value of one if a
firm is in an alliance, and zero otherwise. We use a firm’s combined reporting index as the instrumental variable in the first-stage regressions.
The construction of this instrumental variable follows Bodnaruk et al. (2013). The combined reporting index is constructed using location
information on a firm’s headquarters and its subsidiaries. The p-values (in parentheses) of the F-test for weak instrument with standard errors
clustered at the state-year level are reported in the last row. Panel B presents the second-stage treatment regressions. The dependent variables
in columns (1)–(3) are the logarithm of one plus the number of exploratory patents, the number of patents with the top 10% most citations, and
the number of patents in new classes, respectively, over the three-year period (year t + 1 to t + 3, where year t is the alliance announcement year).
The estimation of the two-stage regressions using Stata command itreatreg (Brown and Mergoupis 2010) allows for consistent estimation of the
interaction terms involving endogenous variables. Explanatory variables are measured at the fiscal year end before the alliance announcement
year. Definitions of the variables are provided in Table A.1 in the appendix.
∗ ∗∗
and greater tolerance for experimentation and failure their boundaries, little research has addressed the
and thus, are able to benefit more from alliances in pro- organizational choice of technology conglomerates and
ducing exploratory and impactful innovation output. its role in innovative activities (one exception is Bena
In summary, our evidence on knowledge pooling, and Li 2014). In this paper, we extend this line of
cross-fertilization, and exploratory innovation pro- inquiry and ask the following research questions: How
vides strong support for both the synergy and toler- is the extent to which an innovative firm is a tech-
ance for failure motives behind alliance formation by nology conglomerate measured? In addition to M&As,
technology conglomerates. do technology conglomerates use strategic alliances to
gain access to new technology? Do strategic alliances
facilitate technology conglomerates’ innovative activi-
7. Conclusions ties and, if so, how?
Since the late 1990s, many technology firms have sig- We develop a measure of firm-to-economy technolog-
nificantly expanded their innovative capabilities, facil- ical proximity to capture the extent of a firm’s tech-
itated by acquisitions and alliances. Despite an increas- nology conglomeration, that is, its exploration of dif-
ing number of firms exploring various technology ferent technology fields with heated inventive activity.
fields with heated inventive activity and redrawing Using a large patent-alliance data set over the period
1985–2007, we show that technology conglomerates Acknowledgments

are more likely to form alliances. Moreover, we show The authors thank the department editor, Gustavo Manso; an
that alliances lead to higher patent output, especially anonymous associate editor; three anonymous referees; and
for technology conglomerates. In terms of underlying Jan Bena, Andriy Bodnaruk, Ling Cen, Peter Cziraki, Steve
mechanisms, we show that after alliance formation, Dimmock, Alexander Dyck, Felix Feng, Mariassunta Gian-
netti, Jerry Hoberg, Po-Hsuan Hsu, Chuan Yang Hwang, Vĳay
more patents of a technology conglomerate make back-
Jog, Nengjiu Ju, Andrew Karolyi, Laura Lindsey, Roger Loh,
ward citations to its partner’s prealliance patents or
William Mann, Vadim Marmer, Francisco Santos, Charles
backward citations made by these patents, and that Trzcinka, Jamie Weathers, Ting Xu, and Shan Zhao; they also
more patents of a technology conglomerate are in thank seminar participants at Carleton University, Nanyang
the technology classes of its partner’s patents. More- Technological University, Norwegian School of Economics,
over, we show that after alliance formation, technology Shanghai Advanced Institute of Finance, Singapore Manage-
conglomerates produce more patents that are novel ment University, Stockholm School of Economics, University
and/or with greater impact. Our evidence on knowl- of Hawaii, University of Notre Dame, University of Toronto,
edge pooling, cross-fertilization, and exploratory inno- and Wilfrid Laurier University; and they thank conference
vation is consistent with both the synergy and toler- participants at the 2015 Eastern Finance Association Meet-
ance for failure motives behind alliance formation by ings (New Orleans), the 2015 Conference on Entrepreneurial
technology conglomerates. Finance and Innovation around the World (Beĳing), the
2015 China International Conference in Finance (Shenzhen),
Our analysis in the paper is silent on answering
the 2015 European Finance Association Meetings (Vienna),
the important question of why some innovative firms and the 2015 Northern Finance Association Meetings (Lake
become technology conglomerates while others remain Louise) for helpful comments. The authors also thank Andriy
technologically focused. We believe answering this Bodnaruk for providing data on combined reporting and
fundamental question will be an important and fruitful Joyce Guan and Alice Guo for excellent research assistance.
direction for future research. All errors are those of the authors.
Appendix
Table A.1. Variable Definitions
Firm-to-economy Firm-to-economy technological proximity is computed as a cosine similarity measure

