You are on page 1of 26

JID: JAE

ARTICLE IN PRESS [m3Gsc;August 20, 2019;11:56]


Journal of Accounting and Economics xxx (xxxx) xxx

Contents lists available at ScienceDirect

Journal of Accounting and Economics


journal homepage: www.elsevier.com/locate/jacceco

Strategic reactions in corporate tax planning


Christopher S. Armstrong a,∗, Stephen Glaeser b, John D. Kepler a
a
The Wharton School, University of Pennsylvania, 1300 Steinberg-Dietrich Hall, Philadelphia, PA 19104-6365, USA
b
The Kenan-Flagler Business School, University of North Carolina, 4000 McColl Building, Chapel Hill, NC 27514, USA

a r t i c l e i n f o a b s t r a c t

Article history: We find that firms’ tax planning exhibits strategic reactions: firms respond to changes in
Received 30 May 2017 their industry-competitors’ tax planning by changing their own tax planning in the same
Revised 4 December 2018
direction. We document evidence of these strategic reactions in two distinct research set-
Accepted 8 March 2019
tings that entail an exogenous increase and decrease in competitors’ tax planning. We also
find evidence that strategic reactions stem from concerns about appearing more tax ag-
Keywords: gressive than industry competitors, some evidence that they stem from firms learning from
Corporate income tax the tax planning of their industry competitors, and no consistent evidence that they stem
Tax planning from leader-follower dynamics.
Industry competition © 2019 Elsevier B.V. All rights reserved.
Strategic interaction
Tax havens
Difference-in-differences
Partially-overlapping groups
Reflection problem

1. Introduction

We examine whether firms adjust their own tax planning in response to their peers’ tax planning. Prior game-theoretic
and strategy studies show that many corporate decisions (e.g., R&D, advertising, and capital expenditures) can be char-
acterized as strategic reactions that describe how a firm’s decisions depend on the anticipated behavior of its peers (e.g.,
Fudenberg and Tirole, 1984; Boone, 20 0 0; Leary and Roberts, 2014).1
Prior tax planning studies have alluded to the possibility of strategic reactions by suggesting that firms’ tax planning
decisions may be influenced by managers’ desire to avoid “standing out” from their peers (e.g., Mills et al., 1998; Rego,
2003; Graham et al., 2014; Gallemore et al., 2014; Boning et al., 2018). For example, tax authorities’ audit and enforcement
policies that target firms with relatively extreme tax planning could induce a strategic component to firms’ tax planning
decisions. Similarly, the presence of customers or other stakeholders who do not want to purchase from or contract with
a firm whose tax planning is significantly different from that of its competitors could also induce a strategic component
to firms’ tax planning decisions. In either case, firms may be less inclined to adopt risky tax positions if doing so might
cause their tax planning to seem particularly aggressive relative to that of their peers. Similarly, managers considering tax


Corresponding author.
E-mail addresses: carms@wharton.upenn.edu (C.S. Armstrong), stephen_glaeser@kenan-flagler.unc.edu (S. Glaeser), jkepler@wharton.upenn.edu (J.D. Ke-
pler).
1
As we explain in greater detail in Section 2, strategic reactions are distinct from other types of competitive forces, such as product market competition,
in that strategic reactions arise when a firm chooses a particular attribute (e.g., its level of R&D, advertising, or tax planning) depending—at least in part—on
what it expects its peers (e.g., competitors) to choose for the same attribute.

https://doi.org/10.1016/j.jacceco.2019.03.003
0165-4101/© 2019 Elsevier B.V. All rights reserved.

Please cite this article as: C.S. Armstrong, S. Glaeser and J.D. Kepler, Strategic reactions in corporate tax planning, Journal
of Accounting and Economics, https://doi.org/10.1016/j.jacceco.2019.03.003
JID: JAE
ARTICLE IN PRESS [m3Gsc;August 20, 2019;11:56]

2 C.S. Armstrong, S. Glaeser and J.D. Kepler / Journal of Accounting and Economics xxx (xxxx) xxx

strategies with greater adjustment costs, such as corporate inversions, may be less inclined on the margin to implement the
strategy if it would cause their tax planning to seem particularly aggressive.
The potential for strategic reactions in corporate tax planning has important implications for researchers, legislators,
regulators, and other stakeholders. For example, debates about the benefit of policies designed to curb avoidance typically
focus on the expected increase in tax revenues from firms directly targeted by the policy.2 However, the presence of strategic
reactions implies that firms that are not directly affected by the policy will also change their tax planning in response to
policy-induced changes in their directly affected peers’ tax planning. To the extent that tax authorities fail to consider these
strategic reactions, they will arrive at inaccurate estimates of the total effect of a given tax policy.
Relatedly, empirical studies of tax policy that fail to account for strategic reactions may also produce systematically
biased estimates of the policy’s direct effect. For instance, a common research design for estimating the effect of a particular
tax policy involves comparing the tax planning of firms directly targeted by the policy—i.e., the treatment group—with
that of firms that are not directly targeted—i.e., the control group. This approach implicitly assumes that control firms are
unaffected by the policy (i.e., the stable unit treatment value assumption, Murnane and Willett, 2010; Glaeser and Guay,
2017; Armstrong and Kepler, 2018).3 However, if there is a strategic component to firms’ tax planning, this assumption
is problematic since the tax policy can still have an indirect effect on control firms as they respond to policy-induced
changes in the tax planning of their directly targeted counterparts. Consequently, empirical estimates of the direct effect
of tax policies will be understated if there are positive strategic reactions and, conversely, overstated if there are negative
strategic reactions.
Finally, theoretical studies of tax policy typically model corporate tax planning as a strategic game between firms and
regulators, ignoring potentially important strategic considerations among firms (e.g., Desai et al., 2007). However, modeling
tax planning as a zero-sum game between a single firm and the tax authority may be inadequate to the extent there is a
strategic component of firms’ tax planning decisions.
Despite multiple a priori reasons to expect strategic reactions in firms’ tax planning and their potential importance for
formulating and evaluating the efficacy of tax policy, prior work has not investigated the possibility of strategic reactions
in tax planning. One likely explanation for this gap in the literature is that the inherently simultaneous nature of strategic
reactions poses a particularly difficult econometric challenge that Manski (1993) refers to as the “reflection problem.” In
the context of corporate tax planning, the reflection problem arises because peer firms (e.g., firms in the same industry)
are exposed to the same background characteristics (e.g., similar product market competition) and tend to have similar
firm-level characteristics (e.g., similar R&D spending), both of which influence their tax planning. These similarities make it
difficult to isolate—or identify—firms’ strategic response to their peers’ tax planning, and vice versa.
We identify the extent to which firms’ tax planning contains a strategic component using a partially-overlapping groups
research design in two distinct research settings. This research design isolates firms that are not directly exposed to a change
in tax regulation that affects some firms’ tax planning, but are indirectly exposed by way of shared industry membership
with these directly exposed firms (Bramoullé et al., 2009; De Giorgi et al., 2010). We follow prior work and identify peer
groups based on industry classifications that correspond to how regulators and other stakeholders either implicitly or ex-
plicitly benchmark firms against their industry competitors (e.g., Leary and Roberts, 2014; Bozanic et al., 2016). The extent
to which firms with no direct exposure to an exogenous change in tax policy respond to changes in their directly exposed
industry competitors’ tax planning triggered by the policy change reflects the strategic component of tax planning.
Our first research setting focuses on the significant corporate tax rate cuts in Ireland (from 32% to 12.5%) that occurred
between 1998 and 2003. These tax cuts provide a powerful source of arguably exogenous variation in the tax planning of
U.S. multinational firms with operations in Ireland (“Irish-haven firms”), as many of these firms shifted income to Ireland to
take advantage of the lower tax rate.4 Although Irish-haven firms’ industry competitors that had no Irish operations were
not directly affected by the Irish tax cuts, they were, nevertheless, indirectly exposed by way of their shared industry mem-
bership with Irish-haven firms. Consequently, the extent to which these indirectly exposed firms altered their tax planning
in response to changes in their Irish-haven competitors’ tax planning captures the strategic component of corporate tax
planning.5

2
For example, https://www.finance.senate.gov/imo/media/doc/58198.pdf. Here, regulators focus on the lost revenues due to corporations shifting income
to the Cayman Islands. However, they do not consider the possibility that curbing the avoidance of firms that take advantage of tax havens in the Cayman
Islands may have indirect spillover effects on these firms’ peers.
3
The stable unit treatment value assumption (SUTVA) requires no interference between units (e.g., firms), so that the value of a unit’s outcome when
exposed to the treatment is the same, regardless of the treatment status of other units. See, e.g., Rubin (1986), Heckman et al. (1999), Horowitz and Manski
(1995), Sobel (1996), Manski (2011), and Armstrong et al. (2018a) for additional discussion.
4
See, e.g., the 20 0 0 OECD report on identifying and eliminating harmful tax practices (“Towards Global Tax Co-operation,” available at https://www.oecd.
org/tax/harmful/2090192.pdf), the related discussion in Hanlon and Heitzman (2010), and articles in the business press, such as https://www.bloomberg.
com/graphics/2016- apple- profits/.
5
To illustrate our partially-overlapping groups research design in the Irish-haven setting, assume that directly-exposed Irish-haven firms experience an
8% increase in their tax planning following the Irish rate cuts, comprised of the following: 3% directly due to the Irish tax cuts, 2% due to common industry
factors, and 3% due to other macroeconomic factors. Further suppose that the indirectly exposed competitors of the Irish-haven firms strategically react by
changing their own tax planning by 2%. In addition, because they are in the same industry, they also experience the common 2% industry effect and the
3% macroeconomic effect. The control firms in this example only experience the 3% macroeconomic effect (plus any of their own industry effects, which is
assumed to be the same as the industry effect for the treatment firms, in expectation, under the parallel trends assumption).

Please cite this article as: C.S. Armstrong, S. Glaeser and J.D. Kepler, Strategic reactions in corporate tax planning, Journal
of Accounting and Economics, https://doi.org/10.1016/j.jacceco.2019.03.003
JID: JAE
ARTICLE IN PRESS [m3Gsc;August 20, 2019;11:56]

C.S. Armstrong, S. Glaeser and J.D. Kepler / Journal of Accounting and Economics xxx (xxxx) xxx 3

Our second research setting uses an arguably exogenous decrease in certain firms’ ability to engage in tax planning. In
particular, we follow Dyreng et al. (2013) who describe and document how firms use Delaware as a domestic tax haven.
Because Delaware does not tax income from intangible assets, firms can reduce their state tax liabilities by transferring
intangible assets to Delaware subsidiaries and making royalty payments from their non-Delaware subsidiaries for the use
of these assets. Dyreng et al. (2013) also explain how (and when) many states changed their tax policies to curtail firms’
ability to use Delaware subsidiaries to avoid taxes. We use these changes in various states’ corporate tax policies as a series
of staggered exogenous shocks that directly reduced targeted firms’ ability to avoid taxes, to which their unexposed industry
competitors should have responded if there are strategic reactions in corporate tax planning.
In both research settings, we find evidence that firms’ tax planning has a positive strategic effect on competitors’ tax
planning using several common empirical measures of tax planning, including GAAP effective tax rates (ETRs), Cash ETRs,
long-run GAAP and Cash ETRs in the Irish setting, and state ETRs in the Delaware setting. Our evidence suggests that this
positive strategic effect exists for both increases and decreases in competitors’ tax planning. In the Irish setting, we find that
firms increase their tax planning in response to an exogenous increase in their competitors’ tax planning. Conversely, in the
Delaware setting we find that firms decrease their tax planning in response to an exogenous decrease in their competitors’
tax planning. We adopt game-theoretic terminology and refer to these positive, symmetric strategic reactions as strategic
complementarities in firms’ tax planning (Bulow et al., 1985).
Our estimates of the magnitude of the strategic complementarities are economically significant and comparable to those
from related studies that examine the role of other competitive forces (e.g., product market competition) that shape firms’
tax planning (e.g., Kubick et al., 2014; Donohoe et al., 2015). For example, our results suggest that firms’ ETRs decrease
(increase) by roughly 0.63 (0.27) percentage points in response to a one percentage point decrease (increase) in their com-
petitors’ ETRs. In terms of dollar magnitude, estimates from the Irish setting imply that firms’ strategic reaction to changes
in their competitors’ tax planning represent a reduction in annual cash taxes paid and income tax expense of approximately
$6.5 and $5.9 million, respectively (corresponding to a 3.2 percentage point reduction in their Cash ETR and a 2.9 percentage
point reduction in their GAAP ETR).6
Having documented evidence of economically significant strategic complementarities in multiple measures of corporate
tax planning in two distinct research settings, we turn our attention to identifying the specific mechanism(s) through which
these strategic effects arise. Prior studies suggest that firms may seek to avoid drawing unwanted scrutiny from tax author-
ities or incur reputational costs with stakeholders (Graham et al., 2014; Bozanic et al., 2016; Boning et al., 2018). Anecdotal
evidence also suggests that firms that “stand out” in their industry by having relatively extreme or conspicuous levels of
tax planning are more likely to be audited by tax authorities or face legislative and regulatory scrutiny (Heitzman and
Ogneva, 2018).7 Consistent with firms’ concerns about appearing overly aggressive relative to their competitors inducing
strategic complementarities in tax planning, we find evidence that tax aggressive firms are more responsive to changes in
their competitors’ tax planning.
Additionally, we find some evidence that less tax aggressive firms are responsive to changes in their competitors’ tax
planning. This evidence is consistent with the strategic complementarities in tax planning partially resulting from firms
learning about new avoidance strategies and/or inferring the “acceptable” amount of tax planning by observing their com-
petitors’ tax planning (Brown, 2011; Brown and Drake, 2013). However, we do not find consistent evidence that smaller firms
are more responsive to changes in the tax planning of their larger, and presumably better informed, industry competitors in
a “leader-follower” relation (e.g., Leary and Roberts, 2014; Xiao, 2017).
Finally, because of the inherent difficulty in credibly identifying strategic reactions (e.g., Angrist, 2014), we conduct a
“falsification test” to assess the robustness of our inferences to alternative explanations. We do so by repeating each of
our primary difference-in-differences analyses using firms’ pre-tax financial performance as the dependent variable. Pre-
tax financial performance should not exhibit strategic reactions in our setting because the mechanisms that drive strategic
reactions in firms’ tax planning directly affect the costs and benefits of tax planning, but not the costs and benefits of pre-
tax financial performance. Furthermore, in the absence of unresolved agency conflicts, firms should always try to maximize
pre-tax firm performance and there is no obvious reason why firms would willingly reduce their pre-tax income in response
to reductions in competitors’ tax planning. Therefore, we should also find evidence of spurious strategic reactions in their
pre-tax performance if the strategic complementarities that we document are an artefact of an endogenous relation among
industry competitors’ tax planning. Supporting the validity of our inferences, we find no evidence of strategic reactions in
firms’ pre-tax net income in either research setting.

