-
MANAGEMENT SCIENCEVol. 60, No. 4, April 2014, pp. 867–887ISSN
0025-1909 (print) ó ISSN 1526-5501 (online)
http://dx.doi.org/10.1287/mnsc.2013.1801
©2014 INFORMS
Entrepreneurial Exits and Innovation
Vikas A. AggarwalINSEAD, 77305 Fontainebleau, France,
[email protected]
David H. HsuThe Wharton School, University of Pennsylvania,
Philadelphia, Pennsylvania 19104, [email protected]
We examine how initial public offerings (IPOs) and acquisitions
affect entrepreneurial innovation as mea-sured by patent counts and
forward patent citations. We construct a firm-year panel data set
of all venturecapital-backed biotechnology firms founded between
1980 and 2000, tracked yearly through 2006. We addressthe
possibility of unobserved self-selection into exit mode by using
coarsened exact matching, and in two addi-tional ways: (1)
comparing firms that filed for an IPO (or announced a merger) with
those not completing thetransaction for reasons unrelated to
innovation, and (2) using an instrumental variables approach. We
find thatinnovation quality is highest under private ownership and
lowest under public ownership, with acquisitionintermediate between
the two. Together with a set of within-exit mode analyses, these
results are consistentwith the proposition that information
confidentiality mechanisms shape innovation outcomes. The results
arenot explained by inventor-level turnover following exit events
or by firms’ preexit window dressing behavior.
Keywords : entrepreneurial exits; innovation; information
confidentialityHistory : Received February 16, 2012; accepted July
29, 2013, by Lee Fleming, entrepreneurship and innovation.
Published online in Articles in Advance December 19, 2013.
1. IntroductionEquity investments in entrepreneurial start-ups
areilliquid until an exit (or liquidity) event such as aninitial
public offering (IPO) or acquisition by anotherentity.1 As a
result, a leading performance mea-sure that researchers in the
entrepreneurship litera-ture investigate is the likelihood of an
exit event.The main motivation for studying such outcomes isthat
these events offer liquidity and financial returnsto the
entrepreneurial founders, their investors, andother shareholders.
We know little, however, aboutthe relationship between
entrepreneurial exit modesand organizational innovation,
particularly when tak-ing into account self-selection.
Understanding the linkbetween exits modes and innovation outcomes
isimportant to start-up entrepreneurs and managers atestablished
companies alike. For entrepreneurs, alter-nate exit mode choices
involve trade-offs in organi-zational structure, governance,
incentives, resources,and degree of information disclosure—all of
whichcan shape innovation outcomes. For industry incum-bents, a
deeper understanding of the consequencesof organizational changes
accompanying the goingpublic process and the entrepreneurial
acquisition
1 We use the terms “exit event” and “liquidity event”
interchange-ably. These refer to the ability of the entrepreneur or
venture capi-talist (VC) to fully or partially sell their equity
stake in a VC-backedstart-up firm.
process can be important in assessing the innova-tion profile of
potential competitors.2 We thereforeexamine the research question
of the relationshipbetween entrepreneurial exit mode and
innovationwhile taking into account the role of
(unobserved)entrepreneurial self-selection into exit mode.To
illustrate the phenomenon we study, consider
the example of Genentech. Tom Perkins, cofounderof the venture
capital firm Kleiner Perkins and chair-man of Genentech’s board
from 1976 through 1990,reflected on the company’s possible sale to
Eli Lillyprior to Genentech’s 1980 public offering: “We didhave
some preliminary discussions with Lilly. Theymade one of the
biggest mistakes in business his-tory in that they didn’t try to
push us very hard tosell the company. I think if Lilly, a year
before thepublic issue, had made an attractive offer we prob-ably
would have gone for it. Because there were no
2 Another motivation for investigating the relationship
betweenentrepreneurial exit modes and innovation outcomes is to
betterassess the public policy implications of the shifting balance
ofentrepreneurial exit modes away from initial public offerings
andtoward mergers and acquisitions (M&As). Panel A of Figure 1
plotsthe ratio of deals (and deal value) from VC-backed M&As to
IPOsover the 1992–2007 time period. The same data series are
plottedfor VC-backed biotechnology firms (the industry subject of
thisstudy) in panel B of Figure 1. Acquisitions have clearly
outstrippedIPOs as the modal form of entrepreneurial exit. Although
assessingthe welfare implications of this shift is beyond the scope
of thispaper, the innovation consequences are a key component to
suchan analysis.
867
Dow
nloa
ded
from
info
rms.o
rg b
y [1
65.1
23.3
4.86
] on
14 Ju
ly 2
014,
at 1
1:59
. Fo
r per
sona
l use
onl
y, a
ll rig
hts r
eser
ved.
-
Aggarwal and Hsu: Entrepreneurial Exits and Innovation868
Management Science 60(4), pp. 867–887, © 2014 INFORMS
precedents to follow; we would have had a goodreturn on the
investment. That would be that. Butthey didn’t. So we gave up that
idea and decided topursue the public issue” (Perkins 2002). On the
likelyconsequences of such a buyout, Genentech’s CEO inthe early
1990s, Kirk Raab speculated: “Genentechwould not be the wonderful
place it is today if somelarge pharmaceutical firm had bought us 0
0 0and likean amoeba absorbed us, which is what big companiesoften
do” (Raab 2003).This anecdote exemplifies a key difficulty in
design-
ing a study investigating the innovation consequencesof
entrepreneurial liquidity mode: the possible issueof self-selection
into mode based on unobserved fac-tors. Clearly, the gold standard
of random assign-ment of ventures to exit mode is not available.
Notonly is being in the position to consider a liquid-ity event (of
any sort) not a random occurrence,the choice between exit modes may
be importantlyinfluenced by unobserved factors. While we recog-nize
that disentangling the comingling of exit modeselection and
treatment effects is challenging, weemploy three approaches enabled
by our panel dataset of the universe of VC-funded U.S.
biotechnol-ogy start-ups founded between 1980 and 2000. First,we
employ a coarsened exact matching (CEM) algo-rithm to our data to
define more closely aligned treat-ment and control samples. Second,
we conduct aquasi-experiment in which we compare the innova-tion
profiles of firms experiencing a given exit eventto subsamples of
firms that “nearly” experienced theevent, but for reasons unrelated
to innovation, didnot complete the exit process. Finally, we
employan instrumental variables strategy centered on therelative
liquidity of alternative exit channels in thebiotechnology
industry.Across the range of our comparisons, we find a
decline in innovation quality (as measured by patentcitations)
as a causal effect of both the IPO and M&Atreatments, with the
IPO effects larger in magnitude.Although the quantity of
innovations (as measuredby patent counts) also declines following
an IPO, wefind an increase in this measure following an
M&A.These results are consistent with an information
con-fidentiality mechanism, in which different levels ofinformation
disclosure associated with alternative exitmodes influence
innovation rates (going public entailsthe largest information
disclosure, while remainingprivately held involves the least, with
being acquiredin-between). We conduct within-exit mode analysesto
sharpen our evidence for this mechanism. Forfirms going public,
there is a significant negativeinteraction on innovation quality
between stock mar-ket analyst attention and the level of
preclinical trialproducts firms have in their pipeline. For
biotechnol-ogy firms, the veil of secrecy may be most important
during the preclinical phase of drug development,and the
interaction with analyst coverage is con-sistent with an
information disclosure mechanism.Furthermore, among acquired firms,
we find beingacquired by a private rather than a public
acquirer(the latter associated with higher information dis-closure)
results in higher innovation quality amongM&As. In addition,
our results point to an importantrole for managerial incentives in
M&As: greater tech-nology overlap between the acquiring and
acquiredfirms boosts patent quantity but reduces quality,
sug-gesting that in more competitive settings, once thefirm becomes
part of another organization, acquiredfirm managers prefer
short-run observable outcomes(patent quantity) at the expense of
outcomes that maynot be observable until the longer run (patent
quality).Finally, we investigate the extent to which
inventorturnover following liquidity events might account forthese
empirical patterns by constructing an inventor-year panel data set
covering inventor histories bothin and out of sample with regard to
our focal firms.We find that the inventor-level turnover effects
cannotexplain the firm-level patterns, which are instead
con-sistent with information confidentiality mechanisms.
2. LiteratureA key precondition to the entrepreneurial
choiceamong exit modes is building a significant businessto warrant
further expansion. Conditional on this,there have been just a few
papers, to our knowl-edge, that deal with this choice; these papers
suggestfour categories of explanatory factors. In the contextof
significant VC involvement, a first set of expla-nations suggests
that financing contractual designcan influence exit outcomes,
because VCs negotiatecertain control rights based on their
assessment ofentrepreneurial quality (e.g., Hellmann 2006, Cum-ming
2008). A second set of explanations centers onindustry or market
characteristics, such as the indus-try degree of leverage and
concentration, or publicequity hotness (e.g., Brau et al. 2003;
Bayar and Chem-manur 2011, 2012). A third set of explanations
relatesto the role of firm and product market characteristics,such
as growth potential, capital constraints, degreeof information
asymmetry, and complementarity withthe potential acquirer (e.g.,
Poulsen and Stegemoller2008; Bayar and Chemmanur 2011, 2012).
Finally,founder characteristics, most notably
entrepreneurialpreferences for control versus value creation, can
playa role. Schwienbacher (2008) argues in a theoreti-cal model
that because entrepreneurs value control,which is more likely under
an IPO exit, they aredriven to be more innovative to reduce the
likelihoodof being acquired.
Dow
nloa
ded
from
info
rms.o
rg b
y [1
65.1
23.3
4.86
] on
14 Ju
ly 2
014,
at 1
1:59
. Fo
r per
sona
l use
onl
y, a
ll rig
hts r
eser
ved.