technological proximity
(Technology Si, t S−i, t
Firm-to-Economy Technological Proximityi, t , .
conglomeration) kSi, t k kS−i, t k
where the vector Si, t (s i, 1, t , . . . , s i, k, t , . . . , s i, K, t ) captures the scope of innovative activity through patent output of
firm i, and the vector S−i, t (s −i, 1, t , . . . , s −i, k, t , . . . , s−i, K, t ) captures the scope of innovative activity through patent
output of all other public firms in the economy excluding firm i. The subscript k in (1, K) is the technology class
index. The scalar s i, k, t (s−i, k, t ) is the ratio of the number of awarded patents to firm i (all other public firms in the
economy excluding firm i) in technology class k with application years from t − 2 to t (application year t) to the total
number of awarded patents to firm i (all other public firms in the economy excluding firm i) applied for over the
same period. This scalar is set to zero if an innovative firm does not have any patent application over year t − 2 to t.
Product market fluidity This variable is a measure of the competitive threats faced by a firm in its product market space that captures
changes in rival firms’ products relative to the firm’s products (see Hoberg et al. 2014, for more detailed
discussion). We obtain the data from the Hoberg-Phillips Data Library at http://alex2.umd.edu/industrydata/.
Technology spillovers Following Bloom et al. (2013), technology spillovers is computed as the following:
X Si · S j
Technology Spilloversi, t R&D stock j, t ×
j ,i kSi k kS j k
where Si (S j ) is the technology vector capturing patent output of firm i (j) during the period 1976–2006. R&D
capital stock is calculated using a perpetual inventory method with a 15% depreciation rate. So
R&D stock j, t R&Dt + (1 − δ)R&D stock j, t−1 , where R&Dt is R&D expenditures in year t, δ 0.15,
R& D stock j, 0 R&D0 /g, and g is the steady-state growth rate of the R&D stock, estimated as the average annual
R&D growth rate of Compustat firms that are in the same two-digit SIC industry as firm j.
Patent count This variable is constructed in three steps. First, for each technology class k and patent application year t, we calculate
the median value of the number of awarded patents in technology class k with application year t across all firms
that were awarded at least one patent in technology class k with application year t. Second, we scale the number of
awarded patents to firm i in technology class k with application year t by the corresponding (technology class and
application year) median value from the first step. Finally, for firm i, we sum the scaled number of awarded patents
from the second step across all technology classes and across application years from t − 2 to t. Since firms’
patenting activities tend to cluster over technology classes and time, patent count thus measures a firm’s relative
productivity in innovation by excluding those clustering effects. This variable is set to zero if an innovative firm
does not have any patent application over year t − 2 to t.
Table A.1. (Continued)
Patent diversification Following Hirshleifer et al. (2014), patent diversification is constructed as
K
2
X n i, k, t
Patent Diversificationi, t 1 − PK ,
k1 k1 n i, k, t
where n i, k, t is the number of awarded patents to firm i in technology class k with application years from
t − 2 to t and K is the total number of patent classes.
R&D R&D expenditures scaled by total assets.
Total assets Book value of total assets in millions of 2007 constant dollars.
Firm age The number of years since a firm is covered by CRSP.
HHI The Herfindahl index of a two-digit SIC industry based on sales.
Leverage Total debt scaled by total assets, where total debt is the sum of short-term debt and long-term debt.
ROA Earnings before interest, taxes, depreciation, and amortization scaled by total assets.
Cash holdings Cash and short-term investment scaled by total assets.
Tobin’s Q Market value of total assets scaled by book value of total assets, where market value of total assets is