6
As we explain in more detail below, this result is consistent with firms exhibiting stronger strategic reactions to increases in competitors’ tax planning,
potentially because firms are more willing to increase rather than decrease their tax planning. However, as we also explain, this inference should be
interpreted cautiously since these estimates are not structural and therefore depend on the economic, policy, and other prevailing conditions in the research
settings.
7
For example, the IRS describes its audit process as follows: “Computer programs give each return numeric ‘scores’. The Discriminant Function System
(DIF) score rates the potential for change, based on past IRS experience with similar returns. The Unreported Income DIF (UIDIF) score rates the return for
the potential of unreported income. IRS personnel screen the highest-scoring returns, selecting some for audit and identifying the items on these returns
that are most likely to need review” (https://www.irs.gov/newsroom/the- examination- audit- process). The predictors that are used in these “computer
scoring” systems are at least partially based on the income distributions within company activity codes (see p. 2 of https://www.irs.gov/pub/irs-soi/puidif2.
pdf). The activity codes used to classify firms are very similar to—if not effectively the same as—SIC codes, which is precisely how we identify firms’
industry competitors in our primary analyses (see pp. 17–18 of https://www.irs.gov/pub/irs-pdf/i1040sc.pdf).

Please cite this article as: C.S. Armstrong, S. Glaeser and J.D. Kepler, Strategic reactions in corporate tax planning, Journal
of Accounting and Economics, https://doi.org/10.1016/j.jacceco.2019.03.003
JID: JAE
ARTICLE IN PRESS [m3Gsc;August 20, 2019;11:56]

4 C.S. Armstrong, S. Glaeser and J.D. Kepler / Journal of Accounting and Economics xxx (xxxx) xxx

In addition to the potential importance for formulating and evaluating the efficacy of tax policy, our evidence of strategic
reactions contributes to the literature by documenting an important determinant of corporate tax planning (see Wilde and
Wilson, 2018 for a review). We also contribute to the emerging literature that examines the role of “peer” and “network”
effects on various corporate attributes (e.g., Beck et al., 2014; Leary and Roberts, 2014; Bird et al., 2018). Studies in this
literature have found that managers and directors of peer firms play a role in shaping their firm’s policies via social inter-
actions (e.g., changes in social norms). We contribute to this literature by providing evidence that competitors’ (or “peers’”)
decisions can affect firms’ decisions not only through social interactions, but also by way of any monetary costs and benefits
associated with strategic interactions.
The remainder of our paper is as follows. Section 2 provides the theoretical background and motivation for our study.
Section 3 describes our two complementary research designs. Section 4 describes our data, sample construction, and variable
measurement. Section 5 presents our results and Section 6 concludes.

2. Background and motivation

2.1. Strategic reactions in corporate tax planning

A long line of game-theoretic studies highlights the role of strategic reactions in corporate decisions, whereby economic
agents’ decisions are not made in isolation, but rather account for the anticipated behavior of other economic agents (e.g.,
Diamond and Dybvig, 1983; Fudenberg and Tirole, 1984; Morris and Shin, 1998). A central result in this literature is that
in a strategic setting, a firm’s optimal decision is characterized as a reaction curve, which is a function that describes the
firm’s best response to the anticipated behavior of its peers—which are typically interpreted as its competitors. Firms with
downward-sloping reaction curves will make decisions that vary inversely with their competitors’ expected decisions, and
these decisions are described as “strategic substitutes.” Conversely, firms with upward-sloping reaction curves will make
decisions that vary in the same direction as their competitors’ expected decisions, and these decisions are described as
“strategic complements.”
There are several reasons to believe that corporate tax planning might exhibit strategic reactions in general, and strategic
complementarities in particular. For example, if firms incur “political costs” or are subject to enforcement on a relative
basis, then a firm that wishes to avoid being “singled out” would ensure that its tax planning closely conforms to that of its
industry peers (Zimmerman, 1983; Dyreng et al., 2016). Graham et al.’s (2014) survey of nearly 600 corporate tax executives
corroborates this conjecture: respondents rank public reputational concerns and the risk of IRS detection as the second and
third most important considerations that would prevent them from pursuing a tax planning strategy.
Firms may also be able to learn new avoidance strategies by observing their competitors’ tax planning. In this case,
strategic complementarities would arise because competitors’ tax planning reduces the firm’s own cost of learning new tax
planning strategies or the degree of tax planning that is unlikely to draw unwanted regulatory or public scrutiny. Consistent
with learning in corporate tax planning, Brown (2011) and Brown and Drake (2013) present evidence that tax planning
strategies propagate through board connections and Cen et al. (2018) present evidence that tax planning strategies propagate
along supply chains.
Finally, firms may mimic or copy the decisions of their industry competitors who are thought to be better informed
about the efficient—or acceptable—level of tax planning (i.e., “leader-follower” dynamics). Consistent with firms mimicking
industry leaders, Kubick et al. (2014) find that product market leaders engage in more tax planning, and that firms mimic
their industry competitors’ tax planning (see also Donohoe et al., 2015, who find that competition from tax-advantaged
banks leads to an increase in the tax aggressiveness of other banks).
Our study differs from Brown (2011), Brown and Drake (2013), Kubick et al. (2014), and Donohoe et al. (2015) in several
important ways. First, evidence from our mechanism and falsification tests suggests that our results do not primarily reflect
learning or leader-follower dynamics, but instead primarily reflect concerns about enforcement or reputation. Second, while
these other studies document whether changes in the opportunities for learning or changes in product market competition
lead to changes in firms’ tax planning, our study examines whether direct changes in competitors’ tax planning causes
changes in a firm’s own tax planning, and vice versa. Consequently, unlike our study, these studies do not speak to the
existence and magnitude of strategic reactions.
Our study is closest to Bird et al. (2018), which relies on manager fixed effects estimated—i.e., identified—using managers
who change firms within the sample of Execucomp firms as a source of plausibly exogenous variation in peer firms’ tax
planning. They find evidence of peer effects in GAAP ETRs, but not Cash ETRs, and conclude that peer effects operate via
earnings management incentives. In contrast to their findings, we document evidence of strategic complementarities in
multiple measures of tax planning, including Cash and State ETRs. Our evidence also suggests that enforcement herding is
one mechanism that is responsible for producing these strategic complementarities.

2.2. The reflection problem

The inherently simultaneous nature of strategic reactions makes it challenging to empirically identify these effects.
Manski (1993) refers to these econometric issues as the “reflection problem,” which he explains in the context of the

Please cite this article as: C.S. Armstrong, S. Glaeser and J.D. Kepler, Strategic reactions in corporate tax planning, Journal
of Accounting and Economics, https://doi.org/10.1016/j.jacceco.2019.03.003
JID: JAE
ARTICLE IN PRESS [m3Gsc;August 20, 2019;11:56]

C.S. Armstrong, S. Glaeser and J.D. Kepler / Journal of Accounting and Economics xxx (xxxx) xxx 5

following linear-in-means model:

Yi = β0 + βȲ Ȳ j + βX Xi + βX̄ X̄ j + βZ̄ Z̄ j + εi , (1)


where i indexes the firm and j indexes the firm’s reference group (e.g., industry). Yi is the outcome of interest (e.g., tax
planning) for firm i, Ȳ j is the average outcome of the firms in firm i’s reference group j, Xi is a vector of firm-level charac-
teristics of firm i, X̄ j is the average value of these characteristics for the firms in firm i’s reference group j, and Zj is a vector
of reference group-level characteristics that have the same values for all of the firms in firm i’s reference group.
Manski (1993) categorizes the determinants of outcome Yi into three mutually exclusive groups, which we now describe
in our specific context of corporate tax planning. First, firms in the same industry can exhibit similar levels of tax planning
because they share a common industry environment. For example, firms in the same industry can have similar tax planning
opportunities, such as specific income tax credits based on their particular line of business (e.g., energy, utilities, and phar-
maceuticals). As another example, firms in the same industry could face similar competitive forces that lead them to adopt
similar strategies (e.g., Kubick et al., 2014; Donohoe et al., 2015) that manifest in empirical measures of tax planning (e.g.,
GAAP ETR). Manski refers to this source of similarity as exogenous effects, which are captured by the coefficients on Zj .
Second, firms in the same industry can exhibit similar tax planning because their individual (i.e., firm-level) charac-
teristics are correlated with each other. For example, firms in the same industry tend to have similar capital structures,
performance, and risk profiles, which can affect their tax planning. Manski refers to this source of similarity as correlated
effects, which are captured by the coefficients on X̄ j .
Third, a firm’s tax planning may vary with the tax planning of the other firms in its industry because of a direct strategic
reaction, which is referred to as an outcome-on-outcome effect, and is captured by the coefficient on Ȳ j . Outcome-on-outcome
effects describe how a firm’s tax planning is directly affected by the tax planning of the other firms in its industry, beyond
any similarity induced by common exposure to industry-level forces (i.e., exogenous effects) and similarities in firm-level
characteristics within industries (i.e., correlated effects).
Manski explains how empirically identifying strategic reactions is complicated because the average outcome in the firm’s
industry, Ȳ j , is a linear combination of competitors’ exogenous and correlated effects. This introduces two distinct identifi-
cation problems. The first is the well-known endogeneity problem related to potentially omitted correlated effects. Second,
even if there are no omitted correlated effects, the resulting linear dependence makes it difficult to separately identify direct
strategic reactions (i.e., outcome-on-outcome effects). However, one solution for the reflection problem is the same as that
for identifying the effect of an endogenous variable in general: namely identifying a source of variation in competitors’ tax
planning that is otherwise exogenous with respect to the firm’s own tax planning. We adopt this approach for breaking the
simultaneity associated with the reflection problem and we describe how we do so in detail in Section 3. However, we first
provide a brief overview of several alternative approaches that are also capable of providing insight and drawing inferences
in the presence of the reflection problem.

2.3. Alternative designs for identifying strategic reactions8

Because of the growing interest in identifying strategic effects in a variety of settings and the difficulty of reliably iden-
tifying exogenous sources of variation, we provide a brief discussion of several alternative approaches for addressing the
reflection problem in the absence of a credible source of exogenous variation in peer behavior. These techniques differ in
their assumptions and data requirements and, in turn, their ability to detect, capture, isolate, and identify strategic reac-
tions. Our brief survey of these alternative techniques also provides insight into the nature of the reflection problem and
the intuition for our identification strategy.

2.3.1. “Coarsening” the unit of analysis


A relatively simple approach is to redefine the unit of analysis at a higher level (e.g., industries rather than firms). By
“coarsening” the unit of analysis, the resulting (aggregate) variables will capture any strategic reactions that occur at that
and lower levels of analysis. Although this approach might not be suitable for answering certain research questions, it may
nevertheless provide insight into the existence and nature of any strategic reactions. For example, in our setting, this ap-
proach might be capable of identifying certain industries that exhibit larger or stronger strategic reactions, although quanti-
fying the magnitude of the strategic reactions may not be feasible.

2.3.2. Exclusion of certain group-level averages as contextual effects


A necessary condition for identification is the existence of at least one individual characteristic whose group-level aver-
age is not a contextual effect, which is analogous to the row condition for identification in a simultaneous equations system
(Brock and Durlauf, 2001a; Durlauf and Ioannides, 2010, p. 459). Therefore, if the group-level averages of certain individual
characteristics can be justifiably excluded, this can be used to identify peer effects. Grinblatt et al. (2008) use this approach

8
Our discussion in this section draws heavily on Ioannides (2006), Durlauf and Ioannides (2010), and Blume et al. (2011). We refer interested readers to
those papers and the references therein for a more exhaustive treatment of identifying strategic effects, the reflection problem, and the various techniques
that we discuss.

Please cite this article as: C.S. Armstrong, S. Glaeser and J.D. Kepler, Strategic reactions in corporate tax planning, Journal
of Accounting and Economics, https://doi.org/10.1016/j.jacceco.2019.03.003
JID: JAE
ARTICLE IN PRESS [m3Gsc;August 20, 2019;11:56]

6 C.S. Armstrong, S. Glaeser and J.D. Kepler / Journal of Accounting and Economics xxx (xxxx) xxx

to show that individuals’ car purchases are influenced by their neighbors’ car purchases. They do so by excluding the neigh-
borhood averages of demographic characteristics as contextual effects by arguing that there is no reason why the average
age of an individual’s neighbors should directly affect the individual’s preferences for different types of cars.