-
Aggarwal and Hsu: Entrepreneurial Exits and InnovationManagement
Science 60(4), pp. 867–887, © 2014 INFORMS 869
Figure 1 Relative Intensity of M&As to IPOs, 1992–2007
Panel B: VC-backed biotechnology
0
5
10
15
20
25
1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004
2005 2006 2007
Ratio of M&A/IPO amount raised
Ratio of M&A/IPO number of deals
0
1
2
3
4
5
6
7
8
9
1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004
2005 2006 2007
Ratio of M&A/IPO amount raised
Ratio of M&A/IPO number of deals
Panel A: All VC-backed start-ups
Source. Dow Jones/VentureSource.Note. Data for M&A deal
value for panel B is unavailable for 1992–1999.
Although there is a limited but growing literatureexamining the
entrepreneurial choice among multi-ple exit modes, the paper by
Schwienbacher (2008)is the only one, to our knowledge, that aims
tolink this choice directly to entrepreneurial innova-tion.
Although we do not believe that any empir-ical study has addressed
this topic, there are twomechanisms through which this choice might
impactinnovation: a first mechanism relates exit mode inno-vation
outcomes to project selection incentives underdifferent ownership
regimes; a second mechanismrelates to whom information is revealed
under var-ied ownership structures to innovation outcomes.
Although both relate innovation to the degree ofinformation
confidentiality the enterprise is able toretain without disclosing
to various parties, we dis-cuss each mechanism and its associated
empiricalimplications separately because each operates in a
dif-ferent way.
2.1. Organizational Ownership, Project Selection,and
Innovation
Under private ownership, the classic agency issuesassociated
with the separation of ownership and cor-porate control are
typically not as severe, becausethe (concentrated) insiders are
also the managers.
Dow
nloa
ded
from
info
rms.o
rg b
y [1
65.1
23.3
4.86
] on
14 Ju
ly 2
014,
at 1
1:59
. Fo
r per
sona
l use
onl
y, a
ll rig
hts r
eser
ved.
-
Aggarwal and Hsu: Entrepreneurial Exits and Innovation870
Management Science 60(4), pp. 867–887, © 2014 INFORMS
By contrast, under public ownership, because of theexpanded
number of possible shareholders, the reg-ulatory requirements
associated with going publicinclude regular public disclosures on
firm opera-tions. These disclosures may have innovation effects.If
managers know that they will have to report projectstatus on a
regular basis, they may be incentivizedto select projects that are
more likely to yield steadyprogress. Developing important
innovations, how-ever, is a process that not only involves a
longertime horizon, but also offers returns with highervariance
relative to more certain investment activi-ties. Moreover,
innovation often requires experimen-tation, which may be curtailed
if managers know theyhave to report results on a quarterly basis.
As such,for situations in which managers want to
incentivizeexploratory (rather than exploitative) behavior,
pri-vate rather than public firm ownership might be opti-mal
(Ferreira et al. 2012). The empirical results ofLerner et al.
(2011) are consistent with these ideas. Inthat study, the authors
use the private equity contextto evaluate whether firms’ innovation
profiles changeas a result of being acquired via buyout, finding
anoverall increase in the innovative output of
privateequity-acquired firms over the long term (as a result,going
private from a publicly held status improvesinnovation outcomes).
Therefore, information disclo-sure to a broad audience under public
ownershipcan negatively impact innovation quantity and qual-ity by
reducing the tolerance for failure (Manso 2011,Ferreira et al.
2012).With regard to acquisitions, whereas in con-
cept there are synergies of personnel and organiza-tions that
should benefit the acquisition target (theentrepreneurial firm),
the act of merging, typicallyinto a larger organization, can impose
costs that mightdampen innovation. Seru (2013) argues that as a
divi-sion within a conglomerate, the acquired firm mayhave skewed
managerial incentives to oversell thetrue prospects of a given
technology in an effort toacquire more resources for the business
unit (or to tar-get projects with near-term as opposed to
longer-termpayoffs). The result is that managers in the
conglom-erate are less willing to fund innovative projects inthe
first place, because they are not able to assess thetrue quality of
projects.We would therefore expect the following order-
ing of innovation outcomes associated with projectselection
incentives resulting from varied ownershipstructures: privately
held would be ahead of theother two exit modes of publicly held and
acquisi-tion with regard to innovation quality. There is evi-dence
in the literature consistent with this ordering,but little or no
within-industry evidence taking intoaccount the full spectrum of
entrepreneurial liquidityoptions, while also addressing issues of
self-selection
into exit mode. There have, however, been a fewefforts to order
innovation outcomes by ownershipstructure on a pairwise basis
(publicly held versusprivately held and acquisition versus
privately held)while taking into account possible selection
effects.For example, research contemporaneous with ourstudy
suggests that firms pursuing an IPO realize adecline in the quality
of their innovations, largely dueto skilled inventor departures and
post-IPO produc-tivity decreases (Bernstein 2012). However, the
samestudy finds that more entrenched managers experi-ence a smaller
decline in innovation productivity. TheBernstein (2012) study
complements our own by eval-uating a multi-industry context, with a
focus solelyon the IPO mode of exit (and so is unable to assesshow
acquisitions fit in comparatively). In addition,in a study using
the medical device industry as theempirical context, Wu (2012)
finds similar post-IPOeffects as Bernstein (2012) does, with
respect to inno-vation quality (a decrease in patent impact
follow-ing an IPO), but at the same time finds an increasein the
quantity of patents after an IPO (in contrastwith Bernstein (2012),
who finds no effect of the IPOtreatment on this same metric).
Likewise, for acquisi-tions, Seru (2013) finds lower patent grants
and for-ward citations following acquisition as compared
toexogenously uncompleted acquisitions, especially forfirms with
active internal capital markets.3 Of course,studies comparing only
one liquidity mode to pri-vate ownership cannot estimate the
relative orderingof expected outcomes among a broader set of
alter-natives in a causal way, which is our objective in
thispaper.Together, these studies point to the likely impor-
tance of factors determining within-event hetero-geneity, as
well as the need to examine multipledimensions of innovative
outcomes—e.g., patentquantity and quality. Overall, according to
thisfirst information confidentiality mechanism of
projectselection, private ownership appears to dominateIPOs and
acquisitions with regard to innovation out-put, though the latter
two exit modes are not clearlyordered among themselves in this
regard.
2.2. Organizational Ownership, InformationDisclosure, and
Innovation
Under private ownership, details of a product or ser-vice
innovation can more likely remain hidden from
3 According to this “dark side” explanation of internal capital
mar-kets of conglomerates, however, it would seem that business
unitmanagers (including those acquired) would have incentives
tooverrepresent their innovation potential as measured by
innovationquantity, even if doing so may be at the cost of
developing higher-quality inventions. Seru (2013) does not find
this effect, though wedo in our empirics.
Dow
nloa
ded
from
info
rms.o
rg b
y [1
65.1
23.3
4.86
] on
14 Ju
ly 2
014,
at 1
1:59
. Fo
r per
sona
l use
onl
y, a
ll rig
hts r
eser
ved.
-
Aggarwal and Hsu: Entrepreneurial Exits and InnovationManagement
Science 60(4), pp. 867–887, © 2014 INFORMS 871
potential competitors. For acquired firms, such infor-mation is
only disclosed to a small set of outsiderswho are evaluating the
firm as a suitable acquisitiontarget.4 By contrast, publicly held
organizations mustroutinely make public disclosures, which can
provideimportant information to organizational outsiders.Consider
the following quotes regarding the biotech-nology industry: “The
biotech industry is fiercelycompetitive and disclosure costs are
generally highbecause most companies develop only a few
products,and the entrance of a competitor poses a serious sur-vival
threat” (Guo et al. 2004, p. 320). Furthermore, asPerkins (2002)
noted in regard to possible disclosureof contractual details with
Eli Lilly around the timeof the company’s IPO, “We figured that our
competi-tors would try to ferret out the details of those
con-tracts. They were literally inked out in the SEC
files.”Entrepreneurs therefore sacrifice the opportunity tooperate
“under the radar” with respect to announc-ing their offerings, in
exchange for liquidity and otherbenefits of a public offering.
Nevertheless, the deci-sion to go public likely involves a
trade-off betweenearly liquidity and the risks of information
disclo-sure to product market competitors, as the theoreticalmodels
of Bhattacharya and Ritter (1983), Maksimovicand Pichler (2001),
and Spiegel and Tookes (2007)suggest.In an empirical analysis of
U.S. manufacturing
firms, Chemmanur et al. (2012) build on these models,finding
that product market characteristics can drivefirms’ choice of exit
mode in ways that are consis-tent with predictions based on the
relative degree ofexpected information confidentiality under
alternateownership structures (private, M&A, and public).
TheChemmanur et al. (2012) study examines a cross-industry sample
of manufacturing firms and findsevidence for a greater decrease in
total factor produc-tivity following an IPO as compared to an
acquisi-tion, consistent with the mechanism of
informationconfidentiality. This study provides
complementaryinsights to ours, with our study in the context
ofentrepreneurial biotechnology firms differing in itsemphasis on
innovative output, as compared to pro-duction and product market
characteristics.A small, related literature is the connection
between corporate governance and innovation out-comes. With
private ownership, in addition to innateentrepreneurial preferences
or benefits associatedwith control, less distributed control rights
allow
4 If the acquirer is publicly held, however, the transaction
couldreceive more scrutiny by antitrust authorities and/or
shareholdersof the acquiring firm (in which case there would be
more infor-mation disclosed to a broader audience). We exploit this
within-acquisition event heterogeneity in our empirics to sharpen
ourempirical evidence beyond across exit mode innovation
orderingfor this information confidentiality mechanism.
entrepreneurs to retain relative autonomy in mak-ing decisions
in the face of differences of opinionwith outsiders (Boot et al.