computed as book value of total assets minus book value of common equity plus market value of common
equity.
Sales growth The ratio of sales in year t to sales in year t − 1 minus one.
Capex Capital expenditures scaled by total assets.
Sample An indicator variable that takes the value of one if it is an alliance firm, and zero otherwise.
Combined reporting index Following Bodnaruk et al. (2013), for each subsidiary (including headquarters) of a firm, the combined
reporting indicator variable is set to one if the subsidiary (or headquarters) is located in a state that
imposes the combined reporting rule, and zero otherwise. The firm’s combined reporting index is the
average value of the combined reporting indicator variable across all of its subsidiaries (including
headquarters) when the information on subsidiary location is available, and the value of the combined
reporting indicator variable based on headquarters location otherwise.
Firm-to-firm technological Following Jaffe (1986), firm-to-firm technological proximity is computed as a correlation coefficient
proximity
Si, t S0j, t
Firm-to-Firm Technological Proximityi, j, t √ √ ,
Si, t S0i, t S j, t S0j, t
where the vector Si, t (s i, 1, t , . . . , s i, k, t , . . . , s i, K, t ) captures the scope of innovative activity through patent
output of firm i, and the vector S j, t (s j, 1, t , . . . , s j, k, t , . . . , s j, K, t ) captures the scope of innovative activity
through patent output of firm j. The subscript k in (1, K) is the technology class index. The scalar s i, k, t
(s j, k, t ) is the ratio of the number of awarded patents to firm i (j) in technology class k with application
years from t − 2 to t to the total number of awarded patents to firm i (j) applied for over the same period.
Same industry An indicator variable that takes the value of one if both partners to a deal operate in the same two-digit SIC
industry, and zero otherwise.
Same state An indicator variable that takes the value of one if both partners to a deal are incorporated in the same state,
and zero otherwise.
No. of patents citing partner’s The number of patents over a three-year period (year t − 2 to t) citing a partner’s existing knowledge, which
existing knowledge refers to its portfolio of patents over the past five years (year t − 7 to t − 3 where year t is the last year of the
three-year period) and citations made by these patents.
No. of patents in partner’s The number of patents over a three-year period in the same technology classes as a partner’s patents over the
technology classes same period.
No. of exploratory patents For each patent applied by firm i over a three-year period (year t − 2 to t), we first calculate its number of
citations on the firm’s own existing knowledge, which refers to its portfolio of patents over the past five
years (year t − 7 to t − 3 where year t is the last year of the three-year period) and citations made by these
patents. The patent is categorized as “exploratory” if more than or equal to 80% of its citations are outside
of firm i’s existing expertise. We then compute the total number of exploratory patents over the three-year
period.
No. of patents with top 10% For each patent applied by firm i over a three-year period (year t − 2 to t), we first calculate the number of
citations citations it receives over the five-year period subsequent to its grant year. A patent is categorized as “with
top 10% citations” if its number of citations received is among the top 10 percentile for the application year
and patent class. We then compute the total number of patents with the top 10% most citations over the
three-year period.
No. of patents in new classes The number of a firm’s patents over a three-year period (year t − 2 to t) not in the same technology classes as
its patents over the prior five-year period (year t − 7 to t − 3 where year t is the last year of the three-year
period).
Inventor overlapping An indicator variable that takes the value of one if at least one of an inventor’s patents owned by an alliance