2.3.3. Non-linear and dynamic models


Since the reflection problem arises from linear dependence between group outcomes and certain group-level aggre-
gates, non-linear-in-means models are not susceptible to this concern (Brock and Durlauf, 2001a). Relatedly, Brock and
Durlauf (2001b) derive a dynamic linear model that does not exhibit the reflection problem since it avoids linear dependence
between the contextual and endogenous variables. Despite the potential promise of non-linear and dynamic identification,
Durlauf and Ioannides (2010, p. 459) caution that these approaches rely on parametric identification and are therefore po-
tentially sensitive to the assumed functional form.

2.3.4. Variance-based approaches


Glaeser et al. (1996, 2003) and subsequent authors rely on the existence of “social multipliers”—whereby endogenous
interactions amplify differences in average group behavior—to provide estimates of strategic reactions based on group-level
data. For example, Glaeser et al. (1996) use city-level crime data, coupled with the intuition that social interactions will slow
the convergence of sample means, to estimate the magnitude of cross-city differences in crime rates beyond that explained
by city-level characteristics. They argue that the covariance of per capita crime rates across a city’s residents generates
over-dispersion in cross-city crime rates relative to the absence of social interactions. Glaeser et al. (1996) show that this
measure is largest for petty crimes, moderate for more serious crimes, and negligible for the most serious crimes. Because
the statistic is related to the expected size of the interacting group of criminals, their results suggest that more serious
crimes are associated with smaller group interactions.9

3. Research design

Our research designs identify the extent to which similarities in the tax planning of firms in the same industry are the
result of their strategic response to their industry competitors’ tax planning. We develop a model of corporate tax planning
that is similar to the specification used in prior corporate finance studies that examine within-industry strategic interactions
(e.g., Rajan and Zingales, 1995; Leary and Roberts, 2014). In particular, we model a firm’s tax planning as a function of (i) its
own firm-level characteristics, which are likely to be similar to those of other firms in its industry (i.e., correlated effects),
(ii) its common industry-level characteristics, which are shared by the other firms in its industry (i.e., exogenous effects),
and (iii) its industry competitors’ tax planning (i.e., outcome-on-outcome effects). This last effect is our object of interest,
and it represents a firm’s strategic response to its industry competitors’ tax planning.
As noted in Section 2, it is challenging to empirically identify direct (i.e., outcome-on-outcome) strategic reactions be-
cause of the risk of correlated omitted variables (Angrist, 2014; Gormley and Matsa, 2014; Leary and Roberts, 2014). Further,
the reflection problem implies that even in the absence of correlated omitted variables, the simultaneous nature of strate-
gic reactions—firms responding to their competitors, and vice versa—produces near perfect collinearity between a firm’s
characteristics and its competitors’ tax planning (Manski, 1993; Blume and Durlauf, 2005; Blume et al., 2015).
To break the simultaneity inherent in the reflection problem we develop a partially-overlapping groups research design
that uses changes in tax policy that directly affect one group of firms as an exogenous source of variation that should
indirectly affect the tax planning of partially-overlapping members of the other group (Bramoullé et al., 2009; De Giorgi
et al., 2010). To the extent the tax policy changes are exogenous with respect to the members of the other group, they should
have no direct effect on the overlapping group members, thus providing a way to break the simultaneity inherent in the
reflection problem. We use this research design in two different research settings: one that involves an exogenous increase
and one that involves and exogenous decrease in certain firms’ tax planning. Importantly, by examining both an exogenous
increase and an exogenous decrease in tax planning, we can assess whether there is symmetry in any strategic effects that
we document (e.g., whether there are strategic complementarities for both increases and decreases in tax planning).

3.1. Partially-overlapping groups difference-in-differences design: Irish-haven

Our first research setting uses significant corporate tax rate cuts in Ireland as an arguably exogenous source of variation
in firms’ tax planning. In December of 1997, Ireland’s finance minister introduced legislation that implemented a phased
reduction of the then 32% corporate tax rate to 12.5%, by January 1, 2003. This led to a large increase in the tax planning of
multinational U.S. firms with operations in Ireland (“Irish-haven firms”), as many of these firms shifted income to Ireland to
take advantage of the lower tax rate. To the extent Irish-haven firms have industry competitors without Irish operations—
and who were therefore not directly affected by the Irish tax cuts—these competitors should alter their tax planning if there
are strategic complementarities in corporate tax planning.
In this research setting, the first group consists of Irish-haven firms, which partially overlap with the second group—
wholly-domestic firms with only U.S. operations—based on shared industry membership. Changes in Irish tax rates should

9
See also Graham (2008), who proposes a method to identify social interactions using conditional variance restrictions.

Please cite this article as: C.S. Armstrong, S. Glaeser and J.D. Kepler, Strategic reactions in corporate tax planning, Journal
of Accounting and Economics, https://doi.org/10.1016/j.jacceco.2019.03.003
JID: JAE
ARTICLE IN PRESS [m3Gsc;August 20, 2019;11:56]

C.S. Armstrong, S. Glaeser and J.D. Kepler / Journal of Accounting and Economics xxx (xxxx) xxx 7

Fig. 1. Partially overlapping groups research design.


This figure illustrates the sample construction for our partially-overlapping groups research design used in our Irish-haven difference-in-differences analysis,
consisting of three groups: (i) Irish-haven industry firms with an Irish subsidiary (withheld from our sample), (ii) domestic firms in an industry with Irish-
haven competitor (treatment group), and (iii) domestic firms without an Irish-haven competitor (control group).

have no direct effect on the tax planning of wholly-domestic firms, but should have an indirect effect to the extent these
firms strategically respond to changes in their Irish-haven competitors’ tax planning. Wholly-domestic firms that partially
overlap with Irish-haven firms via common industry membership form the treatment group, and wholly-domestic firms with
no overlap with Irish-haven firms form the control group. Fig. 1 graphically illustrates how treatment and control groups
are constructed in this setting (as well as which firms are explicitly withheld from our sample in this research design).
To ensure that this test isolates strategic reactions to changes in Irish-haven firms’ tax planning, we only include firms
without foreign assets or foreign sources of income in the sample (i.e., both “treatment” and “control” firms are wholly
domestic). This restriction ensures that the firms in this test are not directly affected by the Irish tax cuts, but are only
indirectly affected through their strategic reactions to their directly exposed industry competitors. We then estimate the
following difference-in-differences specification:

Tax Planningijt = α + β1 Treated Irish Exposureij + β2 Post Rate Cut × Treated Irish Exposureij + θ Firm Controlsit−1
+ ϕ Industry Controls−ijt−1 + δt + εijt , (2)

where Tax Planning is one of several alternative measures of corporate tax planning that we describe in more detail below,
and the i and j subscripts again refer to firms and industries, and the t subscript refers to years. Firm Controls is a vector of
the determinants of tax planning drawn from prior work (e.g., Hanlon and Heitzman, 2010; Armstrong et al., 2012), which
we define in Appendix A. We measure all controls at the end of the fiscal year. Industry Controls are the industry-year
market-weighted average of the determinants of tax planning, calculated excluding firm i.
In our initial estimation of Eq. (2), we include year fixed effects, δ t , to control for common shocks in each year and
general time trends in tax planning (Dyreng et al., 2017). We also separately estimate Eq. (2) with industry-year and firm
fixed effects to assess whether similar inferences obtain using within-firm and within-industry-year variation, respectively.
However, our primary inferences are based on estimates from the specification that includes only year fixed effects, as the
construct of strategic reactions includes a cross-sectional component. Accordingly, including only year fixed effects preserves
most of the relevant variation and provides a more powerful test of strategic effects, albeit at the expense of not confining
the analyses to within-firm and within-industry-year variation.10
We include an indicator, Treated Irish Exposure, which equals one if the firm has industry competitors with Irish oper-
ations. Treated Irish Exposure controls for time-invariant differences between firms with Irish-haven competitors and firms
without Irish-haven competitors. We use the Exhibit 21 data described in Dyreng and Lindsey (2009) to identify firms with

10
Roberts and Whited (2013, p. 559) caution against “overusing” fixed effects because (i) using the economic magnitude of pooled OLS coefficient es-
timates are often more conducive for drawing qualitative inferences, (ii) including fixed effects can exacerbate measurement error, and (iii) certain fixed
effects structures may inadvertently eliminate the variation that is of inherent interest to a particular research question.

Please cite this article as: C.S. Armstrong, S. Glaeser and J.D. Kepler, Strategic reactions in corporate tax planning, Journal
of Accounting and Economics, https://doi.org/10.1016/j.jacceco.2019.03.003
JID: JAE
ARTICLE IN PRESS [m3Gsc;August 20, 2019;11:56]

8 C.S. Armstrong, S. Glaeser and J.D. Kepler / Journal of Accounting and Economics xxx (xxxx) xxx

Irish-haven operations.11 Post Rate Cut is an indicator that equals one after 1998 and delineates the period after indirectly
affected firms had an opportunity to react to Irish-haven firms’ responses to the December 1997 announcement that Ireland
would progressively lower its corporate tax rate.12 The coefficient on Post Rate Cut × Treated Irish Exposure in Eq. (2) captures
the component of firms’ tax planning that represents a strategic reaction to their industry competitors’ tax planning.
A positive coefficient on Post Rate Cut × Treated Irish Exposure implies that wholly-domestic firms with Irish-haven in-
dustry competitors decreased their tax planning following the Irish tax cuts, relative to their wholly-domestic counterparts
without Irish-haven industry competitors (as we discuss in Section 4, all of our measures of Tax Planning are decreasing
in the level of a firm’s tax planning). Because wholly-domestic firms with Irish-haven industry competitors witnessed an
increase in their industry competitors’ expected tax planning, a positive relation would be evidence of strategic substitution
in corporate tax planning. Conversely, a negative relation would be evidence of strategic complementarities in corporate tax
planning. Finally, failing to find a relation would be evidence that firms make their tax planning decisions in a non-strategic
fashion, or that our tests lack sufficient power to reliably detect any strategic reactions.

3.2. Partially-overlapping groups differences-in-differences design: Delaware-haven

Our second research setting uses the staggered adoption of state tax policies designed to curb inter-state tax planning as
an arguably exogenous source of variation in affected firms’ tax planning. This research setting complements the Irish setting
in that it entails a decrease rather than an in increase in firms’ tax planning. The collective results from these two different
settings allows us to assess whether there is symmetry in firms’ strategic reactions to their competitors’ tax planning (e.g.,
whether firms respond to both increases and decreases in their competitors’ tax planning). Moreover, examining two distinct
sources of exogenous variation in firms’ tax planning also allows us to draw stronger inferences and reduces the likelihood
that our results are either spurious or due to omitted correlated variables. It also allows us to estimate multiple local average
treatment effects (LATEs), enhancing the generalizability of our inferences (Glaeser and Guay, 2017).
We follow Dyreng et al. (2013), who describe and document Delaware’s role as a domestic tax haven. In particular,
Dyreng et al. explain how firms can transfer intangible assets to Delaware subsidiaries and make royalty payments from non-
Delaware subsidiaries for the use of these assets. Because Delaware does not tax income earned from intangible assets, these
royalty payments are not subject to Delaware taxation. However, since these royalty payments are deductible, they reduce
the taxable income of non-Delaware subsidiaries. Dyreng et al. (2013) refer to this income-shifting strategy as the Delaware
Passive Investment Company, or Delaware PIC, strategy. They argue that firms with a large number of Delaware subsidiaries
and valuable intangible assets, or Delaware PIC firms, are likely to be taking advantage of this strategy. Consistent with their
conjecture, they find that Delaware PIC firms have state effective tax rates that are between 15 and 24 percent lower than
otherwise similar firms.
Dyreng et al. (2013) also discuss how states have responded to the erosion of their tax base from the widespread use
of the Delaware PIC strategy. In particular, they explain how states either started to require combined reporting or enacted
the economic nexus doctrine to curtail firms’ ability to use the Delaware PIC strategy. Combined reporting requires firms to
include the net profits of all of their subsidiaries in a combined tax return, preventing—or, at a minimum, severely limiting—
the inter-company transfer payments that are essential for the Delaware PIC strategy. The economic nexus doctrine requires
firms to pay taxes based on their economic presence (or “nexus”), which allows states to tax the royalty income that would
otherwise escape taxation by being transferred to Delaware. Dyreng et al. (2013) find that the adoption of combined report-
ing or the economic nexus doctrine led to large reductions in the state tax planning of Delaware PIC firms, as evidenced by
a relatively large increase in these firms’ state effective tax rates.
We use the staggered adoption of combined reporting and the economic nexus doctrine in various states at different
times as a series of exogenous shocks that curtailed the tax planning of Delaware PIC firms. Fig. 4 provides a map that illus-
trates the timing of states’ adoption of these measures. To the extent that Delaware PIC firms had competitors that did not
have Delaware subsidiaries—and therefore were not directly affected by the changes in the state tax policies—these competi-
tors should have altered their tax planning if there is a strategic component of corporate tax planning. In this application of
a partially-overlapping groups research design, the first group consists of Delaware PIC firms that partially overlap with the
second group, non-Delaware subsidiary firms, based on shared industry membership. Firms with no Delaware subsidiaries
(i.e., no direct exposure) that have Delaware PIC firms in their industry (i.e., indirect exposure) form the treatment group,
and firms with neither Delaware subsidies nor Delaware PIC firms in their industry (i.e., neither direct nor indirect exposure)
form the control group.
Finally, we aggregate subsidiaries at the firm level for this analysis rather than use disaggregated subsidiary-firm-level
data. We do so for several reasons. First, subsidiary-level measures of tax planning are not available. Second, a partially-
overlapping groups design at the subsidiary level would require the exclusion of firms with Delaware subsidiaries (analogous
to our exclusion of firms with foreign operations in the Irish setting). However, since the vast majority of U.S. corporations

11
Available on Scott Dyreng’s website: https://sites.google.com/site/scottdyreng/Home/data- and- code. We thank the authors for making these data pub-
licly available.
12
A prior version of this paper used 1997 to delineate the period after indirectly firms had an opportunity to react, but we now use 1998 due to expected
readjustment timing for strategic reactions to take place. Our main results are qualitatively unchanged under this prior measurement choice.