2006). The net impact ofconcentrated versus more distributed
ownership (aswould be the case with public ownership) on
innova-tion, however, is theoretically ambiguous because itdepends
on the relative productivity differences asso-ciated with more
versus less concentrated corporategovernance. Typically, the
corporate board of direc-tors expands in the ramp-up to an IPO
(Baker andGompers 2003). Unfortunately, there is little
literatureon the direct impact of expanded boards or of
tightercorporate governance more generally on innovation.Whereas
earlier literature found a negative relation-ship between
antitakeover provisions and innovationinvestments (e.g., Meulbroek
et al. 1990), a recentstudy (O’Connor and Rafferty 2012) finds no
rela-tion between broad measures of corporate governanceand
innovation levels once simultaneity is taken intoaccount in their
empirical models.Taken together, this second information
confiden-
tiality mechanism, focusing on to whom informationis disclosed
under different ownership structures, pre-dicts acquisitions as
middling in innovation perfor-mance, with better outcomes than
going public andworse outcomes relative to remaining private.
3. Methodology3.1. OverviewExamining the causal implications of
alternate exitmode choices requires a methodology that takes
intoaccount possible self-selection of firms into particu-lar modes
based on unobserved factors. In additionto our aim of drawing
causal inferences on the effectsof exit mode treatments, we also
seek to frame ourresults in the context of the prior literature. As
dis-cussed in the previous section, there are two streamsof work
related to information confidentiality that arepotentially helpful
in understanding the mechanismsat work: altered project selection
and disclosures toexternal parties, both of which operate through
man-agerial channels. Although these two mechanisms areconceptually
distinct, they yield similar predictionswith regard to the
relationship between ownershipstructure and innovation patterns. As
a result, wewill not be able to untangle the mechanisms
empiri-cally, especially because the first mechanism of
projectselection from a choice set of alternatives is unob-servable
to us. Nevertheless, the two mechanismsof information
confidentiality imply an ordering ofinnovation outcomes across
ownership modes andsome empirical patterns within exit mode. Our
anal-yses are accordingly structured to test the empiricalsalience
of the two information confidentiality mecha-nisms. To the degree
that our effects might be alterna-tively explained solely through
inventor-level changes
Dow
nloa
ded
from
info
rms.o
rg b
y [1
65.1
23.3
4.86
] on
14 Ju
ly 2
014,
at 1
1:59
. Fo
r per
sona
l use
onl
y, a
ll rig
hts r
eser
ved.
-
Aggarwal and Hsu: Entrepreneurial Exits and Innovation872
Management Science 60(4), pp. 867–887, © 2014 INFORMS
(rather than managerial-level effects associated withproject
selection incentives or direct information dis-closure), however,
we supplement our firm-level anal-yses with inventor-level
analyses, examining the roleof inventor movements and inventor
productivityaround exit events.
3.2. SampleWe sample the universe of VC-funded biotechnol-ogy
firms founded between 1980 and 2000, identi-fying these firms using
the VentureXpert database.We focus on start-ups receiving venture
capital fund-ing because the quality screen of VC
involvement(Kortum and Lerner 2000) offers a desirable dimen-sion
of homogeneity among firms in the sample, withliquidity needs
arising from the venture capital cycle(Gompers and Lerner 2004,
Inderst and Muller 2004)creating pressures to pursue exit
opportunities. A sec-ond desirable dimension of homogeneity is the
useof biotechnology as the industry context. The impor-tance of
patenting to the appropriation and valuationof innovations is
particularly important in biotechnol-ogy relative to other sectors
(e.g., Levin et al. 1987).A single-industry context enables us to
obtain rele-vant measures of the value and importance of
inno-vations, an objective that would be significantly
morechallenging in a multi-industry setting. We focus onfirms
founded in the 21-year period between 1980 and2000 to ensure that
our results are generalizable acrossa range of initial industry
conditions, as well as toensure that we can observe firm outcomes
for a suf-ficiently long period of time postfounding. The sam-ple
consists of the 476 U.S.-based firms in the humanbiotechnology
industry (Standard Industrial Classifi-cation codes 2833–2836)
founded during these years.The primary data set is structured as an
unbalanced
firm-year panel, with observations for each firm start-ing with
the year of founding. Because the mostrecent founding year is 2000,
and the data are col-lected through 2006, we observe each firm for
a min-imum of seven years, except in cases where the firmis
dissolved prior to 2006.5 Our data set thus includesobservations at
the firm-year level for each year inwhich the firm is in operation,
including those yearsfollowing an exit event (which can be either
an IPO oran M&A). We do not, however, include observationsfor
those years after which a firm ceases to exist asa consequence of a
dissolution event. Left-censoringis not an issue because we observe
firms beginningwith their date of founding. The final observation
year
5 The average life span of a venture fund during this time frame
is8 to 10 years and so VC-backed firms in this industry thus
havestrong incentives to pursue an exit event within 5 to 7 years
post-founding.
of 2006 is chosen in accordance with our use of for-ward
citations as one of our two measures of inno-vative output
(described in more detail in §3.4), forwhich we utilize a four-year
postapplication observa-tion window. In addition to the firm-year
panel, weassemble an inventor-year panel data set (describedin more
detail in §3.9) to understand the role of indi-vidual inventors in
influencing our results.We utilize several archival sources to
assemble our
data sets. For exit events, this includes news arti-cle searches
from Factiva, combined with data fromThomson One Banker, Zephyr,
and U.S. Securities andExchange Commission (SEC) filings. For
measures ofinnovation, we draw on the IQSS Patent Networkdatabase
(see Lai et al. 2011 for a description), whichincorporates the U.S.
Patent and Trademark Officedata on all patents applied for since
1975. This allowsus to construct patent-based measures of
innovationoutput at the firm-year level, and in addition, to
iden-tify unique inventors associated with these patents,thereby
enabling the construction of inventor careerhistories. We also
collect data on firms’ VC fundinghistories, strategic alliances,
product pipelines, as wellas (for post-IPO firms) coverage from
stock marketanalysts. These data draw, respectively, on the
fol-lowing sources: VentureXpert, Deloitte Recap
RDNA,Pharmaprojects and Inteleos, and I/B/E/S. Finally, toconstruct
an instrument for the level of “heat” in theIPO market relative to
the M&A market, we collectdata on IPO and M&A market volume
from multi-ple sources, including Jay Ritter’s IPO data website6and
SDC.
3.3. Empirical StrategyOur main empirical strategy employs the
CEM pro-cedure (Iacus et al. 2011, 2012) to construct treatmentand
control samples that are balanced on pretreat-ment covariates
(discussed in more detail in §3.8).We use the matched control group
to run, for exam-ple, difference-in-differences estimates of the
treat-ment effect of alternate exit modes.7 We employ twoadditional
empirical strategies on the CEM-matcheddata to mitigate any
additional concerns of bias due tounobserved pretreatment
characteristics: (1) a quasi-experiment based on “near” exit
events—those thatwere started but not completed for exogenous
rea-sons; and (2) an instrumental variables strategy toaddress the
possible endogenous selection of IPO ver-sus M&A liquidity
events. To better understand the
6 This data, updated through 2012, uses the methodology
inIbbotson et al. (1994), with the most recent version found
athttp://bear.warrington.ufl.edu/ritter/ipoisr.htm.7 Recent
examples of studies employing the CEM technique to con-struct
matched control samples include Azoulay et al. (2010) andSingh and
Agrawal (2011).
Dow
nloa
ded
from
info
rms.o
rg b
y [1
65.1
23.3
4.86
] on
14 Ju
ly 2
014,
at 1
1:59
. Fo
r per
sona
l use
onl
y, a
ll rig
hts r
eser
ved.
-
Aggarwal and Hsu: Entrepreneurial Exits and InnovationManagement
Science 60(4), pp. 867–887, © 2014 INFORMS 873
mechanisms driving our results, we then conductwithin-exit mode
analyses, along with an analysis atthe inventor-year level. We
first describe the construc-tion of the various measures in our
firm-year dataset, including innovation outcomes, exit events,
firmcharacteristics, and an instrument for the IPO versusM&A
choice. We then discuss the CEM process weemploy to generate the
matched control samples. Weend the section by detailing how we
construct ourinventor-year data set.
3.4. Innovation OutcomesWe begin with our measures of
innovation, for whichwe utilize patent data. To identify all
patents asso-ciated with the firms in our sample, we first
extractfrom the IQSS Patent Network database (Lai et al.2011) all
patents applied for between 1975 and 2010where the “assignee” name
matches our focal firms’current or former name(s). To ensure that
we are com-prehensive in our data collection process, we conductthe
search using an algorithm that matches variouspermutations of the
company name (e.g., we wouldcode patents from “Amgen” and “Amgen
Inc.” asbeing associated with the same firm). The patentnumbers we
collect for our focal firms enable us tocollect a range of other
patent-based characteristicsincluding forward citations and patent
classes. Oursample of 476 firms includes 15,439 patents and
45,789forward citations associated with these patents.Identifying
patents for firms undergoing an M&A
exit raises the issue that post-M&A patent applica-tions
associated with inventions of the acquired firmmay be made with the
acquirer listed as the assignee.As a consequence, it may be
difficult to track the inno-vation outcomes of firms after an
acquisition, unlessthe acquired firm operates as an independent
entity,with future patents accruing to the subsidiary ratherthan to
the parent. We use an inventor-matchingalgorithm to address this
issue. We first assemble adatabase of inventors associated with
preacquisitionpatents applied for by the focal (acquired) firm.
Wethen search patent applications where the acquirer isthe assignee
during the postacquisition period, andconsider patents from this
set of inventors as havingoriginated from the acquired firm. Thus,
the list ofpatents for a focal firm in our sample undergoing
anM&A includes those patents associated directly withthe
acquired firm before and after the acquisition, aswell the subset
of the acquiring firm’s patents thatwere invented by the acquired
entity (i.e., the focalfirm) after the acquisition.8
8 We verify that all of our results are robust to excluding all
acquir-ers who assign any post-M&A patents to the corporate
parent orother entity. We thank an anonymous reviewer for
suggesting thisrobustness test.