technology classes participant over a three-year period is in the same technology classes as their partner’s patents over the
same period, and zero otherwise.
Table A.1. (Continued)
Inventor geographic An indicator variable that takes the value of one if an inventor affiliated with an alliance participant resides
proximity less than 200 km away from their partner’s headquarters, and zero otherwise.
No. of an inventor’s patents The number of an inventor’s patents owned by an alliance participant over a three-year period in the same
in overlapping technology technology classes as a partner’s patents over the same period.
classes
Firm-to-firm product An indicator variable that takes the value of one if a given firm pair is from the same industry, constructed
similarity using the text-based analysis of the firms’ product descriptions in the 10-K filings by Hoberg and Phillips
(2010), and zero otherwise. (We downloaded the data from Hoberg-Phillips
Data Library website: http://www.rhsmith.umd.edu/industrydata/.)
Note. Firm characteristics are measured as of the fiscal year end before the alliance deal announcement and are winsorized at the 1st and 99th
percentiles.
Endnotes 13
Specifically, we use 10 years of alliance data to construct the pre-
1 sample average of a firm’s alliance formation. As noted in Bloom
We focus on innovation outcome of strategic alliances given that
Bena and Li (2014) have provided an extensive analysis of post-M&A et al. (2013), this approach can serve as an initial condition to proxy
innovation outcome. for unobserved heterogeneity under the assumption that the first
2 moments of all observables are stationary. Using patent data over
Martin and Sayrak (2003) provide a comprehensive review of the
literature on product diversification. 1965–1979 as the output (including using 1965–1989 as the presample
3
period) and R&D expenditures over 1972–1979 as the input, Blundell
Given the much-touted capabilities of technology conglomerates, et al. (2002) provide Monte Carlo evidence showing that this presam-
the inevitable question arises: Why have they not developed more
ple mean scaling estimator performs well compared to alternative
new technologies themselves? The fact that all big pharmaceuti-
econometric estimators for dynamic panel data models with weakly
cal companies keep forming strategic alliances and acquiring small
endogenous variables.
biotech firms suggests that in-house R&D and external acquisition of
14
innovation are complements. Indeed, there is a longstanding man- As shown in Table IA1 in the online appendix, some alliances serve
agement literature starting with the seminal work by Cohen and multiple purposes, as would be the case for example, with a R&D
Levinthal (1989), Rosenberg (1990), and Cassiman and Veugelers agreement and a marketing agreement, in which the former could
(2006) arguing that firms engaging in R&D for two purposes. First, be driven by explorations in technology fields, while the latter could
they invest in R&D to generate innovation. Second, R&D enhances be due to explorations in product markets.
the “absorptive capacity” of a firm so that it can better exploit exter- 15
See McFadden (1974) for an introduction to the conditional logit
nally generated innovation. regression, and Bena and Li (2014) for a recent application in finance.
4 16
According to SDC, their data come from SEC filings and their inter- To be included in the M&A sample, we require that (1) the deal is
national counterparts, trade publications, wires, and news sources. eventually completed; (2) the form of the deal is coded by SDC as a
5 merger, an acquisition of majority interest, or an acquisition of assets;
SDC only covers alliances between for-profit firms, and we acknowl-
edge that our study is limited to such collaborations. and (3) the acquirer owns less than 50% of the target firm prior to
6
It is worth noting that our main findings remain unchanged if we the deal and is seeking to own more than 50% of the target firm. We