Please cite this article as: C.S. Armstrong, S. Glaeser and J.D. Kepler, Strategic reactions in corporate tax planning, Journal
of Accounting and Economics, https://doi.org/10.1016/j.jacceco.2019.03.003
JID: JAE
ARTICLE IN PRESS [m3Gsc;August 20, 2019;11:56]

C.S. Armstrong, S. Glaeser and J.D. Kepler / Journal of Accounting and Economics xxx (xxxx) xxx 9

have Delaware subsidiaries, excluding these firms would result in a highly restricted sample that would severely limit the
generalizability of our inferences.13
Following Dyreng et al. (2013), we use the location of a firm’s subsidiaries in various states to measure the firm’s
economic presence in those states. Although we do not exclude firms with Delaware subsidiaries, we do exclude these
firms’ Delaware subsidiaries from the analysis.14 We estimate the following differences-in-differences specification at the
firm level:
 
State ET Ri jt = α + β1 Avg Competitor DE PICi jt + β2 Avg(Competitor Post Lawist )
 
+ β3 Avg Competitor DE P ICi jt × Competitor Post Lawist
+ β4 Avg(Own DE P ICist ) + β5 Avg(Own Post Lawist )
+ β6 Avg(Own DE P ICist × Own Post Lawist ) + θ F irm Control sit−1
+ ϕ Industry Control s−i jt−1 +  Avg(State Control sit−1 )
 
+ Avg Competitor State Control s−i jt−1 + δrt + εi jt , (3)

We include an indicator, Competitor DE PIC, that equals one if the subsidiary has an in-state industry competitor that is
capable of utilizing the Delaware PIC strategy. We assume that a firm is capable of utilizing the Delaware PIC strategy if it
has both a relatively large number of Delaware subsidiaries (those in the upper third of the sample in a given year) and
relatively high intangible assets (those with above median market-to-book ratios in a given year). Our definition follows that
of Dyreng et al. (2013), except that we do not require a minimum number of subsidiaries in states that require combined
reporting or follow the economic nexus doctrine. Instead, we include an indicator, Post Law, that equals one if the state
requires combined reporting, follows the economic nexus doctrine, or both.15
We allow the year fixed effects in Eq. (3) to differ for each of the eight U.S. Bureau of Economic Analysis (BEA) regions
of firms’ headquarters to ensure that the specification compares firms with similar geographic activity (δ rt ).16 We include a
vector of state controls,  , which includes GSP Growth, State Corporate Tax Rate, and State Personal Tax Rate (Armstrong et al.,
2018b). We also require industry competitors to operate in the same state for Competitor DE PIC to equal one because we
expect all of the mechanisms that we consider to operate based on shared tax jurisdictions.17 We also augment Eq. (3) with
state-year, industry-year, and firm fixed effects—both separately and simultaneously—to assess the sensitivity of our infer-
ences to using variation within firms, industry-years, and state-years, respectively. However, we continue to estimate subse-
quent variants of Eq. (3) for our mechanism, falsification, and robustness tests using region-year fixed effects to ensure that
the specification captures the cross-sectional component of strategic effects.
The prefix Avg refers to subsidiary-weighted averaging across non-Delaware subsidiaries. For competitor variables, we
average across all non-Delaware subsidiaries in the firm’s industry, excluding the firm’s own subsidiaries. For variables with
the prefix Own, we average over the firm’s own non-Delaware subsidiaries. We then take the subsidiary-weighted average
of the interaction terms after calculating the interactions.
The coefficients on Avg(Own DE PIC) and Avg(Own Post Law) measure whether the firm is capable of taking advantage of
the Delaware PIC strategy and the firm’s direct exposure to various states’ attempts to neutralize the strategy, respectively.
The coefficient on Avg(Own DE PIC × Own Post Law) measures the traditional differences-in-differences estimate of the ef-
fect of states’ policies to neutralize the Delaware PIC strategy on directly targeted firms’ tax planning. The coefficients on
Avg(Competitor DE PIC) and Avg(Competitor Post Law) measure the extent to which firms’ competitors are capable of taking
advantage of the Delaware PIC strategy and competitors’ exposure to state measures to neutralize the strategy, respectively.
Avg(Competitor DE PIC × Competitor Post Law) is the main variable of interest in Eq. (3), and its coefficient measures
whether firms’ tax planning has a strategic component that is a direct response to their industry competitors’ tax planning.
A positive coefficient implies that firms with industry competitors that are likely taking advantage of the Delaware PIC
strategy also decreased their own tax planning following the enactment of state policies designed to neutralize the strategy,
despite not being directly affected by these policies. Because these firms witnessed a decrease in their industry competitors’
expected tax planning, a positive relation would be evidence of strategic complementarities in corporate tax planning, and
a negative relation would be evidence of strategic substitutions in corporate tax planning. No relation would indicate that

13
Approximately 70% of the firms in our sample have at least one Delaware subsidiary, which is slightly more than the approximately 60% of firms in
Dyreng et al.’s (2013) sample.
14
Dyreng et al. (2013) find that about half of all subsidiaries are located in Delaware, which suggests that the proportion of Delaware subsidiaries does
not reflect firms’ true economic presence. To the extent that the location of firms’ other subsidiaries also does not reflect their true economic presence, the
resulting measurement error, if anything, should bias against finding evidence of either positive or negative strategic reactions.
15
Note that economic nexus and combined reporting requirements are not necessarily laws in the statutory sense. However, we refer to them as “laws”
for concision and because Post Law is common nomenclature in differences-in-differences research designs that examine the effect of laws, regulations, or
court rulings.
16
We are able to include Post Law and region-year fixed effects together because different states began to require combined reporting or began to follow
the economic nexus doctrine at different times (i.e., there is staggered adoption).
17
For example, we expect enforcement herding in this setting to operate via the actions of common state tax regulators, customers, media, and other
stakeholders (e.g., a firm that only has operations in California will not expect changes in its competitors’ tax planning in New York to affect the probability
it is targeted by the California Franchise Tax Board).

Please cite this article as: C.S. Armstrong, S. Glaeser and J.D. Kepler, Strategic reactions in corporate tax planning, Journal
of Accounting and Economics, https://doi.org/10.1016/j.jacceco.2019.03.003
JID: JAE
ARTICLE IN PRESS [m3Gsc;August 20, 2019;11:56]

10 C.S. Armstrong, S. Glaeser and J.D. Kepler / Journal of Accounting and Economics xxx (xxxx) xxx

firms make their tax planning decisions in a non-strategic fashion, without regard to their industry competitors’ expected
tax planning, or that our tests lack sufficient power to reliably detect any strategic reactions.

4. Variable measurement and sample construction

4.1. Measurement of corporate tax planning

As noted by Slemrod (1993) and Hanlon and Heitzman (2010), there is no universally accepted definition of what con-
stitutes tax planning. Conventional use of the term is broad and encompasses a range of activities that are both legal (e.g.,
corporate debt policy, transfer pricing decisions, and permanently reinvested earnings) and illegal (e.g., tax evasion and
fraudulent reporting). Following most prior work, we adopt this relatively expansive notion of tax planning and examine
multiple alternative measures that are common in the economics, finance, and accounting literatures (e.g., Chen et al., 2010;
Armstrong et al., 2012; Lisowsky et al., 2013; Hasan et al., 2014; Dyreng et al., 2017; Cen et al. 2017, 2018). Collectively, these
measures should capture a broad range of different activities that are symptomatic of tax planning.
We use several alternative measures of corporate effective tax rates (ETRs) to measure Tax Planning. First, we use Cash
ETR, which is total taxes paid in cash during the year divided by pretax income. Second, we use GAAP ETR, which is total
income tax expense divided by pretax income (Dyreng, et al., 2008). For tests in the Delaware research setting that examine
variation in tax planning at the state level, we use State ETR, which is total state tax expense divided by pre-tax income
(Dyreng et al., 2013). Finally, in subsequent tests, we repeat our analyses using firms’ three- and five-year Cash ETRs, GAAP
ETRs, and State ETRs to capture tax planning activities that might manifest over longer horizons (Dyreng et al., 2008).

4.2. Sample construction

Our sample selection procedure begins with all listed firms in Compustat for fiscal years 1993 through 2013 for which
we have the necessary data to compute each of our primary tax planning measures and control variables. Following
Dyreng et al. (2017, 2016) and others, we exclude firm-years with negative pretax income and, where necessary, we truncate
Cash ETR, GAAP ETR, and State ETR so that they lie in the unit interval (i.e., between zero and one).18 We require non-missing
control and dependent variables for all of our tests, and we lag all firm- and industry-level control variables one year. In the
Irish research setting, we restrict the sample to include only wholly-domestic U.S. firms (i.e., those without foreign assets or
income). Following prior work on strategic reactions and peer effects in corporate behavior, we further restrict the sample
to include only firms with at least 10 listed firms in their three-digit SIC industry and that are listed in the CRSP database
during the year (e.g., Leary and Roberts, 2014). Finally, we winsorize all firm- and industry-level continuous variables at
the 1st and 99th percentiles. The final sample for our Irish research setting consists of 21,490 firm-year observations and
spans the period 1993 to 2013 and includes 138 unique industries measured using three-digit historical SIC codes. The final
sample for our Delaware research setting consists of 8618 firm-year observations.19
Panels A and B of Table 1 present descriptive statistics for variables used in our Irish and Delaware research settings,
respectively. The distributions of the firm-, industry-, and state-level variables for our sample are similar to those in prior
studies of corporate tax planning (e.g., Armstrong et al., 2012; Dyreng et al., 2013; Dyreng et al., 2017, 2016). For example,
the average GAAP ETR for our Irish sample is 28%, and the average Cash ETR is 24%. There are some differences between the
averages in Panel A1 and the industry averages in Panel A2 (e.g., average industry ROA is 8% whereas average firm-level ROA
is 6%). These differences arise from our use of market value-weighted industry averages, while the averages in Panel A are
implicitly equally weighted.

5. Results

We test for evidence of strategic reactions in two distinct research settings. We first examine the Irish-haven setting and
second we examine the Delaware-haven setting.

5.1. Irish-haven difference-in-differences analysis

5.1.1. Irish-haven strategic reactions


Table 2 presents results from estimating Eq. (2), which models tax planning in a partially-overlapping groups difference-
in-differences analysis around the enactment of significant cuts in Irish corporate tax rates. Panel A presents results for Cash
ETR. The coefficient on Post Rate Cut × Treated Irish Exposure in column (1) implies that wholly-domestic firms with Irish-
haven industry competitors responded to their competitors’ increased tax planning by reducing their own Cash ETR by 3.2
percentage points after the enactment of the Irish tax cuts (t-stat of –2.26). This reduction is as compared to the concurrent

18
We truncate 1313 observations to equal zero and 575 observations to equal one in our final sample that is reported in Panel A of Table 1.
19
We require firms to have non-missing pre-tax domestic income and state income tax expense in the Delaware-haven sample, which significantly
reduces the sample size relative to that in the Irish-haven analysis.

Please cite this article as: C.S. Armstrong, S. Glaeser and J.D. Kepler, Strategic reactions in corporate tax planning, Journal
of Accounting and Economics, https://doi.org/10.1016/j.jacceco.2019.03.003
JID: JAE
ARTICLE IN PRESS [m3Gsc;August 20, 2019;11:56]

C.S. Armstrong, S. Glaeser and J.D. Kepler / Journal of Accounting and Economics xxx (xxxx) xxx 11

Table 1
Descriptive Statistics. This table presents descriptive statistics for variables used in our tests. Panels A1 and A2 report descriptive statistics for the firm-
level and market value-weighted peer-firm average (excluding firm i’s own value) variables, respectively, for the sample for the Irish-haven tests. Panel
A3 reports differences in means for firm level characteristics between treatment and control firms for our Irish-haven tests. Panels B1 and B2 report
descriptive statistics for the firm-level and market value-weighted peer-firm average (excluding firm i’s own value) variables, respectively, for the sample
for the Delaware-haven tests.