We utilize two measures of patent-based innova-tion output:
patent applications and forward citations.These two characteristics
of firm-level output proxyfor the quantity and quality of
innovation, respec-tively. Prior work (Trajtenberg 1990) suggests,
more-over, that forward citations in particular have a
strongcorrelation with economic value. We define the firm-year
variables patent applications stock as the numberof patent
applications applied for by the firm up toand including the
firm-year, and forward patent cita-tions four years stock as the
number of patent cita-tions within a four-year postissue window to
patentsapplied for (and subsequently granted) by the focalfirm up
to and including the firm-year.9 We measureboth through 2006 (the
forward citations window con-straints our final observation
year).10
3.5. Exit EventsWe observe variation in the modes by
whichentrepreneurs and their stakeholders achieve exit.From the
time of founding, each firm can undergomultiple exit or “near-exit”
events (those for whichthe process was begun, but never
consummated). ForM&A events we are concerned specifically with
sit-uations in which the focal firm is the target in theacquisition
(thereby creating a liquidity event for thefounders and investors).
We conduct an exhaustivearchival search using news articles from
Factiva, trian-gulated with Thomson One Banker, Zephyr, and
SECfilings, to identify realized exit events for our focalfirms
(from founding through 2006). We utilize in ourspecifications a set
of indicator variables for subsam-ples of firms that underwent an
IPO or M&A, as wellas indicator variables for the three-year
period of timefollowing the IPO or M&A. These latter
variables—focal, post-IPO (113) and focal, post-M&A
(113)—allowus to obtain difference-in-differences estimates of
theIPO and M&A effects on our CEM-matched sample,as we discuss
in detail in §4.11In addition to identifying realized exit events
from
our archival data search, we also identify those exitevents that
were “withdrawn” in the sense that theexit process started but was
never taken to com-pletion. For IPOs, a withdrawn event
represents
9 We also examine the robustness of our results to using our
for-ward citation measure less self-citations (the two versions of
thevariable are pairwise correlated at 92%). Removing
self-citationsstrengthens the results, and so we report the more
conservativefull-forward citations in our empirical tables.10 We
use the stock versions of these variables because we believethese
have a more natural interpretation given our
difference-in-differences approach with firm fixed effects and firm
age controls(as discussed later); our results are, however, robust
to alternativelyusing the flow measures.11 We also researched the
incidence of publicly held firms beingtaken private (as in the
Lerner et al. 2011 study). Among our samplecompanies, we did not
find a single such case.
Dow
nloa
ded
from
info
rms.o
rg b
y [1
65.1
23.3
4.86
] on
14 Ju
ly 2
014,
at 1
1:59
. Fo
r per
sona
l use
onl
y, a
ll rig
hts r
eser
ved.
-
Aggarwal and Hsu: Entrepreneurial Exits and Innovation874
Management Science 60(4), pp. 867–887, © 2014 INFORMS
situations in which the firm filed for an IPO butsubsequently
did not go public due to exogenousmarket conditions. Withdrawn
M&A events representsimilar situations, in which a deal was
announcedbut never consummated. These two sets of eventsenable us
to conduct a quasi-experiment to identifythe treatment effect of
exits (IPO or M&A) using sub-samples that pool realized-exit
and near-exit events(IPO/near-IPO in one case and
M&A/near-M&A inthe other). An assumption of this approach
is thata firm’s withdrawal from a previously planned exitevent is
uncorrelated with its innovation capacity andwith other firm-level
characteristics. For withdrawnIPO events, we verify through news
articles that thewithdrawal is a function of unstable or volatile
mar-ket conditions, factors exogenous to our model spec-ifications.
For withdrawn M&A events we similarlyverify that withdrawals
are due to shareholder objec-tions or to regulatory oversight.12
Furthermore, weregress the likelihood of (IPO or M&A)
withdrawalon our innovation variables and our full set of
(time-lagged) firm characteristics (described later), and findall
effects to be insignificant. This increases our con-fidence that
exit withdrawals are not systematicallyrelated to either innovation
or firm characteristics.Our firm-year data set is structured to
account for
the fact that a firm can undergo multiple “near”-and
“realized”-exit events throughout its lifetime. Wecode the full
history of such events, and can there-fore observe situations
where, for example, the firmexperiences a withdrawn IPO or M&A
event, andsubsequently exits via one of these modes. Similarly,we
can observe situations in which one mode of exit(e.g., an IPO) is
followed by another (an M&A). Oneadditional category of
firm-level outcomes is the com-plete dissolution, or liquidation,
of a firm in our sam-ple. Such situations differ importantly from
our twomodes of exit (IPO and M&A) in that the firm ceasesto
exist as a going concern and can thus no longercontinue its
innovation output. For firms that are dis-solved, we use the year
of dissolution as the finalobservation year for the firm in our
firm-year paneldata set.In addition to the indicator variables for
differ-
ent exit modes and the three-year postexit windows,we utilize
two additional exit event-related measuresthat are specific to the
subsample of acquired firms.First, we create an indicator variable
for whether theacquiring entity is private (the private dummy).
Sec-ond, following Jaffe (1986), we define technology over-lap as
the angular separation between the primaryU.S. patent class vectors
of the acquiring and acquired
12 Although we were able to confirm that the reasons for
M&Awithdrawal were externally driven, we cannot determine
whetherthe decision to withdraw came from the target or from the
acquirer.
(focal) firms. Each vector has a dimension of 987 andis indexed
by unique patent classes; a given valuewithin a vector represents
the proportion of the firm’sstock of patents (applied for prior to
and until thedate of acquisition) assigned to the patent class
asso-ciated with the index for that value. The technologyoverlap
measure is the angular dot product of thetwo vectors: a value of 1
represents vectors with per-fect overlap, whereas a value of 0
represents orthog-onal patent class vectors. We interact both the
privatedummy and the technology overlap measure with thefocal,
post-M&A (113) indicator variable to examinethe role of
particular organizational mechanisms ininfluencing innovation
output within the M&A modeof exit.
3.6. Firm CharacteristicsWe employ a set of firm-level controls
to account forany residual time-varying unobserved heterogeneityin
our models (we utilize firm fixed effects in mostspecifications).
To account for firm-level quality andlife-cycle considerations we
use firm age, which is theage of the firm since founding, along
with VC inflowsstock, which measures the cumulative amount of
VCfunding received by the firm through the currentfirm-year
(collected using VentureXpert). In addition,we use the Deloitte
Recap RDNA database to collectdata on the cumulative stock of
strategic alliances afirm has entered up to the current firm-year,
strategicalliance stock, a further measure of firm quality
(e.g.,Stuart et al. 1999).13In addition to age, VC funding, and
strategic
alliances, we use a firm’s product portfolio as afinal
firm-level characteristic. In the empirical con-text of
biotechnology, a relevant metric for prod-uct development is the
stage of an individual drugcompound in the U.S. Food and Drug
Administra-tion (FDA) approval process. To construct our
twoproduct-related measures, we utilize the Inteleos
andPharmaProjects databases to compile the number ofproducts each
firm has at different stages of devel-opment in a given firm-year.
We track the trajec-tory of an individual drug compound over timeby
combining Inteleos, for which we have data foryears 1990–2001, and
PharmaProjects (which we useto collect 2002–2006 data, matching
these with drugcompounds identified in Inteleos).14 We measure
the
13 Firms’ strategic alliance stock is correlated with VC inflows
stockat the 66% level, and so in the empirical tables we only use
thelatter variable, although the results are robust to using the
formervariable instead.14 We compile product pipeline data only for
firms founded post-1989 because of time-period coverage limitations
associated withthese two data sources. However, since our unit of
analysis is anindividual drug compound as it moves through the FDA
approvalprocess, we are able to track product portfolios
post-M&A as well.
Dow
nloa
ded
from
info
rms.o
rg b
y [1
65.1
23.3
4.86
] on
14 Ju
ly 2
014,
at 1
1:59
. Fo
r per
sona
l use
onl
y, a
ll rig
hts r
eser
ved.
-
Aggarwal and Hsu: Entrepreneurial Exits and InnovationManagement
Science 60(4), pp. 867–887, © 2014 INFORMS 875
number of products in a given firm-year at four stagesof the FDA
approval process: preclinical, stage 1,stage 2, and stage 3. Our
measure of early-stage inno-vations, preclinical products, enables
us to test the con-ditions under which information disclosure may
bemost significant. In addition, as an aggregate measureof a firm’s
product portfolio value in a given firm-year (which we use as a
control variable), we con-struct a measure, weighted products,
which weights thenumber of products based on their stage, putting
arbi-trary values of 1, 2, 5, and 10, respectively, on the
fourdevelopment stages, reflecting the relative degree ofeconomic
value of the firm’s portfolio based on thelikelihood of eventual
product commercialization (ourresults are similar with unweighted
counts of firmproduct portfolios).Finally, for the subsample of
firms that undergo an
IPO, we collect from I/B/E/S a measure of stock mar-ket analyst
coverage, analyst reports, that measures thetotal number of analyst
reports published about thefirm in the firm-year. Prior studies
have discussedthe role that analysts play in influencing both
infor-mation availability and incentive structures, whichcan
influence innovation (Chemmanur et al. 2012,Ferreira et al. 2012,
He and Tian 2013). We thus usethis variable, which measures the
degree of scrutinyon the firm by outside parties, to examine the
infor-mation confidentiality mechanism in our sample offirms that
have gone public.
3.7. Instrumental VariableAs discussed previously, one component
of our strat-egy for addressing the possibility of unobserved
self-selection into exit mode involves instrumenting forthe
endogenous selection between the IPO and M&Amodes of exit. We
utilize as our instrument the rela-tive level of “heat” in one
market as compared to theother within the biotechnology industry.
Prior litera-ture has typically used volume-based measures of
IPOmarket heat. Yung et al. (2008), for example, definemarket heat
in two ways: first, by comparing thefour-quarter moving average to
the historical quar-terly volume; and second, by examining IPO
marketunderpricing relative to the historical average. Whileother
studies of IPO market heat utilize variants ofthis approach, the
commonality is using volume-basedmeasures (e.g., Helwege and Liang
2004). For M&As,“merger waves” are an analogous concept to “hot
mar-kets” in IPOs (e.g., Harford 2005), and in this casetransaction
volume is similarly used as the key met-ric. We thus focus on
volume in the IPO and M&Amarkets because this offers an
approach to measur-ing market heat that is common to both markets.