use the NBER Patent Database that ends in 2006 (Hall et al. 2005). end up with 16,090 M&A deals by innovative firms over the sample
7 period 1995–2007.
See Table IA1 in the online appendix for examples of alliance agree-
17
ments for our sample deals. To the extent that the alliance sample If a firm has done an acquisition and formed an alliance in the same
includes deals motivated by purposes other than corporate innova- year, it will show up as one observation in the M&A pair and one
tion, our estimates of alliances’ effects on patent output will likely observation in the alliance pair. In our sample of M&A and alliance
understate their magnitude. With the previous caveat in mind, we deals, there are only 53 pairs of firms showing up in both samples,
later conduct a robustness check by limiting our sample of alliances and only seven of those pairs are in the same year.
to those where a R&D agreement is involved. 18
See Li and Prabhala (2007) for an overview of identifying treatment
8 effects in corporate finance.
Lerner et al. (2011) and Bloom et al. (2013) apply the similar filter for
innovative firms. About 40% of the Compustat firms over the period 19
One natural question to ask is why innovative firms would not,
1950–2009 are innovative based on our definition. over time, move their operations to states that do not require com-
9
Whenever the United States Patent and Trademark Office (USPTO) bined reporting (for tax reasons). The last row in Table IA2 panel
changes its technology classification system, it retroactively changes A provides some suggestive evidence: Firms located in combined
the class assignments for all previous patents to maintain consistency reporting states tend to have more peer firms (based on the two-digit
at a particular point in time. Hence, our measures are unaffected by SIC codes) within the same state than those located in noncombined
changes in the classification system. During our sample period, there reporting states, suggesting a possible agglomeration benefit from
are 426 technology classes. states with combined reporting.
10
Among the 20,136 alliance deals between the sample period 1995–
2007, we find only 48 deals involving equity stakes and 153 deals References
resulting in acquisitions.
Aghion P, Van Reenen J, Zingales L (2013) Innovation and institu-
11
Note that patent diversification is set to zero if a firm does not have tional ownership. Amer. Econom. Rev. 103:277–304.
any patents in recent three years. Ahuja G (2000) Collaboration networks, structural holes, and inno-
12
The correlation between patent count and patent diversification is vation: A longitudinal study. Admin. Sci. Quart. 45:425–455.
0.84. It is worth noting that our main findings regarding technologi- Allen JW, Phillips GM (2000) Corporate equity ownership, strate-
cal proximity do not change if we exclude patent diversification from gic alliances, and product market relationships. J. Finance 55:
the regression specification (results available upon request). 2791–2815.
Balsmeier B, Fleming L, Manso G (2017) Independent boards and Hirshleifer DA, Hsu P-H, Li D (2014) Don’t hide your light under a
innovation. J. Financial Econom. 123:536–557. bushel: Innovative diversity and stock returns. Working paper,
Baum JAC, Calabrese T, Silverman BS (2000) Don’t go it alone: University of California, Irvine.
Alliance network composition and startups’ performance in Hoberg G, Phillips G (2010) Product market synergies and compe-
Canadian biotechnology. Strategic Management J. 21:267–294. tition in mergers and acquisitions: A text-based analysis. Rev.
Bena J, Li K (2014) Corporate innovations and mergers and acquisi- Financial Stud. 23:3773–3811.
tions. J. Finance 69:1923–1960. Hoberg G, Phillips G, Prabhala N (2014) Product market threats,
Benner MJ, Tushman M (2002) Process management and technolog- payouts, and financial flexibility. J. Finance 69:293–324.
ical innovation: A longitudinal study of the photography and Holmström B, Roberts J (1998) The boundaries of the firm revisited.
paint industries. Admin. Sci. Quart. 47:676–706. J. Econom. Perspect. 12:73–94.