Panel A1. Irish-haven sample: firm characteristics


Tax planning measures N Mean SD P5 P10 P25 P50 P75 P90 P95

Cash ETR 21,490 0.21 0.22 0.00 0.00 0.00 0.17 0.34 0.46 0.59
GAAP ETR 21,490 0.27 0.19 0.00 0.00 0.05 0.34 0.39 0.42 0.47
3-Year Cash ETR 18,987 0.24 0.22 0.00 0.00 0.03 0.24 0.36 0.46 0.61
5-Year Cash ETR 14,916 0.25 0.22 0.00 0.00 0.03 0.26 0.36 0.47 0.65
3-Year GAAP ETR 19,250 0.28 0.20 0.00 0.00 0.10 0.34 0.39 0.42 0.50
5-Year GAAP ETR 15,502 0.28 0.20 0.00 0.00 0.09 0.34 0.39 0.43 0.55
Treatment Variables N Mean SD P5 P10 P25 P50 P75 P90 P95

Treated Irish Exposure 21,490 0.84


Post Rate Cut 21,490 0.61
Control Variables N Mean SD P5 P10 P25 P50 P75 P90 P95

Book Leverage 21,490 0.23 0.22 0.00 0.00 0.02 0.19 0.38 0.54 0.63
Market Leverage 21,490 0.23 0.23 0.00 0.00 0.01 0.16 0.38 0.57 0.69
ROA 21,490 0.06 0.12 –0.09 –0.02 0.02 0.06 0.11 0.18 0.24
New Investment 21,490 0.06 0.12 –0.05 –0.03 –0.01 0.03 0.10 0.20 0.29
Log(Market Value of Equity) 21,490 5.50 2.01 2.33 2.86 4.00 5.47 6.90 8.11 8.99
Book-to-Market 21,490 0.94 0.62 0.22 0.32 0.54 0.84 1.17 1.60 1.95
Change in Goodwill 21,490 0.01 0.05 0.00 0.00 0.00 0.00 0.00 0.03 0.09
Net Operating Loss 21,490 0.37
Placebo Test Outcomes N Mean SD P5 P10 P25 P50 P75 P90 P95

Pretax Cash Flow 19,851 0.06 0.20 –0.07 –0.02 0.01 0.04 0.09 0.16 0.22

Panel A2. Irish-haven sample: industry average characteristics


Industry Average Control Variables N Mean SD P5 P10 P25 P50 P75 P90 P95

Industry Average Book Leverage 21,490 0.24 0.12 0.08 0.10 0.15 0.23 0.32 0.42 0.45
Industry Average Market Leverage 21,490 0.18 0.12 0.04 0.04 0.07 0.15 0.27 0.34 0.40
Industry Average ROA 21,490 0.08 0.04 0.02 0.04 0.05 0.08 0.11 0.14 0.16
Industry Average New Investment 21,490 0.07 0.06 –0.01 –0.01 0.02 0.06 0.11 0.14 0.16
Industry Average Log(Market Value of Equity) 21,490 8.47 1.07 6.48 7.02 7.71 8.63 9.30 9.78 9.93
Industry Average Book-to-Market 21,490 0.67 0.36 0.27 0.32 0.45 0.62 0.82 1.05 1.18
Industry Average Change in Goodwill 21,490 0.02 0.02 0.00 0.00 0.00 0.01 0.02 0.04 0.05
Industry Average Net Operating Loss 21,490 0.40 0.22 0.09 0.12 0.22 0.39 0.55 0.69 0.78

Panel A3. Irish-haven sample covariate balance


Mean

Control Variables Treatment Control Diff.

Book Leverage 0.23 0.24 –0.01


Market Leverage 0.22 0.25 –0.03
ROA 0.06 0.07 –0.01
New Investment 0.06 0.05 0.01
Log(Market Value of Equity) 5.53 5.39 0.14
Book-to-Market 0.92 1.00 –0.08∗
Change in Goodwill 0.01 0.02 –0.01
Net Operating Loss 0.38 0.36 0.02

Panel B1. Delaware-haven sample: firm characteristics


Tax Planning Measures N Mean SD P5 P10 P25 P50 P75 P90 P95

State ETR 8618 0.03 0.05 0.00 0.00 0.00 0.02 0.04 0.06 0.09
Treatment Variables N Mean SD P5 P10 P25 P50 P75 P90 P95

Avg(Competitor DE PIC × Post Law) 8618 0.11 0.35 0.00 0.00 0.00 0.00 0.04 0.24 0.55
Avg(Competitor DE PIC) 8618 0.03 0.06 0.00 0.00 0.00 0.00 0.03 0.08 0.14
Avg(Competitor Post Law) 8618 0.09 0.14 0.00 0.00 0.00 0.03 0.13 0.27 0.38
Avg(Own DE PIC) 8618 0.09 0.22 0.00 0.00 0.00 0.00 0.00 0.40 0.58
Avg(Own Post Law) 8618 0.45 0.36 0.00 0.00 0.17 0.38 0.71 1.00 1.00
Avg(Own DE PIC × Post Law) 8618 0.07 0.19 0.00 0.00 0.00 0.00 0.00 0.29 0.50
Firm Controls N Mean SD P5 P10 P25 P50 P75 P90 P95

Book Leverage 8618 0.21 0.17 0.00 0.00 0.05 0.20 0.32 0.44 0.53
Market Leverage 8618 0.20 0.20 0.00 0.00 0.03 0.15 0.31 0.49 0.63
ROA 8618 0.05 0.09 –0.11 –0.04 0.01 0.05 0.09 0.14 0.18
(continued on next page)

Please cite this article as: C.S. Armstrong, S. Glaeser and J.D. Kepler, Strategic reactions in corporate tax planning, Journal
of Accounting and Economics, https://doi.org/10.1016/j.jacceco.2019.03.003
JID: JAE
ARTICLE IN PRESS [m3Gsc;August 20, 2019;11:56]

12 C.S. Armstrong, S. Glaeser and J.D. Kepler / Journal of Accounting and Economics xxx (xxxx) xxx

Table 1
(continued)

Tax Planning Measures N Mean SD P5 P10 P25 P50 P75 P90 P95

New Investment 8618 0.07 0.10 –0.03 –0.01 0.01 0.04 0.10 0.18 0.26
Foreign Assets 8618 0.35
Book–to–Market 8618 0.94 0.54 0.29 0.37 0.56 0.84 1.18 1.58 1.92
Change in Goodwill 8618 0.02 0.06 0.00 0.00 0.00 0.00 0.02 0.07 0.13
Net Operating Loss 8618 0.70
Log(Market Value of Equity) 8618 7.21 1.91 3.89 4.75 5.97 7.17 8.47 9.69 10.40
State Controls N Mean SD P5 P10 P25 P50 P75 P90 P95

GSP Growth 8618 0.04 0.03 0.01 0.01 0.02 0.04 0.06 0.09 0.09
State Corporate Tax Rate 8618 0.01 0.02 0.00 0.00 0.00 0.01 0.02 0.04 0.05
State Personal Tax Rate 8618 0.04 0.03 0.00 0.01 0.02 0.03 0.05 0.08 0.10
Placebo outcome N Mean SD P5 P10 P25 P50 P75 P90 P95
Pretax Domestic Income 8616 0.10 0.10 –0.05 0.00 0.05 0.10 0.15 0.21 0.25

Panel B2. Delaware-haven sample: industry average characteristics


Industry Average Control Variables N Mean SD P5 P10 P25 P50 P75 P90 P95

Industry Average Book Leverage 8618 0.19 0.13 0.00 0.01 0.10 0.17 0.26 0.37 0.45
Industry Average Market Leverage 8618 0.14 0.14 0.00 0.00 0.05 0.10 0.20 0.33 0.44
Industry Average ROA 8618 0.07 0.06 –0.01 0.00 0.03 0.07 0.10 0.14 0.16
Industry Average New Investment 8618 0.06 0.06 –0.01 0.00 0.01 0.06 0.10 0.14 0.17
Industry Average Foreign Assets 8618 0.32 0.33 0.00 0.00 0.01 0.22 0.52 0.91 1.00
Industry Average Book-to-Market 8618 0.63 0.42 0.00 0.01 0.37 0.58 0.84 1.15 1.38
Industry Average Change in Goodwill 8618 0.02 0.03 0.00 0.00 0.00 0.01 0.02 0.04 0.06
Industry Average Net Operating Loss 8618 0.63 0.36 0.00 0.00 0.34 0.71 1.00 1.00 1.00
Industry Average Log(Market Value of Equity) 8618 8.06 2.92 0.00 4.78 7.44 8.78 10.09 10.59 10.88

Table 2
Irish-haven difference-in-differences analysis.
This table reports OLS estimates of the following difference–in–differences specification for a sample of firms with operations in the United States only:

Tax Planningi jt = α + β1 T reatedi j + β2 T reat edi j × Postt + C ontrols + εi jt

Where Tax Planning is measured as either Cash ETR or GAAP ETR as defined in Appendix A and i, j, and t index firms, industries, and years, respectively.
Treated is an indicator equal to one if firm i has competitors—defined as firms with the same three-digit SIC code—in industry j that have operations in
Ireland and zero otherwise. Post is an indicator equal to one after 1998, which corresponds to the beginning of the progressive reduction of corporate tax
rates in Ireland. Controls is a vector of time-varying firm and industry-average characteristics as defined in Panel A of Table 1. For parsimony we do not
tabulate coefficients for the control variables and fixed effects. All variables are defined in Appendix A. t-statistics are reported below coefficient estimates
and are calculated based on robust standard errors clustered by industry. ∗ , ∗ ∗ , ∗ ∗ ∗ indicate statistical significance (two-sided) at the 0.1, 0.05, and 0.01
levels, respectively.

Panel A. Cash ETR


Dependent variable: Cash ETRijt Cash ETRijt Cash ETRijt Cash ETRijt
(1) (2) (3) (4)

Treated Irish Exposureij –0.057∗ 0.007 · ·


(–1.96) (0.61) · ·
Post Rate Cutt × Treated Irish Exposureij –0.032∗ ∗ –0.056∗ ∗ ∗ –0.024∗ ∗ ∗ –0.055∗ ∗
(–2.26) (–3.88) (–3.63) (–2.35)
Firm Controlsijt–1 Yes Yes Yes Yes
Industry Average Controls–ijt–1 Yes Yes Yes Yes
Year Fixed Effects Yes No No No
Industry-Year Fixed Effects No Yes No Yes
Firm Fixed Effects No No Yes Yes
Observations 21,490 21,490 21,490 21,490
R2 0.106 0.215 0.496 0.531

Panel B. GAAP ETR


Dependent variable: GAAP ETRijt GAAP ETRijt GAAP ETRijt GAAP ETRijt
(1) (2) (3) (4)

Treated Irish Exposureij –0.065∗ 0.024∗ ∗ ∗ · ·


(–1.82) (3.32) · ·
Post Rate Cutt × Treated Irish Exposureij –0.029∗ ∗ –0.068∗ ∗ ∗ –0.012∗ ∗ –0.042∗ ∗ ∗
(–2.19) (–5.70) (–2.17) (–2.98)
Firm Controlsijt–1 Yes Yes Yes Yes
Industry Average Controls–ijt–1 Yes Yes Yes Yes
Year Fixed Effects Yes No No No
Industry-Year Fixed Effects No Yes No Yes
Firm Fixed Effects No No Yes Yes
Observations 21,490 21,490 21,490 21,490
R2 0.128 0.308 0.642 0.667

Please cite this article as: C.S. Armstrong, S. Glaeser and J.D. Kepler, Strategic reactions in corporate tax planning, Journal
of Accounting and Economics, https://doi.org/10.1016/j.jacceco.2019.03.003
JID: JAE
ARTICLE IN PRESS [m3Gsc;August 20, 2019;11:56]

C.S. Armstrong, S. Glaeser and J.D. Kepler / Journal of Accounting and Economics xxx (xxxx) xxx 13

change in the Cash ETR of wholly-domestic firms with no Irish-haven industry competitors—but that are otherwise similar—
and therefore represents the strategic component of tax planning. The magnitude of this strategic response is economically
significant, as it represents 14.5% of the sample standard deviation of Cash ETR. Furthermore, this result is robust to the
inclusion of two-digit SIC code industry-year, firm, and industry-year and firm fixed effects in columns (2), (3), and (4),
respectively.
We find similar results for GAAP ETR in Panel B of Table 2. The coefficient on Post Rate Cut × Treated Irish Exposure in
column (1) implies that wholly-domestic firms with Irish-haven industry competitors had a relative reduction in their GAAP
ETRs of 2.9 percentage points following the Irish tax cuts (t-stat of –2.19). This relative decrease is economically significant,
as it represents 15.3% of the sample standard deviation of GAAP ETR. We draw similar inferences when we include two-
digit SIC code industry-year, firm, and both industry-year and firm fixed effects in columns (2), (3), and (4), respectively.
Multiplying the coefficients when Cash ETR (GAAP ETR) is the dependent variable by the sample average pre-tax income
implies that wholly-domestic firms with Irish-haven competitors reduced their tax expense by $6.5 million ($5.9 million)
relative to their wholly-domestic counterparts with no Irish-haven competitors following the Irish tax cuts.20 , 21
To estimate the direct effect of the Irish tax cuts on Irish-haven firms we estimate Eq. (2) to compare the sample of Irish-
haven firms (i.e., those with Irish operations) with a control sample of domestic-only firms with no Irish-haven industry
competitors. We include Post 1997, an indicator for years after 1997, to capture the direct effect of the reduction in Irish
corporate tax rates on Irish-haven firms’ ETRs. Panels A and B of Table 3 present results using Cash ETR and GAAP ETR to
measure Tax Planning, respectively. In each panel, we find that firms with Irish operations experienced significant reductions
in their effective tax rates following the reduction in Irish corporate tax rates. For example, in column (4) of Panels A and
B, which rely on within-firm variation in changes in effective tax rates, we find that firms with Irish operations experienced
a 5.2 and 6.9 percentage point reduction in their Cash ETR and GAAP ETR, respectively (t-stats of –1.81 and –3.53).
Comparing the results in Tables 2 and 3 suggests that the magnitude of strategic reactions is large. For example, the
results in Table 2, Panel B column (1) indicate that the indirect effect of the Irish tax cuts is a 2.9 percentage point reduction
in GAAP ETRs. The corresponding results in Table 3, Panel B, column (1) indicate that the direct effect of the Irish tax cuts
is a 4.3 percentage point reduction in GAAP ETRs. Together, these estimates imply that a one percentage point increase in
competitors’ tax planning results in an approximately 0.63 percentage point increase in firms’ tax planning (2.9/4.3).
We also take three steps to ensure that the maintained parallel trends assumption inherent in the difference-in-
differences design is satisfied in the Irish-haven setting. First, we assess covariate balance between treatment and control
groups in Table 1, Panel A3. Of the eight controls we examine, only Book-to-Market is statistically different between the
treatment and control firms at the 10% level.22 Second, we plot the average Cash ETR and GAAP ETR over time for the
treated and control firms in Fig. 2. For both measures we find that treated and control firms exhibit parallel trends prior to
the Irish tax cut, and that treated firms’ ETRs decrease relative to control firms’ ETRs after 1998.
Third, in addition to the non-parametric evidence in Fig. 2, we present regression evidence that treatment and control
firms shared parallel trends in Fig. 3. We estimate Eq. (2), including all controls and fixed effects, after replacing the Post
indicator with separate indicators for each year relative to the treatment year. We then plot the coefficient estimates and
95% confidence intervals for the interaction between Treatment and each indicator when using Cash ETR (Panel A) and GAAP
ETR (Panel B) as the dependent variable. We find no evidence of differential trends across the treatment and control groups
in the period before the Irish rate cut. This finding is consistent with the evidence in both Fig. 2 and Table 1, Panel A3 and
suggests that the parallel trends assumption is valid in this setting.