Webuild on the methodology used in Yung et al. (2008) todevelop our
metric for relative IPO market attractive-ness. Using IPO volume
data from Jay Ritter’s website,and M&A volume data from SDC, we
first identify
the number of quarters in each firm-year where thefour-quarter
moving average of biotechnology IPO (orM&A) volume is 25% above
the quarterly averagefrom the prior five years. We then construct a
measureof IPO relative to M&A market heat (IPO versus
M&Abiotechnology industry liquidity) by taking the ratio ofthe
IPO measure to the M&A measure. When we dis-cuss our results
using this instrumental variable (IV)in the next section, we will
also discuss how the IV isboth correlated with the possibly
endogenous variablebut satisfies the exclusion restriction by being
unre-lated to firm-level innovation outcomes.
3.8. Coarsened Exact Matching ProcedureTable 1 summarizes the
definitions and descrip-tive statistics of the measures used in our
analyses.In Table 2, we show that the CEM procedure helpsbalance
the preevent subsamples, which we use asthe basis for our main
difference-in-differences spec-ifications. As Iacus et al. (2011)
note, CEM is partof a general class of methods termed
“monotonicimbalance bounding” (MIB), which has beneficial
sta-tistical properties as compared to prior “equal per-cent bias
reducing” (EPBR) models (Rubin 1976), ofwhich propensity score
matching and Mahalanobisdistance are examples.15 MIB generalizes
the EPBRclass, eliminating many of the assumptions requiredfor
unbiased estimates of treatment effects, and out-performing EPBR in
most situations, including thosespecifically designed to meet the
EPBR assumptions(Iacus et al. 2011, 2012). A key difference in
practicelies in the sequence of data preprocessing: whereasmethods
such as propensity score matching (PSM)require determining ex ante
the size of the matchedcontrol sample, then ensuring balance ex
post, CEMperforms the balancing ex ante (Iacus et al. 2012).CEM
entails “coarsening” a set of observed covari-ates, performing
exact matching on the coarseneddata, “pruning” observations so that
strata haveat least one treatment and one control unit, thenrunning
estimations using the original (but pruned)uncoarsened data
(Blackwell et al. 2009).A key goal of any matching process is to
ensure that
the treated and control groups are “balanced” in thesense that
their covariates have (approximately) equaldistributional
characteristics. Table 2 shows the out-come of our application of
the CEM process. We focuson four key pretreatment observables, age,
VC inflowstock, strategic alliance stock, and weighted
products,
15 In summarizing a series of analytical and numerical tests of
theCEM method, Iacus et al. (2011, p. 359) note, “[CEM] 0 0
0generatesmatching solutions that are better balanced and estimates
of thecausal quantity of interest that have lower root mean square
errorthan methods under the older existing class, such as based
onpropensity scores, Mahalanobis distance, nearest neighbors,
andoptimal matching.”
Dow
nloa
ded
from
info
rms.o
rg b
y [1
65.1
23.3
4.86
] on
14 Ju
ly 2
014,
at 1
1:59
. Fo
r per
sona
l use
onl
y, a
ll rig
hts r
eser
ved.
-
Aggarwal and Hsu: Entrepreneurial Exits and Innovation876
Management Science 60(4), pp. 867–887, © 2014 INFORMS
Table 1 Descriptive Statistics and Variable Definitions
(Firm-Year Level of Analysis)
Variable Definition Mean Std. dev.
Dependent variablesPatent applications stock Stock of patent
applications to firm i in year t 17017 57004Forward patent
citations
four years stockForward patent citations to firm i ’s stock of
patents within four years of patents granted in year t 55001
156045
Independent variablesEvent and time variables
Focal IPO sample Dummy= 1 only for all firm-years (pre- and
postevent) associated with a firm undergoing an IPO 0048 0050Focal,
post-IPO window Dummy= 1 for the time window one to three years
(inclusive) after the IPO event 0008 0027IPO year indicator Dummy=
1 only for the year in which a firm undertook an IPO 0004 0020Focal
M&A sample Dummy= 1 only for all firm-years (pre- and
postevent) associated with a firm undergoing an M&A 0040
0049Focal, post-M&A window Dummy= 1 for the time window one to
three years (inclusive) after the M&A event 0007 0026Focal,
post-M&A window,
private acquirerInteraction term for the time window one to
three years (inclusive) after the M&A event if acquired
by a privately held entity (indicator variable)0004 0020
Focal, post-M&A window,technology overlap
Interaction term for the time window one to three years
(inclusive) after the M&A event with anormalized angular
separation between vectors of primary patent classes of acquired
andacquiring firms (see text; formula follows Jaffe 1986)
0010 0026
Biotechnology firm characteristicsAge Age in years of the focal
firm as of year t 8042 6012VC inflows stock Cumulative VC inflows
invested in the focal firm to year t (in $M) 16039 27087Strategic
alliance stock Cumulative number of strategic alliances the focal
firm had entered into as of year t as reported
by Deloitte Recap RDNA10039 17091
Weighted products a Aggregate measure of focal firm’s product
portfolio in year t created by weighting the number ofproducts
along the FDA approval process: preclinical (weighted 1), stage 1
(2), stage 2 (5), andstage 3 (10)
75054 143038
Preclinical products a Number of preclinical products in a
firm-year 1005 3039Analyst reports For firms going public, number
of analyst reports issued on focal firm in year t 61025 128094
Instrumental variableIPO vs. M&A biotechnology
industry liquidityRatio of number of quarters in a focal year in
which the deal volume of IPOs in the biotechnology
industry exceeded by 25% the rolling average over the prior
five-year window to the samecount for M&As
0056 0063
aData compiled only for firms founded post-1989.
creating separate treatment and control samples post-CEM for the
IPO and M&A treatments. The fourvariables we use to balance the
treatment and con-trol samples represent observable quality
dimensionsthat we expect would be correlated with the IPOand
M&A treatments. As the “pre-CEM” columnshows, the IPO and
M&A treatment and control sub-samples are significantly
different (at the 5% level)across the board for the full set of
covariates. Thesedifferences are reduced, however, post-CEM,
withnone of the treatment-control differences significant athigher
than 5%, suggesting balance in the two sets ofsamples.
3.9. Inventor-Year Data SetFinally, although our primary aim is
to empiricallyassess the role of information confidentiality in
therelationship between entrepreneurial exits and inno-vation, an
alternative to such mechanisms based inhuman resource turnover
might instead be a primarydriver of innovation patterns. For
example, Stuart andSorenson (2003) suggest that IPO and M&A
liquid-ity events are organizationally disruptive for the
focalenterprise, and link the geographic distribution ofnew firm
foundings to the regional pattern of such
entrepreneurial liquidity events. They find supportfor an
employee spinoff mechanism behind the empir-ical pattern. More
generally, in acquisitions, there maybe personnel adjustment costs
that can result fromchanges in corporate culture and/or from
turnoverin personnel composition. Similarly, for employeesholding
stock options, IPOs could loosen the bondsof employment for
personnel not subject to lock-uprestrictions. We therefore wish to
assess the degreeto which our firm-year results are wholly
explainedby inventor-level turnover. If they are, the informa-tion
confidentiality mechanisms, which operate at themanagerial policy
level rather than at the inventorlevel, may be less important in
explaining the firm-level empirical patterns. We therefore
construct aninventor-year data set by identifying all
inventorsassociated with patents of our focal firm sample
andconstructing full inventor histories for each of
theseindividuals.These inventor histories include patenting
activities
both within and outside our focal firms,16 with the
16 We track inventor histories starting from 1975 to ensure that
wecapture a sufficient window of history for inventors prior to
theirjoining the focal firm.
Dow
nloa
ded
from
info
rms.o
rg b
y [1
65.1
23.3
4.86
] on
14 Ju
ly 2
014,
at 1
1:59
. Fo
r per
sona
l use
onl
y, a
ll rig
hts r
eser
ved.
-
Aggarwal and Hsu: Entrepreneurial Exits and InnovationManagement
Science 60(4), pp. 867–887, © 2014 INFORMS 877
Table 2 Firm Characteristics Before and After Coarsened
ExactMatching Procedure
Pre-CEM Post-CEM
IPO Control Controlsample sample IPO sample
L Age 2004 1090⇤⇤ 2019 2020400835 400835 400605 400685
L VC inflow stock 2010 1055⇤⇤ 2016 2000410535 410405 410555
410495
L Strategic alliance stock 2011 1010⇤⇤ 2010 2025410185 410035
400795 400775
L Weighted products 1022 0091⇤⇤ 0059 0044420155 410685 410615
410405
M&A Control Controlsample sample M&A sample
L Age 2000 1094⇤⇤ 2047 2050400835 400835 400475 400435
L VC inflow stock 2005 1066⇤⇤ 2044 2037410485 410485 410345
410335
L Strategic alliance stock 1081 1058⇤⇤ 2023 2025410235 410215
400965 400775
L Weighted products 1012 1002⇤⇤ 1003 0099410865 410975 410765
410935
Notes. The mean and standard deviation (in parentheses) are
reported. Thenatural logarithm of a variable, X , is denoted L X .
The CEM procedureinvolves matching on the log values of age, VC
inflow stock, alliance stock,and weighted products.
⇤⇤Indicates difference is significant at the 5% or higher level
compared tothe “treated” sample.
resulting inventor-year data set consisting of 12,769inventors
associated with 15,439 focal firm patents,each observed, on
average, for 11.3 years (the totalnumber of patents within and
outside the focal firmassociated with these inventors is 57,803).