Bloom N, Schankerman M, Van Reenen J (2013) Identifying tech- Huddleston J, Sicilian S (2008) History and considerations for com-
nology spillovers and product market rivalry. Econometrica 81: bined reporting: Will states adopt a model combined reporting
1347–1393. statute? State Local Tax Lawyer Sympos. Edition, 3–20.
Blundell R, Griffith R, Van Reenen J (1999) Market shares, market Jaffe AB (1986) Technological opportunity and spillovers of R&D:
value and innovation in a panel of British manufacturing firms. Evidence from firms’ patents, profits, and market value. Amer.
Rev. Econom. Stud. 66:529–554. Econom. Rev. 76:984–1001.
Blundell R, Griffith R, Windmeĳer F (2002) Individual effects and
Jensen MC, Meckling WH (1992) Specific and general knowledge,
dynamics in count data models. J. Econometrics 128:113–131.
and organizational structure. Werin L, Wĳkander H, eds. Con-
Bodnaruk A, Massa M, Simonov A (2013) Alliances and corporate
tract Economics (Blackwell, Oxford, UK), 251–274.
governance. J. Financial Econom. 107:671–693.
Katila R, Ahuja G (2002) Something old, something new: A longitu-
Boone AL, Ivanov VI (2012) Bankruptcy spillover effects on strategic
dinal study of search behavior and new product introduction.
alliance partners. J. Financial Econom. 103:551–569.
Branstetter LG, Sakakibara (2002) When do research consortia work Acad. Management J. 45:1183–1194.
well and why? Evidence from Japanese panel data. Amer. Kogan L, Papanikolaou D, Seru A, Stoffman N (2017) Technological
Econom. Rev. 92:143–159. innovation, resource allocation, and growth. Quart. J. Econom.
Brav A, Jiang W, Ma S, Tian X (2017) How does hedge fund activism 132:665–712.
reshape corporate innovation? J. Financial Econom. 130:237–264. Kortum S, Lerner J (2000) Assessing the contribution of venture cap-
Brown G, Mergoupis T (2010) Treatment interactions with non- ital to innovation. RAND J. Econom. 31:674–692.
experimental data in Stata. Bath Economics Research Paper Lerner J, Shane H, Tsai A (2003) Do equity financing cycles mat-
No. 10/10, University of Bath, Bath, UK. ter? Evidence from biotechnology alliances. J. Financial Econom.
Cassiman B, Veugelers R (2006) In search of complementarity in 67:411–446.
innovation strategy: Internal R&D and external knowledge Lerner J, Sorensen M, Strömberg P (2011) Private equity and long-
acquisition. Management Sci. 52:68–82. run investment: The case of innovation. J. Finance 66:445–477.
Chan SH, Kensinger JW, Keown AJ, Martin JD (1997) Do strategic Levinthal DA, March JG (1993) The myopia of learning. Strategic
alliances create value? J. Financial Econom. 46:199–221. Management J. 14:95–112.
Chemmanur T, Loutskina E, Tian X (2014) Corporate venture capital, Li G-C, Lai R, D’Amour A, Doolin DM, Sun Y, Torvik VI, Yu AZ,
value creation, and innovation. Rev. Financial Stud. 27:2434–2473. Fleming L (2014) Disambiguation and co-authorship networks
Chemmanur T, Shen Y, Xie J (2017) Innovation beyond firm bound- of the U.S. patent inventor database (1975–2010). Res. Policy
aries: Common blockholders, strategic alliances, and corporate 43:941–955.
innovation. Working paper, Boston College, Chestnut Hill, MA. Li K, Prabhala N (2007) Self-selection models in corporate finance.
Coase RH (1937) The nature of the firm. Economica 4:386–405. Eckbo BE, ed. Handbook of Corporate Finance: Empirical Corporate
Cohen WM, Levinthal DA (1989) Innovation and learning: The two Finance (Elsevier/North-Holland, Amsterdam), 37–86.
faces of R&D. Econom. J. 99:569–596. Lindsey L (2008) Blurring firm boundaries: The role of venture capi-
Fee CE, Hadlock CJ, Thomas S (2006) Corporate equity ownership tal in strategic alliances. J. Finance 63:1137–1168.
and the governance of product market relationship. J. Finance Liu T, Servilir M, Tian X (2016) Acquiring innovation. Working paper,
61:1217–1250. Wharton School, Philadelphia.
Frésard L, Hoberg G, Phillips G (2015) Innovative activities and Manso G (2011) Motivating innovation. J. Finance 66:1823–1869.