5.1.2. Irish-haven economic mechanism analysis


In this section, we consider several non-mutually exclusive potential mechanisms through which strategic complemen-
tarities in corporate tax planning might arise. We first examine whether these strategic complementarities are the result of
“leader-follower” dynamics within industries. Several recent studies suggest that firms might mimic or emulate the actions
of the perceived “leaders” in their industry (e.g., Leary and Roberts, 2014; Xiao, 2017).
To identify leaders and followers, we follow prior work and assume that larger firms are more likely to be industry lead-
ers (e.g., Leary and Roberts, 2014; Xiao, 2017). Larger firms arguably have more resources and stronger incentives to gather

20
Based on our sample average pre-tax income of $202 million.
21
A prior version of this paper used peers’ idiosyncratic returns as an instrument for peer tax planning (e.g., Leary and Roberts, 2014). Following Leary and
Roberts (2014), we calculated idiosyncratic returns from an industry-return adjusted market model (where industry is based on three-digit SIC code)
estimated for each firm on a rolling annual basis using historical monthly CRSP stock returns during the prior five years. We required a minimum of
twelve months of prior stock return data when estimating each regression. We draw similar inferences using this alternative identification strategy (first-
stage coefficient of 0.067 (0.140) and associated t-stat of 6.40 (10.57) and second-stage coefficient on peer tax planning of 0.143 (0.036) and associated
t-stat of 3.82 (2.13) when using Cash ETR (GAAP ETR) to measure tax planning. The data requirements for this test reduce our sample by 6433 observations
(23%). Relaxing these data requirements does not qualitatively affect our main inferences (coefficient estimate of –0.021 and corresponding t-stat of –2.03
for the results in Column (1) of Table 2 Panel A and coefficient estimate of –0.023 and corresponding t-stat of –2.45 for the results in Column (1) of
Table 2 Panel B after relaxing the data requirements).
22
Under the null of random assignment, about one out of every ten variables would be expected to statistically differ between the treatment and control
groups by chance.

Please cite this article as: C.S. Armstrong, S. Glaeser and J.D. Kepler, Strategic reactions in corporate tax planning, Journal
of Accounting and Economics, https://doi.org/10.1016/j.jacceco.2019.03.003
JID: JAE
ARTICLE IN PRESS [m3Gsc;August 20, 2019;11:56]

14 C.S. Armstrong, S. Glaeser and J.D. Kepler / Journal of Accounting and Economics xxx (xxxx) xxx

Table 3
Irish-haven direct effect.
This table reports OLS estimates of the following difference–in–differences specification for a sample of firms with Irish-haven operations compared to a
control sample of domestic-only firms with no Irish-haven industry competitors around the passage of the Irish corporate tax rate cuts announced in 1997:

Tax Planningi jt = α + β1 T reatedi j + β2 T reatedi j × Postt + Controls + εi jt

Where Tax Planning is measured as either Cash ETR or GAAP ETR as defined in Appendix A, and i, j, and t index firms, industries, and years, respectively.
Treated is an indicator equal to one if firm i has operations in Ireland as of 1997, and zero otherwise. Post is an indicator equal to one during years 1998
onward, which corresponds to the beginning of the progressive reduction of corporate tax rates in Ireland. Controls is a vector of time-varying firm and
industry-average characteristics as defined in Panel A of Table 1. For parsimony we do not tabulate coefficients for the control variables and fixed effects.
All variables are defined in Appendix A. t-statistics are reported below coefficient estimates and are calculated based on robust standard errors clustered
by industry. ∗ , ∗ ∗ , ∗ ∗ ∗ indicate statistical significance (two-sided) at the 0.1, 0.05, and 0.01 levels, respectively.

Panel A. Cash ETR


Dependent variable: Cash ETRijt Cash ETRijt Cash ETRijt Cash ETRijt
(1) (2) (3) (4)

Treated Irish Exposureij –0.0 0 0 0.008 · ·


(–0.01) (0.49) · ·
Post 1997t × Treated Irish Exposureij –0.029∗ ∗ –0.038∗ ∗ –0.028∗ –0.052∗
(–2.11) (–2.32) (–1.94) (–1.81)
Firm Controlsijt–1 Yes Yes Yes Yes
Industry Average Controls–ijt–1 Yes Yes Yes Yes
Year Fixed Effects Yes No No No
Industry-Year Fixed Effects No Yes No Yes
Firm Fixed Effects No No Yes Yes
Observations 8329 8329 8329 8329
R2 0.043 0.150 0.428 0.501

Panel B. GAAP ETR


Dependent variable: GAAP ETRijt GAAP ETRijt GAAP ETRijt GAAP ETRijt
(1) (2) (3) (4)

Treated Irish Exposureij –0.018∗ –0.001 · ·


(–1.91) (–0.07) · ·
Post 1997t × Treated Irish Exposureij –0.043∗ ∗ ∗ –0.068∗ ∗ ∗ –0.033∗ ∗ ∗ –0.069∗ ∗ ∗
(–3.90) (–5.25) (–2.62) (–3.53)
Firm Controlsijt–1 Yes Yes Yes Yes
Industry Average Controls–ijt–1 Yes Yes Yes Yes
Year Fixed Effects Yes No No No
Industry-Year Fixed Effects No Yes No Yes
Firm Fixed Effects No No Yes Yes
Observations 8329 8329 8329 8329
R2 0.085 0.185 0.508 0.569

private information about the optimal level of tax planning.23 Smaller firms might monitor their larger industry competi-
tors to infer their private information from their tax planning decisions. Evidence that smaller firms’ tax planning is more
responsive to the tax planning of their larger industry competitors would be consistent with leader-follower dynamics gen-
erating strategic complementarities in corporate tax planning. Therefore, in our first mechanism test we examine whether
there is a differential responsiveness of smaller and larger firms to their industry competitors’ tax planning.
Firms may also rationally herd together with respect to their tax planning in order to avoid unwanted public or regu-
latory scrutiny (i.e., enforcement herding). For example, a firm whose tax planning is noticeably more aggressive than the
“typical” level of tax planning in its industry may face a greater likelihood of an audit by tax authorities or invite tax poli-
cies aimed at curbing tax planning.24 Consequently, firms may be less willing to take an aggressive tax position if they
expect their competitors to curb their tax planning. Conversely, firms may be more inclined to adopt aggressive tax po-
sitions if they expect their competitors to engage in more aggressive tax planning. Indeed, competitive pressure may force
them to do so.25 We consider rational herding in corporate tax planning as another mechanism that could generate strategic
complementarities in corporate tax planning.

23
For example, large firms may invest relatively more in tax planning or work more closely with tax authorities to determine the maximum “acceptable”
amount of tax planning (Ayers et al., 2015). Furthermore, larger firms are more likely to have the legal resources and wherewithal to challenge unfavorable
tax enforcement outcomes, suggesting that larger firms are also more likely to establish a precedent that other firms can follow.
24
The notion of rational herding has been extensively studied in the finance, economics, and management literatures (e.g., Nofsinger and Sias, 1999;
Hirshleifer and Teoh, 2008). Rational herding refers to the phenomenon in which individual agents simultaneously converge to similar choices by examining
the same information, such as market prices, or exposure to the same institutional environments and regulatory pressures. A rational herding equilibrium
can result from firms inferring information from the actions of other agents or as the result of their payoffs increasing in the number of firms that take
the same action. Our notion of rational herding corresponds to the latter type of equilibrium, as the reputational or enforcement costs of tax planning are
arguably decreasing in industry-competitors’ tax planning.
25
For example, firms might have tax planning opportunities at the margin that they are reluctant to pursue if they are already more aggressive than
their industry competitors, but will do so if ever their competitors become more tax aggressive.

Please cite this article as: C.S. Armstrong, S. Glaeser and J.D. Kepler, Strategic reactions in corporate tax planning, Journal
of Accounting and Economics, https://doi.org/10.1016/j.jacceco.2019.03.003
JID: JAE
ARTICLE IN PRESS [m3Gsc;August 20, 2019;11:56]

C.S. Armstrong, S. Glaeser and J.D. Kepler / Journal of Accounting and Economics xxx (xxxx) xxx 15

Panel A. Cash ETR


0.28
0.27
0.26

Average Cash ETR


0.25
0.24
0.23
0.22
0.21
0.2
0.19
1994 1995 1996 1997 1998 1999 2000 2001 2002
Year
Contol Group Treatment Group

Panel B. GAAP ETR


0.32

0.31
Average GAAP ETR

0.3

0.29

0.28

0.27

0.26

0.25
1994 1995 1996 1997 1998 1999 2000 2001 2002
Year
Contol Group Treatment Group

Fig. 2. Irish-haven parallel trends (non-parametric pattern).


This figure presents average effective tax rates for wholly-domestic U.S. firms that have competitors with Irish-haven operations (Treatment Group) and
wholly-domestic U.S. firms that do not have competitors with Irish-haven operations (Control Group). Panel A plots average cash effective tax rates (Average
Cash ETR) and Panel B plots average GAAP effective tax rates (Average GAAP ETR). Ireland began a progressive reduction of its corporate tax rate in 1998.

Finally, given the complex, dynamic, and interconnected nature of tax laws, regulations, treaties, and judicial rulings,
firms may learn tax planning strategies from observing their competitors’ tax planning (Hanlon and Heitzman, 2010; Brown,
2011; Brown and Drake, 2013). For example, firms may observe their competitors reduce their tax burden via corporate
inversions to low-tax jurisdictions. These firms may then monitor the success of these tax planning strategies, as well as
learn how to take advantage of these strategies themselves, by studying their competitors. Anecdotal evidence also suggests
that tax authorities are concerned about firms’ ability to benchmark and reverse engineer the tax planning strategies of their
competitors, prompting some tax authorities to advocate “tax privacy” (Blank, 2014). We refer to this potential mechanism
as learning.
To test for evidence of learning and enforcement herding in corporate tax planning, we examine the strategic responses
of firms that have either higher or lower levels of tax planning than their industry competitors. Firms with relatively low
effective tax rates in a given industry are more likely to be engaging in tax planning. Therefore, evidence that firms with
higher effective tax rates respond to their industry competitors by lowering their effective tax rates via increased tax plan-
ning would suggest that learning is an important mechanism. Similarly, firms that are more aggressive than their competi-
tors are likely to be more sensitive to changes in enforcement and public scrutiny. Therefore, finding that firms with lower
effective tax rates respond to their industry competitors by changing their tax planning would be evidence that enforcement
herding is an important mechanism.
Although we conduct these tests for learning and enforcement simultaneously, we note that these two explanations are
not mutually exclusive. Moreover, these tests are not designed to assess whether strategic effects are larger or smaller for
one group of firms, but rather to determine whether there is a detectable effect for a given group. It is therefore possible to

Please cite this article as: C.S. Armstrong, S. Glaeser and J.D. Kepler, Strategic reactions in corporate tax planning, Journal
of Accounting and Economics, https://doi.org/10.1016/j.jacceco.2019.03.003
JID: JAE
ARTICLE IN PRESS [m3Gsc;August 20, 2019;11:56]

16 C.S. Armstrong, S. Glaeser and J.D. Kepler / Journal of Accounting and Economics xxx (xxxx) xxx

Panel A. Cash ETR


0.08
0.06
0.04
0.02
0.00
Cash ETR

-0.02
-0.04
-0.06
-0.08
-0.10
-0.12
1995 1996 1997 1998 1999 2000+
Year

Panel B. GAAP ETR


0.04

0.02

0.00

-0.02
GAAP ETR

-0.04

-0.06

-0.08

-0.10

-0.12

-0.14
1995 1996 1997 1998 1999 2000+
Year
Fig. 3. Irish-haven parallel trends (treatment effect timing).
This figure presents OLS regression coefficient estimates and two-tailed 95% confidence intervals based on standard errors clustered by industry after re-
estimating Eq. (2), including all controls and fixed effects, and replacing the Post indicator with separate indicators for each year relative to the treatment
year of 1998. Panel A presents results for Cash ETR and Panel B presents results for GAAP ETR.

find evidence consistent with both learning and enforcement herding in the same test. In contrast, the tests of the leader-
follower mechanism do imply a differential effect because they suggest that strategic effects will be concentrated among
smaller firms.
To operationalize our mechanism tests, we modify Eq. (2) by splitting the sample based on firm size and firms’ relative
level of tax planning. Panel A of Table 4 presents results from re-estimating Eq. (2) for the separate subsamples when using
Cash ETR as the measure of Tax Planning. In columns (1) and (2) we split our main sample based on firm size. Large Firm is
defined as an indicator equal to one if a firm’s size is greater than the industry average, and zero otherwise.26 Column (1)
presents results for large firms (Large Firm = 1) while column (2) presents results for small firms (Large Firm = 0). The results
suggest that small firms are not more sensitive to the planning of their industry competitors and are therefore inconsistent
with leader-follower dynamics producing strategic complementarities in corporate tax planning. If anything, the difference

26
We use mean rather than median firm size—which produces an unequal sample split with fewer firms above the sample mean—because it is unlikely
that half the sample firms are “leaders.”