We definethe variables change in (mean: 0.46; s.d.: 0.50) andchange
out (mean: 0.02; s.d.: 0.15) as indicators forwhether a given
inventor either joined or departed afocal firm in a given year. For
inventors joining a focalfirm in our sample, we set the variable
change in toequal 1 in the first year in which the inventor
appliesfor a patent in the focal firm. A departure, capturedby
change out, is identified when an inventor who haspatented in one
of our focal firms is observed to sub-sequently patent outside this
same focal firm. Thisvariable is equal to 1 in the year the
inventor patentsin the “new” firm. We additionally define the
vari-able years since first invention at the inventor-year levelto
reflect the length of the inventor’s career to date.Finally, we
create patent outcome measures similar tothe firm-year measures
discussed previously (patentapplications stock and forward patent
citations four yearsstock), except that these are specific to the
inventorand defined at the inventor-year level.
4. Empirical Results4.1. Post- vs. Preevent ComparisonsWe begin
our analysis in Table 3 with a simple regres-sion analysis of the
innovation patterns for firms thatexperienced an IPO or an
acquisition, comparing thepost- and preevent innovation profiles.
This analysisdoes not confine the sample to observations matchedvia
CEM, because we initially want to describe theinnovation patterns
comparing post- versus preeventsfor the sample of firms undergoing
each event. In sub-sequent analyses, we will adopt methods to
addresspossible selection issues associated with firms of
dif-ferent characteristics choosing liquidity modes. Weexamine two
innovation outcomes throughout ourempirics, patent applications
stock and forward patentcitations four years stock, with the former
measure cor-responding to innovation quantity and the latter aproxy
for innovation quality. We take the log valueof these outcome
variables and run firm fixed effectsOLS regressions on our
firm-year sample. Negativebinomial count models (of unlogged
outcomes) yieldsimilar estimates for the specifications that
convergein estimation. For the sake of consistency throughoutthe
tables, we report OLS results.We first compare the innovation
profiles of the
202 firms in our sample undergoing an IPO in the firstfour
columns of Table 3. The first two columns reportthe effect of being
in the post-IPO period, with thefirst column including no controls
beyond the firmfixed effects and the second adding to the model
avariety of (logged) time-varying firm controls: age, VCinflows
stock, and weighted products. VC inflows stockproxies for
differential firm resource inputs, whereasage and weighted products
aim to control for possibleinnovation rate differences across the
firm and prod-uct life cycle. Chemmanur et al. (2010), for
example,find that IPOs occur at the peak of firms’
productivitycycle. The key independent variable, focal,
postevent(113), is negative and significant in both
specifica-tions, with the estimate in (3-2) suggesting a 36%decline
in patent applications in the three years post-IPO. The analogous
specifications for the forwardpatent citation outcome are contained
in the next twocolumns of Table 3. The only difference is that we
nor-malize these forward patent citations regressions byincluding
the log of patent applications stock as a regres-sor (a structure
we adopt throughout our empiricalspecifications when we analyze
this outcome vari-able). Dropping this normalization does not alter
thestatistical significance of the estimates, although
theindependent variable of interest is typically estimatedwith a
larger coefficient. The key independent vari-able, focal, postevent
(113), is positive, but only sig-nificantly so in specification
(3-4), with the estimatesuggesting a 5% increase in forward patent
citations
Dow
nloa
ded
from
info
rms.o
rg b
y [1
65.1
23.3
4.86
] on
14 Ju
ly 2
014,
at 1
1:59
. Fo
r per
sona
l use
onl
y, a
ll rig
hts r
eser
ved.
-
Aggarwal and Hsu: Entrepreneurial Exits and Innovation878
Management Science 60(4), pp. 867–887, © 2014 INFORMS
Table 3 Post- vs. Preevent Innovation Comparisons (Firm-Year
Level of Analysis) OLS Regression Coefficients Reported
Post- vs. pre-IPO innovation comparisons Post- vs. pre-M&A
innovation comparisons
L Forward patent citations L Forward patent citationsDependent
variable: L Patent applications stock four years stock L Patent
applications stock four years stock
(3-1) (3-2) (3-3) (3-4) (3-5) (3-6) (3-7) (3-8)
Focal, postevent 41135 É10069⇤⇤⇤ É00361⇤⇤⇤ 00022 00049⇤⇤
00508⇤⇤⇤ 00223⇤⇤⇤ É00066⇤⇤⇤ É00073⇤⇤⇤4000545 4000325 4000225
4000235 4000515 4000325 4000205 4000215
Firm-level controls No Yes No Yes No Yes No YesEvent year FE No
Yes No Yes No Yes No YesFirm FE Yes Yes Yes Yes Yes Yes Yes
YesConstant 20339⇤⇤⇤ É00308⇤⇤⇤ 00276⇤⇤⇤ 00219⇤⇤⇤ 10759⇤⇤⇤ É00320⇤⇤⇤
00278⇤⇤⇤ 00264⇤⇤⇤
4000215 4000345 4000185 4000245 4000215 4000335 4000155
4000225No. of observations (firms) 3,498 (202) 3,498 (202) 3,498
(202) 3,498 (202) 2,934 (180) 2,934 (180) 2,934 (180) 2,934
(180)
Notes. Firm-level controls include L Age, L VC inflows stock,
and L Weighted products. L Patent applications stock is also a
control for (3-3), (3-4), (3-7), and(3-8). The samples are not CEM
matched because the comparisons simply reflect innovation rates
post versus preevent for the sample of firms undergoingeach
event.
⇤⇤ and ⇤⇤⇤ indicates statistical significance at 5% and 1%,
respectively.
stock within four years of patent application in thethree years
post-IPO.The final four columns of the table report analo-
gous specifications for the 180 firms undergoing anM&A,
comparing post- with pre-M&A innovationrates. With the full
slate of controls, we find that thepost-M&A (113) window is
associated with a 22%increase in patent applications and a 7%
decrease inforward patent citations (both estimates are
statisti-cally significant at the 1% level). These estimates
havenot taken into consideration the possible self-selectioninto
exit mode based on unobservables, however. Wetherefore employ
several strategies including CEMmatching, an instrumental variables
analysis, and acomparison of actual versus “near” liquidity
eventsto better understand the relationship between exitmodes and
innovation patterns.
4.2. Coarsened Exact Matching EstimatesIn Table 4, we use the
CEM technique, balanced onthe log values of age, VC inflows stock,
alliance stock,and weighted products to define an IPO treatment
andcontrol sample (we omit the alliance stock variableas a
regressor in our models because it significantlyreduces our sample
size and because it is significantlycorrelated with our VC inflows
variable). We also testthe robustness of our results to using CEM
matchingon preevent stocks (as of the year prior to the event)of
the dependent variables. We find that our keyresults hold, although
we report our results withoutmatching on the stocks of preevent
outcomes, becausethey are more conservative. The first three
columns ofTable 4 examine the outcome variable log patent
appli-cations stock. Each OLS specification contains our fullset of
firm controls, event year fixed effects, and firmfixed effects. The
specifications differ on the sample
analyzed. We start with the entire CEM-balanced sam-ple
employing 328 firms. The difference-in-differencesestimate, focal,
post-IPO (113), after controlling for thefocal IPO sample, is
negative and significant, with animplied 40% drop in patent
applications post-IPO(the comparison group is therefore firms which
wereeither private or experienced an M&A). The nexttwo columns
restrict the sample successively by firstremoving firms that
remained privately held over theduration of the study window
(reducing the sam-ple size to 200 firms and 1,872 firm-years, with
thecomparison group as firms undergoing an M&A) andthen
examining just the subsample of firms experi-encing both an IPO and
an M&A (yielding 79 firmsand 817 firm-year observations). In
both cases, focal,post-IPO (113) is negative and significant at the
1%level, though the estimated effect drops to 35% and28%,
respectively. These estimates are in line withthe estimates
produced from the simple post- ver-sus pre-IPO analysis of Table 3.
We also note thatthe CEM-balancing procedure seems successful, as
thecoefficient on the focal event sample in these andsubsequent
specifications is not different than zero,suggesting no preevent
differences in trends in thecomparison groups. A final note is that
in (4-3), sincethe sample contains firms undergoing both
liquidityevents (almost always in the order of IPO followedby
M&A), we can also estimate a focal, post-M&A(113) variable.
That estimated coefficient is not differ-ent than zero.The final
three columns of Table 4 examine the for-
ward patent citations outcome, following a parallelmodel
structure and subsample comparison as thefirst group of analyses in
this table. Here, we finda reversal of the empirical patterns
produced by asimple post- versus pre-IPO comparison. Recall thatin
that analysis, we found a positive and significant
Dow
nloa
ded
from
info
rms.o
rg b
y [1
65.1
23.3
4.86
] on
14 Ju
ly 2
014,
at 1
1:59
. Fo
r per
sona
l use
onl
y, a
ll rig
hts r
eser
ved.