the incentives for vertical acquisitions and integration. Working Martin JD, Sayrak A (2003) Corporate diversification and share-
paper, University of Maryland, College Park. holder value: A survey of recent literature. J. Corporate Finance
Fulghieri P, Sevilir M (2009) Organization and financing of innova- 9:37–57.
tion, and the choice between corporate and independent venture
Mazerov M (2009) A majority of states have now adopted a key
capital. J. Financial Quant. Anal. 44:1291–1321.
corporate tax reform—“Combined reporting.” Working paper,
Gans JS, Hsu D, Stern S (2002) When does start-up innovation spur
Center on Budget and Policy Priority, Washington, DC.
the gale of creative destruction? RAND J. Econom. 33:571–586.
McFadden D (1974) Conditional logit analysis of qualitative choice
Gao H, Hsu P-H, Li K (2018) Innovation strategy of private firms.
behavior. Zarembka P, ed. Frontiers of Econometrics (Academic
J. Financial Quant. Anal. 53:1–32.
Press, New York), 105–142.
Gomes-Casseres B, Hagedoorn J, Jaffe AB (2006) Do alliances pro-
mote knowledge flows? J. Financial Econom. 80:5–33. McGrath RG (2001) Exploratory learning, innovative capacity and
Grossman SJ, Hart OD (1986) The costs and benefits of ownership: managerial oversight. Acad. Management J. 44:118–131.
A theory of vertical and lateral integration. J. Political Econom. Miller DJ (2006) Technological diversity, related diversification, and
94:691–719. firm performance. Strategic Management J. 27:601–619.
Hall BH, Jaffe AB, Trajtenberg MM (2005) The NBER patent citation Mody A (1993) Learning through alliances. J. Econom. Behav. Organ.
data file: Lessons, insights and methodological tools. Jaffe AB, 20:151–170.
Trajtenberg M, eds. Patents, Citations and Innovations: A Window Mowery DC, Oxley JE, Silverman BS (1996) Strategic alliances
on the Knowledge Economy (MIT Press, Cambridge, MA), 403–470. and interfirm knowledge transfer. Strategic Management J. 17:
Hart O (1995) A natural-resource-based view of the firm. Acad. Man- 77–91.
agement Rev. 20:986–1014. Peeters C, van Pottelsberghe de la Potterie B (2006) Innovation strat-
Hart O, Moore J (1990) Property rights and the nature of the firm. egy and the patenting behavior of firms. J. Evolutionary Econom.
J. Political Econom. 98:1119–1158. 16:109–135.
He J, Tian X (2018) Finance and corporate innovation: A survey. Asia- Phillips GM, Zhdanov A (2013) R&D and the incentives from merger
Pacific J. Financial Stud. 47:165–212. and acquisition activity. Rev. Financial Stud. 26:34–78.
Podolny JM (1994) Market uncertainty and the social character of Smith WK, Tushman ML (2005) Managing strategic contradictions:
economic exchange. Admin. Sci. Quart. 39:458–483. A top management model for managing innovation streams.
Rhodes-Kropf M, Robinson DT (2008) The market for mergers and Organ. Sci. 16:522–536.
the boundaries of the firm. J. Finance 63:1169–1211. Solomon SD (2014) New buying strategy as Facebook and
Robinson DT (2008) Strategic alliances and the boundaries of the Google transform into web conglomerates. New York Times
firm. Rev. Financial Stud. 21:649–681. (August 5), https://dealbook.nytimes.com/2014/08/05/new
Robinson DT, Stuart TE (2007) Financial contracting in biotech strate- -strategy-as-tech-giants-transform-into-conglomerates/.
gic alliances. J. Law Econom. 50:559–596. Stuart TE (1998) Network positions and propensities to collaborate:
Rosenberg N (1990) Why do firms do basic research? Res. Policy An investigation of strategic alliance formulation in a high-
19:165–174. technology industry. Admin. Sci. Quart. 43:668–698.
Sampson RC (2005) Experience effects and collaborative returns in Stuart TE, Hoang H, Hybels RC (1999) Interorganizational endorse-
R&D alliances. Strategic Management J. 26:1009–1031. ments and the performance of entrepreneurial ventures. Admin.
Seru A (2014) Firm boundaries matter: Evidence from con- Sci. Quart. 44:315–349.
glomerates and R&D activity. J. Financial Econom. 111: Tian X, Wang T (2014) Tolerance for failure and corporate innovation.
381–405. Rev. Financial Stud. 27:211–255.