Please cite this article as: C.S. Armstrong, S. Glaeser and J.D. Kepler, Strategic reactions in corporate tax planning, Journal
of Accounting and Economics, https://doi.org/10.1016/j.jacceco.2019.03.003
JID: JAE
ARTICLE IN PRESS [m3Gsc;August 20, 2019;11:56]

C.S. Armstrong, S. Glaeser and J.D. Kepler / Journal of Accounting and Economics xxx (xxxx) xxx 17

Fig. 4. Delaware experiment affected states.


Panel A presents the geographic distribution of states requiring combined reporting of corporate income (Dyreng et al., 2013). Seventeen U.S. states (grey)
adopted combined reporting prior to 1995 and seven states (striped) adopted combined reporting between 1995 and 2009. Panel B presents the geographic
distribution of states that require economic nexus of corporate income (Dyreng et al., 2013). Twenty-six U.S. states (grey) adopted the economic nexus rule
prior to 1995 and twelve states (striped) adopted the economic nexus rule between 1995 and 2008.

between the coefficients in columns (1) and (2) suggests that larger firms are more sensitive to changes in the tax planning
of their industry competitors (untabulated F-statistic of 9.06, p-value of <0.01).
Columns (3) and (4) of Panel A present the results from estimating Eq. (2) after splitting the sample based on whether
the firm exhibited less (Above Industry ETR = 1 in column (3)) or more (Above Industry ETR = 0 in column (4)) tax planning
than its competitors. We do find some evidence of learning: the results suggest that less aggressive firms are sensitive
to increases in their industry competitors’ tax planning (coefficient estimate of –0.023; t-stat of –1.75 in column (3)). We
also find evidence that suggests that enforcement herding induces strategic complementarities in firms’ tax planning. In
particular, firms that are more aggressive than their industry competitors—and are therefore arguably more concerned about
tax enforcement—are sensitive to changes in their competitors’ tax planning (coefficient estimate of –0.025; t-stat of –2.16
in column (4)).

Please cite this article as: C.S. Armstrong, S. Glaeser and J.D. Kepler, Strategic reactions in corporate tax planning, Journal
of Accounting and Economics, https://doi.org/10.1016/j.jacceco.2019.03.003
JID: JAE
ARTICLE IN PRESS [m3Gsc;August 20, 2019;11:56]

18 C.S. Armstrong, S. Glaeser and J.D. Kepler / Journal of Accounting and Economics xxx (xxxx) xxx

Table 4
Irish-haven economic mechanisms.
This table reports OLS estimates of the following difference–in–differences specification using sample restrictions based on the economic mechanisms as
defined in Appendix A:

Tax Planningi jt = α + β1 T reatedi j + β2 T reatedi j × Postt + Controls + εi jt

Where Tax Planning is measured as either Cash ETR or GAAP ETR as defined in Appendix A, and i, j, and t index firms, industries, and years, respectively.
Panel A presents results from using Cash ETR as a measure of Tax Planning. Panel B presents results from using GAAP ETR as a measure of tax planning.
Controls is a vector of time-varying firm and industry-average characteristics as defined in Panel A of Table 1. For parsimony we do not tabulate coefficients
for the control variables and fixed effects. For parsimony we do not tabulate coefficients for our control variables and estimated fixed effects. All variables
are defined in Appendix A. t-statistics are reported below coefficient estimates and are calculated based on robust standard errors clustered by industry. ∗ ,
∗∗ ∗∗∗
, indicate statistical significance (two-sided) at the 0.1, 0.05, and 0.01 levels, respectively.

Panel A. Industry leader, learning, and enforcement herding: Cash ETR


Dependent variable: Cash ETRijt

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

Treated Irish Exposureij 0.020 –0.060∗ ∗ –0.010 –0.048∗ ∗


(0.69) (–2.04) (–0.63) (–2.48)
Post Rate Cutt × Treated Irish Exposureij –0.135∗ ∗ ∗ –0.028∗ –0.023∗ –0.025∗ ∗
(–3.73) (–1.97) (–1.75) (–2.16)
Firm Controlsijt–1 Yes Yes Yes Yes
Industry Average Controls–ijt–1 Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes

Sample restriction Large Firm = Above Industry ETR =

1 0 1 0

Observations 1079 20,411 8163 13,327


R2 0.179 0.105 0.143 0.136

Panel B. Industry leader, learning, and enforcement herding: GAAP ETR


Dependent variable: GAAP ETRijt

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

Treated Irish Exposureij –0.045∗ –0.065∗ –0.007 –0.070∗ ∗


(–1.74) (–1.78) (–1.00) (–2.30)
Post Rate Cutt × Treated Irish Exposureij –0.071∗ ∗ –0.028∗ ∗ –0.006 –0.035∗ ∗
(–2.32) (–2.09) (–0.77) (–2.13)
Firm Controlsijt–1 Yes Yes Yes Yes
Industry Average Controls–ijt–1 Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes

Sample restriction Large Firm = Above Industry ETR =

1 0 1 0

Observations 1079 20,411 9581 11,909


R2 0.235 0.127 0.112 0.151

Panel B of Table 4 presents results using GAAP ETR as the measure of Tax Planning. We find no evidence of leader-follower
dynamics inducing strategic complementarities in corporate tax planning (untabulated F-statistic of 1.44, p value of 0.15). In
contrast to the results in Panel A, the results in Panel B provide no evidence that learning induces strategic complemen-
tarities in corporate tax planning. In particular, the estimates in column (3) provide no evidence that firms with relatively
high GAAP ETRs are sensitive to increases in their industry-competitors’ tax planning. However, we again find evidence that
enforcement herding produces strategic complementarities in corporate tax planning. The results in column (4) indicate that
tax aggressive firms respond in a similar fashion to a reduction in their competitors’ tax planning (coefficient estimate of
–0.035; t-stat of –2.13).

5.1.3. Irish-haven placebo analysis


Strategic reactions have a close analogue in the individual peer effects literature, which examines whether individual be-
haviors (e.g., smoking, drug use, and crime) propagate through social interactions.27 Cohen-Cole and Fletcher (2008) demon-
strate an important concern with studies in this literature by using the standard research design to document evidence of

27
The key difference between strategic reactions among industry competitors and social interactions in social networks is that social interactions diffuse
behavior via behavioral mechanisms or through tradeoffs between nonmonetary costs and benefits (e.g., a teenager who trades off the social costs of
not smoking when with a group of friends who smoke). In contrast, strategic reactions are rational, non-behavioral responses to competitors’ expected
decisions, which are the result of a monetary cost-benefit tradeoff.

Please cite this article as: C.S. Armstrong, S. Glaeser and J.D. Kepler, Strategic reactions in corporate tax planning, Journal
of Accounting and Economics, https://doi.org/10.1016/j.jacceco.2019.03.003
JID: JAE
ARTICLE IN PRESS [m3Gsc;August 20, 2019;11:56]

C.S. Armstrong, S. Glaeser and J.D. Kepler / Journal of Accounting and Economics xxx (xxxx) xxx 19

Table 5
Falsification tests: placebo analysis.
This table reports OLS estimates of the following difference–in–differences specification for a sam-
ple of firms with operations in the United States only:

Pretax Cash F lowi jt = α + β1 T reatedi j + β2 T reatedi j × Postt + Controls + εi jt

where i, j, and t index firms, industries, and years, respectively. Controls is a vector of time-varying
firm and industry-average characteristics as defined in Panel A of Table 1. For parsimony we do
not tabulate coefficients for the control variables and fixed effects. All variables are defined in
Appendix A. t-statistics are reported below coefficient estimates and are calculated based on ro-
bust standard errors clustered by industry. ∗ , ∗ ∗ , ∗ ∗ ∗ indicate statistical significance (two-sided) at
the 0.1, 0.05, and 0.01 levels, respectively.

Dependent variable: Pretax Cash Flowijt


(1)

Treated Irish Exposureij 0.008


(1.36)
Post Rate Cutt × Treated Irish Exposureij –0.011
(–1.51)
Firm Controlsijt–1 Yes
Industry Average Controls–ijt–1 Yes
Year Fixed Effects Yes
Observations 19,851
R2 0.140

“network diffusion” in non-social outcomes that should not exhibit such behavior, including height, acne, and headaches.28
Cohen-Cole and Fletcher’s (2008) approach suggests an analogous “falsification” test in our research setting. Specifically,
if strategic reactions in tax planning are an artefact of inadequately accounting for the endogenous relationship between
firms and their industry competitors, we should also find evidence of spurious strategic reactions in outcomes that seem
implausible.
We use firms’ pre-tax performance as an outcome that should not exhibit strategic reactions, since it seems unlikely that
changes in their industry competitors’ tax planning should produce changes in firms’ pre-tax performance.29 We re-estimate
Eq. (2) using firms’ Pretax Cash Flow, measured as pre-tax cash flow from operations scaled by beginning of year total assets,
as the dependent variable. We examine cash flows rather than earnings because there may be strategic complementarities
or peer effects in earnings management via accruals.
The results in Table 5 provide no evidence of strategic reactions in firms’ pre-tax performance. In particular, the results
in column (1) suggest that the enactment of significant Irish tax cuts had no detectable effect on the relative pre-tax income
of firms exposed to Irish-haven competitors (t-stat of –1.51 for the coefficient on Post Rate Cut × Treated Irish Exposure).

5.1.4. Irish-haven Hoberg and Phillips industry classification


A potential concern with our partially-overlapping groups research design is that it relies on an a priori classification
(or grouping) to define firms’ competitors (i.e., three-digit SIC codes). To ensure that our results are robust to alternative
classifications of a firm’s competitors, we re-estimate each of our primary analyses using Hoberg and Phillips (2010, 2016)
text-based industry classifications to identify competitors.30 We report the results in Table 6 and find that all of our primary
inferences continue to hold. We also find that the economic magnitudes of the main coefficients of interest are similar to
those of their counterparts in our earlier analyses that use three-digit SIC codes.

5.1.5. Irish-haven long-run tax planning


A potential concern with one-year ETRs as a measure of corporate tax planning is that they might exhibit volatility from
one year to the next for reasons unrelated to tax planning activities (Hanlon and Heitzman, 2010). Dyreng et al. (2008) ad-
dress this concern by calculating ETRs over a longer horizon, which mitigates potential measurement error due to

28
Cohen-Cole and Fletcher’s (2008) primary criticism of the literature on peer effects is that prior studies fail to sufficiently account for endogeneity. As
we discussed earlier, we address endogeneity concerns through our use of two distinct research settings that feature different arguably exogenous shocks
to competitors’ tax planning.
29
Alternatively, since an increase in competitors’ tax planning leads to an increase in their after-tax profitability, it could allow competitors to compete
more aggressively. In this case, increases in competitors’ tax planning may lead to a decline in a firm’s own pre-tax profitability. This, in turn, could cause
firms to alter their own tax planning to “keep up” with their competitors’ after-tax performance. We view this scenario as an alternative mechanism that
could cause strategic complementarities in corporate tax planning. Therefore, this falsification test also speaks to the existence of this alternative mech-
anism. Evidence that an increase in competitors’ tax planning causes a decrease in firms’ pre-tax performance would be consistent with this alternative
mechanism.
30
Available on Gerard Hoberg and Gordon Phillip’s website: http://hobergphillips.usc.edu/industryclass.htm. We thank the authors for making these data
publicly available.

Please cite this article as: C.S. Armstrong, S. Glaeser and J.D. Kepler, Strategic reactions in corporate tax planning, Journal
of Accounting and Economics, https://doi.org/10.1016/j.jacceco.2019.03.003
JID: JAE
ARTICLE IN PRESS [m3Gsc;August 20, 2019;11:56]

20 C.S. Armstrong, S. Glaeser and J.D. Kepler / Journal of Accounting and Economics xxx (xxxx) xxx

Table 6
Competitor classification robustness: Hoberg and Phillips industries.
This table reports OLS estimates of the following difference-in-differences specification for a sample of firms with operations in the United States
only:

Tax Planningi jt = α + β1 T reatedi j + β2 T reatedi j × Postt + Controls + εi jt

Where Tax Planning is measured as either Cash ETR or GAAP ETR as defined in Appendix A, and i, j, and t index firms, industries, and years,
respectively. We define industries for all covariates constructed in this table by FIC 50 code (Hoberg and Phillips, 2010, 2016)—Treated is an indicator
equal to one if firm i has competitors in industry j that have operations in Ireland, and zero otherwise. Post is an indicator equal to one after 1998,
which corresponds to the beginning of the progressive reduction of corporate tax rates in Ireland. Controls is a vector of time-varying firm and
industry-average characteristics as defined in Panel A of Table 1. For parsimony we do not tabulate coefficients for the control variables and fixed
effects. All variables are defined in Appendix A. t-statistics are reported below coefficient estimates and are calculated based on robust standard
errors clustered by industry. ∗ , ∗ ∗ , ∗ ∗ ∗ indicate statistical significance (two-sided) at the 0.1, 0.05, and 0.01 levels, respectively.