-
Aggarwal and Hsu: Entrepreneurial Exits and InnovationManagement
Science 60(4), pp. 867–887, © 2014 INFORMS 879
Table 4 IPO Treatment vs. Control Sample (Post-CEM) Innovation
Comparisons (Firm-Year Level of Analysis) OLS Regression
Coefficients Reported
Dependent variable: L Patent applications stock L Forward patent
citations four years stock
Removing firms Firms experiencing both Removing firms Firms
experiencing bothSample: All remaining private an IPO and M&A
All remaining private an IPO and M&A
(4-1) (4-2) (4-3) (4-4) (4-5) (4-6)
Focal, post-IPO 41135 É00399⇤⇤⇤ É00352⇤⇤⇤ É00279⇤⇤⇤ É00190⇤⇤⇤
É00155⇤⇤⇤ É00243⇤⇤⇤4000385 4000395 4000505 4000275 4000295
4000385
Focal IPO sample 10076 É10391 10710 001124202575 4108725 4105885
4103565
Focal, post-M&A 41135 00035 É00119⇤⇤⇤4000555 4000415
Firm controls Yes Yes Yes Yes Yes YesEvent year FE Yes Yes Yes
Yes Yes YesFirm FE Yes Yes Yes Yes Yes YesConstant É00778 É10190
É10762⇤⇤⇤ 00063 00066 É00437
4202075 4005405 4005855 4105555 4103085 4004435No. of
observations (firms) 2,702 (328) 1,872 (200) 817 (79) 2,702 (328)
1,872 (200) 817 (79)
Note. Firm-level controls include L Age, L VC inflows stock, and
L Weighted products; L Patent applications stock is also a control
for 4-4, 4-5, and 4-6 only.⇤⇤⇤Indicates statistical significance at
1%.
effect of citations post-IPO. Using the CEM-balancingprocedure,
we instead find a negative and signifi-cant effect at the 1% level
across the various samples.Using the entire sample, we find a 19%
drop. Underthe logic that firms remaining private for the
entirestudy period may be qualitatively different (in
unob-servables) compared to firms achieving liquidity, andso should
be left aside in the analysis, we estimate a15% drop in forward
citations. Finally, restricting thesample to firms undergoing both
events produces a24% estimated decline in forward citations
post-IPOas compared to a 12% (and statistically significant)decline
post-M&A (the two coefficients are statisti-cally different
from each other). Therefore, using aCEM-balanced sample of IPO
treatment versus con-trol, we find that IPOs are associated with
both worseinnovation quantity and quality.In Table 5, we report a
similar table, but for M&A
treatment and control samples using CEM balancing.We follow an
analogous structure as in Table 4 withregard to model specification
and sample compar-isons. For patent applications, our results are
similarto what we find in the post- versus pre-M&A sam-ple: a
positive and significant effect.17 However, acrossthe range of
samples used in this table, our estimated
17 At the suggestion of an anonymous reviewer, we investigatethe
importance of an alternative “window dressing” mechanismin the
preevent period in which the focal firm files many
patentapplications to attract the relevant audience. This
alternative holdsonly for patent applications rather than forward
patent citations,because only the former is contemporaneously
observed by theaudience (we also checked for any preevent spikes in
forwardpatent citations—we did not find any). Recall that our
findings onpatent applications are declines in the time window
following IPO,but are an increase in the time window following the
average M&A.Therefore, this alternative explanation only
applies to our post-IPO
effects here are 25% to 50% of the economic size ofthe prior
analysis, which did not account for selection.On the other hand,
our analysis of forward patentcitations yields both similar
statistical and economicsignificance as the simple post- versus
pre-M&A anal-ysis: a negative and significant decline in
forwardpatent citations. In addition, the negative and signifi-cant
effect of the post-IPO window for (5-3) and (5-6)associated with
patent applications and forward cita-tions, respectively, is
consistent with the results fromTable 4 (the former coefficient is
statistically differentand of opposite sign than the focal,
post-M&A (113)coefficient in the same specification; the latter
coeffi-cient is statistically lower than its corresponding
focal,post-M&A (113) coefficient). Finally, note that the
focalM&A sample dummy is also not statistically differ-ent than
zero in all the specifications in Table 5, againimplying a
successful CEM-balancing procedure.
4.3. Endogenous Choice of IPO vs. M&AOne concern with the
CEM-balanced estimates pre-sented in the prior two tables is that
the match-ing procedure is only as good as the observablesupon
which we could possibly balance the treatedand control samples. As
a result, there could still beunobserved selection issues
associated with those esti-mates. We therefore employ two
additional empiri-cal strategies to estimate our effects, both of
which
patent application results. To evaluate it, we include dummies
forvarious time window dummies prior to the event year (in
additionto the post-IPO time dummy—and so the interpretation of the
timewindow variables is relative to the event year). We find either
nopreevent spike for a majority of the preevent windows, or for
afew of the windows, a small (relative to the post-IPO window)
andnegative coefficient. We therefore conclude that preevent
windowdressing is unlikely to explain the empirical patterns.
Dow
nloa
ded
from
info
rms.o
rg b
y [1
65.1
23.3
4.86
] on
14 Ju
ly 2
014,
at 1
1:59
. Fo
r per
sona
l use
onl
y, a
ll rig
hts r
eser
ved.
-
Aggarwal and Hsu: Entrepreneurial Exits and Innovation880
Management Science 60(4), pp. 867–887, © 2014 INFORMS
Table 5 M&A Treatment vs. Control Sample (Post-CEM)
Innovation Comparisons (Firm-Year Level of Analysis) OLS Regression
Coefficients Reported
Dependent variable: L Patent applications stock L Forward patent
citations four years stock
Removing firms Firms experiencing both Removing firms Firms
experiencing bothSample: All remaining private an M&A and IPO
All remaining private an M&A and IPO
(5-1) (5-2) (5-3) (5-4) (5-5) (5-6)
Focal, post-M&A 41135 00171⇤⇤⇤ 00177⇤⇤⇤ 00106⇤⇤⇤ É00037⇤⇤⇤
É00036⇤⇤⇤ É00078⇤⇤⇤4000225 4000225 4000275 4000135 4000145
4000185
Focal M&A sample 00693 00522 00011 10431 10430 É104324301985
4302225 4001895 4109325 4109565 4200825
Focal, post-IPO 41135 É00504⇤⇤⇤ É00083⇤⇤⇤4000475 4000335
Firm controls Yes Yes Yes Yes Yes YesEvent year FE Yes Yes Yes
Yes Yes YesFirm FE Yes Yes Yes Yes Yes YesConstant É00290 00372
É00188 00009 00361 00153
4202625 4202805 4202355 4103665 4103835 4104725No. of
observations (firms) 4,711 (396) 3,681 (260) 1,369 (96) 4,711 (396)
3,681 (260) 1,369 (96)
Note. Firm-level controls include L Age, L VC inflows stock, and
L Weighted products; L Patent applications stock is also a control
for 5-4, 5-5, and 5-6 only.⇤⇤⇤Indicates statistical significance at
1%.
use CEM matching as the first step to sample con-struction. In
our first strategy, we conduct extensiveresearch into the firms
within our original sample thatnearly completed a liquidity event,
but for reasonsunrelated to innovation did not complete the
event.
Table 6 Endogenous Choice of IPO vs. M&A Innovation
Comparisons (Firm-Year Level of Analysis) “Near” vs. Actual Events
and Instrumental VariableAnalyses (Post-CEM) OLS Regression
Coefficients Reported
2SLS IV analysis on firms undergoingOLS analysis of “near” vs.
OLS analysis of “near” vs. either an IPO or M&A,
Estimation method and sample: actual IPOs, post-CEM actual
M&As, post-CEM post-CEM balancing (IPO treatment)
L Patent L Forward L Patent L Forward L Patent L
Forwardapplications patent citations applications patent citations
applications patent citations
Dependent variable: stock four years stock stock four years
stock stock (2SLS) four years stock (2SLS)
(6-1) (6-2) (6-3) (6-4) (6-5) (6-6)
Focal, post-IPO 41135 É00320⇤⇤⇤ É00151⇤⇤⇤ É00409⇤⇤⇤
É00150⇤⇤⇤4000415 4000305 4000585 4000435
Focal, post-M&A 41135 00180⇤⇤⇤ É00050⇤⇤4000215 4000145
Focal event sample É00429 00045 00457 É00985 É00145 001524009105
4004575 4201825 4104495 4100335 4007515
IPO year indicator (instrumented) 10157⇤⇤ É000694005275
4003845
Firm-level controls Yes Yes Yes Yes Yes YesEvent year FE Yes Yes
Yes Yes Yes YesFirm FE Yes Yes Yes Yes Yes YesConstant É20373⇤⇤⇤
00807⇤ É00457⇤⇤⇤ 00985⇤⇤⇤ É10000 00809
4007495 4004355 4001245 4001695 4007005 4005105No. of
observations (firms) 1,612 (168) 1,612 (168) 2,154 (175) 2,154
(175) 1,049 (179) 1,049 (179)
Notes. Firm-level controls include L Age, L VC inflows stock,
and L Weighted products; L Patent applications stock is also a
control for 6-2, 6-4, and 6-7 only.For (6-5) and (6-6), the
first-stage logit regression of the endogenous variable, IPO year
indicator, on the instrumental variable, IPO vs. M&A
biotechnologyindustry liquidity, yields a positive and significant
coefficient of 0.036 with a standard error of 0.013 (p < 0001).
The F statistic of the first stage is 22.7,suggesting that the
instrument is not weak.
⇤, ⇤⇤, and ⇤⇤⇤ indicate statistical significance at 10%, 5%, and
1%, respectively.
We compare actual versus “near” IPO events, post-CEM matching,
in the first two columns of Table 6 forboth of our outcome
variables (in unreported analy-ses, we find that the balance
between the treatmentand control samples for the actual versus
near-IPOs
Dow
nloa
ded
from
info
rms.o
rg b
y [1
65.1
23.3
4.86
] on
14 Ju
ly 2
014,
at 1
1:59
. Fo
r per
sona
l use
onl
y, a
ll rig
hts r
eser
ved.