Technology Conglomeration, Strategic Alliances, and Corporate Innovation

Uploaded by

Copyright:

Available Formats

Technology Conglomeration, Strategic Alliances, and Corporate Innovation

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Technology Conglomeration, Strategic Alliances, and Corporate Innovation

Uploaded by

Copyright:

Available Formats

This article was downloaded by: [129.100.58.

76] On: 19 March 2019, At: 20:30

Technology Conglomeration, Strategic Alliances, and

To cite this article:

Published online in Articles in Advance 19 Mar 2019

Full terms and conditions of use: https://pubsonline.informs.org/page/terms-and-conditions

Copyright © 2019, INFORMS

Please scroll down for article—it is on subsequent pages

Technology Conglomeration, Strategic Alliances, and

History: Accepted by Gustavo Manso, finance.

1. Introduction Despite an increasing number of firms becoming

Figure 1. (Color online) Technology Conglomeration Over Time

1980 1990 2000

General Electric Co 0.633 Du Pont (E I) De Nemours 0.666 IBM 0.715

Table 3. Descriptive Statistics for the Panel Data Sample

Panel A: Summary statistics for the panel data sample

Mean S.D. Median

No. of alliances 0.470 2.033 0.000

(1) No. of alliances

Table 4. Forming Alliances: The Panel Data Sample

(1) (2) (3) (4) (5) (6)

Firm-to-economy 2.641∗∗∗ 1.150∗∗∗ 0.891∗∗∗ 0.848∗∗∗ 0.818∗∗∗

Table 5. Descriptive Statistics for the Matching Firm Sample

Alliance firms Control firms

Mean S.D. Median Mean S.D. Median t-test Wilcoxon

Table 6. Forming Alliances: The Matching Firm Sample

Alliance firm 1, control firm 0

Industry match and size match Random match

(1) (2) (3) (4)

Firm-to-economy 4.000∗∗∗ 3.372∗∗∗ 4.967∗∗∗ 4.415∗∗∗

Table 7. Choosing Between Alliances and M&As 5. Technology Conglomerates, Alliances,

Table 8. Instrumental Variable Regressions

Panel A: The first-stage regressions

Alliance firm 1, control firm 0

(1) (2) (3)

Combined reporting index 0.171∗∗∗ 0.169∗∗∗ 0.156∗∗∗

Panel B: The second-stage regressions

Log(1 + patent countt+1, t+3 )

(1) (2) (3)

Table 9. The Underlying Mechanism: Synergy

Panel A: The first-stage regressions

Alliance firm 1, control firm 0

Combined reporting index 0.181∗∗∗ 0.194∗∗∗

Panel B: The second-stage regressions

Log(1 + # patents citing partner’s Log(1 + # patents in partner’s

Sample × Firm-to-economy technological proximity 1.046∗∗∗ 1.428∗∗∗

Table 10. Inventor Closeness to a Partner Firm and Patent Output

Panel A: Inventor patent output

Log(1 + No. of an inventor’s patentst+1, t+3 )

(1) (2) (3)

Inventor overlapping technology classes 0.042∗∗∗ 0.037∗∗∗

Log(1 + No. of an inventor’s patents in overlapping technology classest+1, t+3 )

(1) (2) (3)

Inventor overlapping technology classes 0.752∗∗∗ 0.743∗∗∗

Table 11. The Underlying Mechanism: Tolerance for Failure

Panel A: The first-stage regressions

Alliance firm 1, control firm 0

(1) (2) (3)

Combined reporting index 0.164∗∗∗ 0.134∗∗∗ 0.160∗∗∗