its extent of subsequent diversiﬁcation. We measure a ﬁrm’s innovations by the number
of patents that the ﬁrm has accumulated, weighted by the number of citations that these
patents receive. Our main measure of a ﬁrm’s extent of diversiﬁcation is the number of
industries (at the 4-digit SIC level) in which the ﬁrm is operating, but we also use other
measures for robustness checks. When we examine the effects of innovations on
diversiﬁcation, endogeneity is a major concern. For example, a third variable may affect
both diversiﬁcation and innovation. Also, reverse causality is possible. For example,
Seru (2014) ﬁnds that diversiﬁcation signiﬁcantly affects innovations. To tackle these
endogeneity issues, we utilise three main empirical strategies: (1) the ﬁrm ﬁxed effects
model with one-year lagged innovations as the key independent variable; (2) the 2SLS
(two-stage-least-squares) model using the state-level R&D tax credits as the instrument
for innovations; and (3) the dynamic panel GMM (Generalised Method of Moments)
model. As a robustness check, we also use the Abadie-Imbens matching method.
Our ﬁrm ﬁxed effects model successfully controls for the time-invariant ﬁrm-speciﬁc
factors that are correlated with both innovations and diversiﬁcation. Furthermore, using
lagged innovations helps us mitigate the potential endogeneity issue to some degree. In
this model, we ﬁnd that a ﬁrm’s innovations have signiﬁcant and positive effects on the
extent of its subsequent diversiﬁcation.
However, if innovations exhibit strong autocorrelation, using lagged innovations
would be less effective to mitigate potential endogeneity. To deal with endogeneity more
effectively, we resort to the 2SLS technique. Motivated by Wilson (2009), we use an
exogenous policy shock, i.e., the staggered implementation of state-level R&D tax
credits as the instrument for innovations. It is a valid instrument because the
implementation of R&D tax credits should affect ﬁrms’innovations, but it should not
directly affect ﬁrms’diversiﬁcation decisions. Indeed, Wilson (2009) ﬁnds that the
implementation of state-level R&D tax credits signiﬁcantly boosts ﬁrms’R&D
expenditures, and many studies document that higher R&D expenditures are associated
with more innovations (e.g., Fang et al., 2014). Using this exogeneous policy shock, our
2SLS estimation provides evidence on the positive causal effect of a ﬁrm’s innovations
on its subsequent diversiﬁcation.
One potential problem with the above instrument is that it only varies at the state-year
level, so it may fail to address the possibility that the time-variant ﬁrm-speciﬁc
unobservables affect both innovations and diversiﬁcation. To deal with this dynamic
endogeneity issue, we follow Acemoglu et al. (2008), Hoechle et al. (2012), Wintoki
et al. (2012), O’Connor and Rafferty (2012) and Flannery and Hankins (2013)’s methods
and estimate the dynamic panel GMM model. Doing so enables us to use ﬁrm-level
‘internal instruments’to tackle dynamic endogeneity. Our GMM estimation results
provide further evidence on the positive causal effect of innovations on diversiﬁcation.
Thispositive relationshipbetween a ﬁrm’s innovationsand its subsequentdiversiﬁcation
is consistent with the following two scenarios. Suppose a ﬁrm has some innovations
applicableto industry A but none of the ﬁrm’s innovations is applicableto industry B, and
both industriesare potential industriesthe ﬁrm may enter. Scenario (1) isthat the ﬁrm then
diversiﬁesinto industryA, while scenario (2) is thatthe ﬁrm then diversiﬁes intoindustry B.
In both scenarios, the ﬁrm’s innovations are positively correlated with the extent of its
subsequent diversiﬁcation. If scenario (1) is true, our hypotheses will receive strong
support. However, if scenario (2) is true, it will cast serious doubt on our hypotheses.
To distinguish between these two scenarios, we need to examine the following
question: given that a ﬁrm diversiﬁes into one of these two industries (A or B), and that its
© 2016 John Wiley & Sons, Ltd.
Innovation-Related Diversiﬁcation and Firm Value 477