-
Aggarwal and Hsu: Entrepreneurial Exits and InnovationManagement
Science 60(4), pp. 867–887, © 2014 INFORMS 881
and M&As reported in Table 2 is maintained). Weinclude in
each specification our full set of time-varying firm controls and
report firm fixed-effectsOLS models. Our results are consistent
with the CEManalyses in Table 4, in which we find negative and
sig-nificant difference-in-differences post-IPO time win-dow
effects for patent quantity and quality. In thethird and fourth
columns of Table 6, we conductan analogous examination using actual
versus nearacquisitions. Again, our results echo our findingsfrom
Table 5, with a positive and significant post-M&A window effect
on patent applications, but anegative and significant coefficient
for the same win-dow on forward patent citations. For both pairs
ofactual versus near event analyses, we regressed thelikelihood of
withdrawal on our innovation variablesand our full set of firm
characteristics (with time lags)and found all regressors
insignificant (available uponrequest from the authors). This lends
support to ourquasi-experimental strategy in that withdrawn
eventsare not systematically related to innovation or
firmcharacteristics in a regression framework.Our second empirical
strategy to address selection
of liquidity mode based on unobservables adds aninstrumental
variables strategy to an IPO-treatmentCEM-balanced sample. For this
analysis, we confinethe sample to firms experiencing either an IPO
orM&A liquidity event and instrument for the poten-tially
endogenous variable, IPO year indicator. We doso by constructing a
variable, IPO vs. M&A biotechnol-ogy industry liquidity. As
noted above, this variable isdefined at the biotechnology industry
level and is ameasure of the comparative deal volume of each
liq-uidity mode over a rolling time window. The higherthe value of
IPO vs. M&A biotechnology industry liquid-ity, the “hotter” is
the IPO market relative to the M&Amarket for biotechnology
transactions. As a result, allelse equal, the higher the
instrumental variable (IV),the more likely a given firm will choose
an IPO as aresult of the comparative “money-chasing deals”
IPOenvironment. This logic is borne out when we regressIPO year
indicator on IPO vs. M&A biotechnology indus-try liquidity and
our slate of firm controls. The result-ing coefficient is positive
and statistically significant atthe 1% level. This is the
first-stage regression in bothspecifications (6-5) and (6-6) in
which we run two-stage least squares (2SLS) regressions. The F
-statisticfor our first-stage regression is 22.7, strongly
suggest-ing that our IV is not weak. Durbin and Wu-Hausmantests
(with values of 19 and 15) reject the null hypoth-esis that IPO
year indicator is exogenous.In addition, the requirement that the
IV is uncor-
related with firm innovation outcomes is likely satis-fied in
our case. The IV is a measure of industry-levelrelative liquidity,
whereas our ultimate outcome vari-ables are at the firm level.
Furthermore, the IV is a
measure of relative liquidity of exit mode rather thana measure
of differences in factor inputs that mightbe correlated with
firm-level innovation outcomes.Finally, it is not only notoriously
difficult to predictthe degree to which a financing channel will be
“hot”(e.g., Lowry 2003), but also the relative degree towhich one
market will be more active than another.This suggests that it will
be very difficult or not pos-sible for entrepreneurs with (possibly
unobserved)innovation expectations to correctly anticipate
rela-tively “hot” financing modes. Although our instru-mental
variable allows us to meet the order conditionfor identification,
there is no direct statistical test ofthe exclusion restriction.
Using this empirical frame-work, our results on innovation quantity
and qualityare consistent with the estimates we obtained fromusing
CEM matching alone (Table 4) and CEM match-ing coupled with actual
versus near IPOs (first twocolumns of Table 6). Furthermore, the
2SLS results arerobust to omitting the CEM-balancing scheme
(whichhas the effect of nearly tripling the number of
usablefirm-year observations).The results thus far are consistent
with the informa-
tion confidentiality mechanism in that innovation out-comes are
worse post-IPO relative to post-M&A, andseem to be best under
private ownership. This patternholds for patent applications
(comparing (6-1) to (6-3)and (6-5)) and for forward patent
citations (comparing(6-2) to (6-4) and (6-6)), even after
addressing the roleof possible self-selection into liquidity mode.
Infor-mation confidentiality is best preserved under pri-vate
ownership and is partially compromised underan acquisition
(information is spread to the acquireror candidate acquirers). IPOs
represent the structurewith the most information revelation to the
greatestnumber of outsiders among the ownership
structures,consistent with the predictions of the information
con-fidentiality mechanisms. We now examine situationswithin
liquidity mode in which the information confi-dentiality effects
are likely to be more or less severe toprovide another dimension of
empirical evidence forthis mechanism because it might connect to
firm-levelinnovation outcomes.
4.4. Within-Event HeterogeneityWe begin by examining
heterogeneous within-IPOeffects. Whereas all IPOs in the United
States necessi-tate regulatory compliance with the SEC with
regardto information disclosure, we believe that the negativeeffect
of information confidentiality on innovationoutcomes may be most
salient under two concurrentconditions: namely, when the focal
biotechnology firmhas many early-stage projects (as proxied by the
num-ber of preclinical products), and at the same timethe firm
itself receives considerable scrutiny (leadingto increased
information flows to outsiders) by stock
Dow
nloa
ded
from
info
rms.o
rg b
y [1
65.1
23.3
4.86
] on
14 Ju
ly 2
014,
at 1
1:59
. Fo
r per
sona
l use
onl
y, a
ll rig
hts r
eser
ved.
-
Aggarwal and Hsu: Entrepreneurial Exits and Innovation882
Management Science 60(4), pp. 867–887, © 2014 INFORMS
Table 7 Within-Event Heterogeneity (Firm-Year Level of Analysis,
Post-CEM Matching) OLS Regression Coefficients Reported
Within IPO sample Within M&A sample
L Patent L Forward L Patent L Forward L Patent L
Forwardapplications patent citations applications patent citations
applications patent citations
stock four years stock stock four years stock stock four years
stock
(7-1) (7-2) (7-3) (7-4) (7-5) (7-6)
L Analyst reports 00060⇤⇤⇤ 00033⇤⇤⇤4000135 4000095
L Preclinical products É00134 É000224002385 4001625
L Analyst reports ⇤ 00046 É00084⇤⇤L Preclinical products 4000605
4000415
Focal, post-M&A 41135 00191⇤⇤⇤ É00062⇤⇤⇤ 00043
00137⇤⇤⇤4000245 4000165 4000545 4000395
Focal, post-M&A 41135, É00051 00075⇤⇤private acquirer
4000525 4000345
Focal, post-M&A 41135, 00143⇤⇤ É00298⇤⇤⇤tech overlap 4000755
4000545
Firm controls Yes Yes Yes Yes Yes YesEvent year FE Yes Yes Yes
Yes Yes YesFirm FE Yes Yes Yes Yes Yes YesConstant É00545 10956⇤⇤⇤
00123 10410 00821 00607
4005515 4003755 4201855 4104375 4200605 4104965No. of
observations (firms) 891 (129) 891 (129) 2,089 (170) 2,089 (170)
1,593 (119) 1,593 (119)
Note. Firm-level controls include L Age and L VC inflows stocks;
L Weighted products is included in specifications 7-3 through 7-6,
and L Patent applicationsstock is also a control for 7-2, 7-4, and
7-6 only.
⇤⇤ and ⇤⇤⇤ indicate statistical significance at 5% and 1%,
respectively.
analysts. Stock analysts therefore work in the oppo-site
direction as information confidentiality, exposinginformation and
firm analysis to the outside. In thefirst two columns of Table 7,
we analyze the interac-tion effect of log analyst reports and log
preclinical prod-ucts on our two innovation outcomes. Although wedo
not find a significant patent applications effect, wedo find a
negative and statistically significant effect ofthis interaction on
our measure of innovation quality.Although the direct effect of
analyst coverage is posi-tive on innovation, the interaction effect
suggests thatfor a given level of preclinical products, the
marginalimpact of increasing analyst attention as measuredby
analyst reports by one standard deviation resultsin a decrease of
2.2% in forward patent citations.This effect is consistent with
other research highlight-ing the risks of information disclosure in
the earlystages of the biotechnology product development pro-cess:
“At an early stage in the product developmentcycle, the firm’s lead
time over potential competitorsis short, and managers may
accordingly view the riskof adverse competitor action as high and
thereforebe reluctant to disclose extensive proprietary
informa-tion” (Guo et al. 2004, p. 326).To probe the within-IPO
sample for possible evi-
dence of organizational governance effects, we col-lected
information on whether the executive officers(including the chief
executive officer) of the firm
at the time of IPO were also founders of the firm.Although there
could be varied reasons for observ-ing such instances, we examine
whether there areconsequences for innovation depending on such
exec-utive officer status. On one hand, we might conjec-ture
incentive alignment because founders typicallypossess a large share
of equity, even at the time ofIPO. On the other hand, the
literature has reportedfounder control tendencies (e.g., Boot et
al. 2006,Schwienbacher 2008), and so the net effect is
theoret-ically ambiguous. We define two variables to capturethe
phenomenon: (1) an indicator variable for whetherthe CEO at the
time of IPO is also a founder, and(2) the percentage of executive
officers at the time ofIPO who were founders. In both cases (when
inter-acted with the post-IPO time window), we find nosignificant
effect on forward patent citations, althoughwe do find a
significant positive effect using the CEOvariable on patent
applications (results available uponrequest).Similarly, we examine
heterogeneity within the
M&A sample with an eye to testing the informa-tion
confidentiality mechanism. First, we conjecturethat there might be
differential information disclo-sure effects associated with
acquisition by a publicversus private acquirer. Because such an
acquisi-tion happens only once, to estimate the effect, weinteract
an indicator for private acquirer with the
Dow
nloa
ded
from
info
rms.o
rg b
y [1
65.1
23.3
4.86
] on
14 Ju
ly 2
014,
at 1
1:59
. Fo
r per
sona
l use
onl
y, a
ll rig
hts r
eser
ved.
-
Aggarwal and Hsu: Entrepreneurial Exits and InnovationManagement
Science 60(4), pp. 867–887, © 2014 INFORMS 883
key difference-in-differences variable focal, post-M&A(113)
in our OLS panel firm fixed-effects framework.Although we do not
find an effect of this inter-action on patent applications, the
effect is positiveand significant for forward patent citations.
This sug-gests that relative to the post-M&A window of
publicacquirers, biotechnology targets acquired by privateentities
receive a nearly 8% boost in innovation qual-ity. Naturally,
private acquirers retain more informa-tion confidentiality relative
to public acquirers.Taken together, these two empirical patterns
of
within-event heterogeneity provide additional evi-dence
consistent with the information confidentialitymechanism. With
regard to M&As, the Seru (2013)and related theories suggest an
additional within-M&A pattern. Recall that this theory relates
businessunit manager incentives for innovation in the contextof a
competitive internal capital and labor market of aconglomerate
(which the acquired innovator joins inthe case of an acquisition).
Due to such competitionat least in the short run, individual
managers ma