Dependent variable: Cash ETRijt GAAP ETRijt


(1) (2)

Treated Irish Exposureij 0.042 0.058∗ ∗


(1.61) (2.55)
Post Rate Cutt × Treated Irish Exposureij –0.044∗ –0.064∗ ∗ ∗
(–1.94) (–2.96)
Firm Controlsijt–1 Yes Yes
Industry Average Controls–ijt–1 Yes Yes
Year Fixed Effects Yes Yes
Observations 21,490 21,490
R2 0.100 0.127

Table 7
Long-run tax planning.
This table reports OLS estimates of the following difference-in-differences specification for a sample of firms with operations in the United States only:

Long Run Tax Planningi jt = α + β1 T reatedi j + β2 T reatedi j × Postt + Controls + εi jt

Where Long Run Tax Planning is measured as one of the long–run tax planning measures as defined in Appendix A, and i, j, and t index firms, industries,
and years, respectively. Treated is an indicator equal to one if firm i has competitors in industry j that have operations in Ireland, and zero otherwise.
Post is an indicator equal to one after 1998, which corresponds to the beginning of the progressive reduction of corporate tax rates in Ireland. Controls
is a vector of time-varying firm and industry-average characteristics as defined in Panel A of Table 1. All variables are defined in Appendix A. t-
statistics are reported below coefficient estimates and are calculated based on robust standard errors clustered by industry. ∗ , ∗ ∗ , ∗ ∗ ∗ indicate statistical
significance (two-sided) at the 0.1, 0.05, and 0.01 levels, respectively.

Dependent variable: 3-Year Cash ETRijt 5-Year Cash ETRijt 3-Year GAAP ETRijt 5-Year GAAP ETRijt
(1) (2) (3) (4)

Treated Irish Exposureij 0.007 0.015 –0.010 –0.007


(0.41) (1.04) (–0.90) (–0.65)
Post Rate Cutt × Treated Irish Exposureij –0.056∗ ∗ –0.070∗ ∗ ∗ –0.048∗ ∗ –0.057∗ ∗ ∗
(–2.43) (–3.52) (–2.42) (–3.36)
Firm Controlsijt–1 Yes Yes Yes Yes
Competitor Average Controls–ijt–1 Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes
Observations 18,987 14,916 19,250 15,502
R2 0.101 0.087 0.101 0.097

mismatches between cash taxes paid and earnings from the current period.31 We therefore assess the sensitivity of our
inferences to using long-run ETRs as alternative measures of corporate tax planning.
We calculate 3-year and 5-year ETRs as the sum of GAAP income tax expense and cash taxes paid, scaled by the sum
of pre-tax income during the same period (i.e., 3-Year Cash ETR, 5-Year Cash ETR, 3-Year GAAP ETR, and 5-Year GAAP ETR,
respectively).32 Table 7 presents results using the alternative long-run ETRs as the dependent variable in our Irish setting.
We find evidence of strategic complementarities in long-run corporate tax planning (t-stats ranging from –2.42 to –3.52 on
Post Rate Cut × Treated Irish Exposure). These results suggest that the strategic complementarities in corporate tax planning
that we document in our initial tests are also present in long-run measures of tax planning. Moreover, they also suggest
that our previous findings for one-year ETRs are not an artefact of mismatches between cash taxes paid and pre-tax income.

31
These mismatches can arise from timing differences in tax payments, cash settlements with tax authorities for violations recognized in prior periods,
and any potential U.S. taxes on foreign earnings that are unpaid until repatriation, but are included in current earnings (Hanlon and Heitzman, 2010).
32
Note that data requirements on firms’ future tax expense, tax payments, and pretax income restricts our sample, resulting in fewer observations in
Table 7 relative to each tests’ counterpart in earlier tables.

Please cite this article as: C.S. Armstrong, S. Glaeser and J.D. Kepler, Strategic reactions in corporate tax planning, Journal
of Accounting and Economics, https://doi.org/10.1016/j.jacceco.2019.03.003
JID: JAE
ARTICLE IN PRESS [m3Gsc;August 20, 2019;11:56]

C.S. Armstrong, S. Glaeser and J.D. Kepler / Journal of Accounting and Economics xxx (xxxx) xxx 21

Table 8
Delaware-haven differences-in-differences analysis.
This table reports OLS estimates of the following differences–in–differences specification:

Stat e ET Ri jt = α + β1 T reat edi j + β2 Postt + β3 T reat edi j × Postt + Controls + εi jt

Where Tax Planning is measured as State ETR as defined in Appendix A, and i, j, and t index firms, industries, and years, respectively. Treated
is an indicator equal to one if a firm’s subsidiary i has competitors—defined as firms with the same three-digit SIC code—in industry j that
have operations in Delaware, and zero otherwise. Post is an indicator equal to one after years during which separate reporting or no economic
nexus state laws were enacted, and zero otherwise. All Treated, Post, and Treated × Post variables are aggregated up to the firm-year level using
subsidiary-weighted averages. Controls is a vector of time-varying firm, industry, and state-average characteristics as defined in Panel B of Table 1.
For parsimony we do not tabulate coefficients for the control variables and fixed effects. All variables are defined in Appendix A. t-statistics are
reported below coefficient estimates and are calculated based on robust standard errors clustered by state of incorporation. ∗ , ∗ ∗ , ∗ ∗ ∗ indicate
statistical significance (two-sided) at the 0.1, 0.05, and 0.01 levels, respectively.

Dependent variable: State ETRijt State ETRijt State ETRijt State ETRijt State ETRijt State ETRijt
(1) (2) (3) (4) (5) (6)

Avg(Competitor DE PICijt × Post Lawist ) 0.004∗ ∗ 0.005∗ ∗ 0.004∗ 0.004∗ ∗ 0.003 0.001
(2.20) (2.30) (1.88) (2.09) (1.45) (0.26)
Avg(Competitor DE PICijt ) –0.013 –0.016 –0.012 –0.012 –0.005 0.026∗ ∗
(–1.65) (–1.62) (–1.66) (–1.60) (–0.45) (2.07)
Avg(Competitor Post Lawist ) –0.008 –0.006 –0.010 –0.011∗ ∗ 0.004 –0.004
(–0.88) (–0.71) (–1.33) (–2.02) (0.36) (–0.75)
Avg(Own DE PICijt ) –0.018∗ ∗ ∗ –0.018∗ ∗ ∗ –0.018∗ ∗ ∗ –0.019∗ ∗ ∗ –0.021∗ ∗ ∗ –0.014∗ ∗ ∗
(–3.50) (–3.69) (–3.63) (–3.67) (–4.39) (–3.58)
Avg(Own Post Lawist ) –0.002 –0.001 –0.005∗ –0.001 –0.005∗ 0.001
(–1.01) (–0.56) (–1.73) (–0.61) (–1.76) (0.25)
Avg(Own DE PICijt × Post Lawist ) 0.015∗ 0.016∗ 0.014 0.014 0.015∗ ∗ 0.003
(1.93) (1.88) (1.62) (1.54) (2.08) (0.35)

Firm Controlsijt–1 Yes Yes Yes Yes Yes Yes


Competitor Average Controls-ijt-1 Yes Yes Yes Yes Yes Yes
State Controlsijt-1 Yes Yes Yes Yes Yes Yes
Competitor State Average Controls-ijt-1 Yes Yes Yes Yes Yes Yes
Fixed Effects Year Region-Year State State-Year Industry-Year Firm, State-Year, &
Industry-Year
Observations 8618 8044 8618 8618 8618 8618
R-squared 0.038 0.052 0.096 0.172 0.329 0.414

5.2. Delaware-haven differences-in-differences analysis

5.2.1. Delaware-haven strategic reactions


Table 8 presents results from estimating Eq. (3), which models tax planning in a partially-overlapping groups differences-
in-differences analysis using states’ staggered adoption of policies designed to curb tax planning via inter-state income shift-
ing as shocks to competitors’ tax planning. The coefficient on Avg(Competitor DE PIC × Competitor Post Law) measures the
indirect effect of these state policies on the tax planning of firms that are not directly exposed to these policies, but that are
indirectly exposed by way of shared industry membership with directly exposed firms. The coefficient estimate in column
(1) implies that a one standard deviation increase in competitors’ exposure to state policies that neutralize the Delaware PIC
strategy results in a 0.14 percentage point increase firms’ State ETR (t-stat of 2.20). This increase is economically significant
as it represents 2.8% of the sample standard deviation of State ETR.
We also re-estimate Eq. (3) using alternative fixed effects structures to control for other potential forms of unobserved
heterogeneity. In particular, in columns (2) through (6) of Table 8 we sequentially include (i) region-year, (ii) state, (iii)
state-year, (iii) industry-year, and (iv) firm, state-year, and industry-year fixed effects and find that our inferences remain
qualitatively similar, albeit not statistically significant in columns (5) and (6).33 For parsimony we report estimates of all
subsequent variants of Eq. (3) with year fixed effects, but note that we obtain qualitatively similar inferences when using
these alternative fixed effect structures.
Again, the results suggest that the magnitude of strategic reactions is large. The results in Table 8, column (1) imply that
if all of a firm’s competitors went from unexposed to exposed to state laws designed to curb the Delaware PIC strategy,
the result would be a 0.4 percentage point increase in the firm’s State ETR. The coefficient on Avg(Own DE PIC × Post Law)
suggests that if a firm went from unexposed to exposed to state laws design to curb the Delaware PIC strategy, the result

33
In untabulated analyses we find respective t-stats of 3.00 and 6.26 for the coefficients on Avg(Competitor DE PIC × Competitor Post Law) in the specifica-
tions corresponding to those reported in columns (5) and (6) if we instead scale total state tax expense by pre-tax total income (the t-stats corresponding
to the specification reported in columns (1)–(4) range from 2.90 to 4.13). We scale by pre-tax domestic income to facilitate comparison of our estimates
with those reported by Dyreng et al. (2013).

Please cite this article as: C.S. Armstrong, S. Glaeser and J.D. Kepler, Strategic reactions in corporate tax planning, Journal
of Accounting and Economics, https://doi.org/10.1016/j.jacceco.2019.03.003
JID: JAE
ARTICLE IN PRESS [m3Gsc;August 20, 2019;11:56]

22 C.S. Armstrong, S. Glaeser and J.D. Kepler / Journal of Accounting and Economics xxx (xxxx) xxx

Table 9
Delaware-haven economic mechanisms.
This table reports OLS estimates of the following differences–in–differences specification:

Stat e ET Ri jt = α + β1 T reat edi j + β2 Postt + β3 T reat edi j × Postt + Controls + εi jt

Where Tax Planning is measured as either State ETR as defined in Appendix A, and i, j, and t index firms, industries, and years, respectively. Treated is an
indicator equal to one if a firm’s subsidiary i has competitors—defined as firms with the same three-digit SIC code—in industry j that have operations in
Delaware, and zero otherwise. Post is an indicator equal to one after years during which separate reporting or no nexus state laws were enacted, and zero
otherwise. All Treated, Post, and Treated × Post variables are aggregated up to the firm-year level using subsidiary-weighted averages. Controls is a vector
of time-varying firm, industry, and state-average characteristics as defined in Panel B of Table 1. For parsimony we do not tabulate coefficients for the
control variables and fixed effects. Controls is a vector of time-varying firm, industry, and state-average characteristics as defined in Panel B of Table 1.
For parsimony we do not tabulate coefficients for the control variables and fixed effects. All variables are defined in Appendix A. t-statistics are reported
below coefficient estimates and are calculated based on robust standard errors clustered by state of incorporation. ∗ , ∗ ∗ , ∗ ∗ ∗ indicate statistical significance
(two-sided) at the 0.1, 0.05, and 0.01 levels, respectively.

Dependent variable: State ETRijt

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

Avg(Competitor DE PICijt × Post Lawist ) –0.008 0.006∗ ∗ 0.014∗ ∗ 0.001


(–1.08) (2.45) (2.31) (1.11)
Delaware-haven Own Treatment Variables Yes Yes Yes Yes
Firm Controlsijt–1 Yes Yes Yes Yes
Competitor Average Controls-ijt-1 Yes Yes Yes Yes
State Controlsijt-1 Yes Yes Yes Yes
Competitor State Average Controls-ijt-1 Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes

Sample restriction Large Firm = Above Industry ETR =

1 0 1 0

Observations 2434 6184 3683 4935


R2 0.027 0.028 0.088 0.196

would be a 1.5 percentage point increase in the firm’s State ETR.34 Together, these estimates imply that a one percentage
point decrease in competitors’ tax planning results in an approximately 0.27 percentage point decrease in firms’ tax planning
(0.4/1.5).
The magnitude of strategic reactions documented in our Irish-haven tests (an elasticity of roughly 0.63) are larger than
those documented in our Delaware-haven tests (an elasticity of roughly 0.27). In other words, firms appear to respond more
to increases in their competitors’ tax planning. One potential explanation for the differential magnitude is that firms are
more willing to reduce their tax burden and are therefore more responsive to increases in their industry-competitors’ tax
planning. However, we interpret these differential magnitudes with caution since our estimates of strategic reactions are not
structural in the sense that they depend on economic, policy, and other conditions.
For example, federal tax regulators may be less constrained than state tax regulators, causing enforcement herding in-
centives to be weaker in the federal setting. Similarly, measurement error may differ across the two settings (e.g., if our
state subsidiary data is noisy, this may attenuate the coefficient estimates in the Delaware-haven setting). Therefore, we
cautiously interpret the above evidence of differential magnitudes as consistent with firms exhibiting stronger strategic re-
actions to increases in their competitors’ tax planning. However, we believe there is opportunity for future research to
compare strategic reactions in a setting where economic and policy conditions and measurement error are similar for both
increases and decreases in firms’ competitors’ tax planning.

5.2.2. Delaware-haven economic mechanism analysis


Table 9 presents the results of modifying Eq. (3) to accommodate the mechanism tests in the Delaware PIC setting.
Columns (1) and (2) present results from estimating Eq. (3) using the subsample of large firms (Large Firm = 1) and small
firms (Large Firm = 0), respectively. We find some evidence of leader-follower dynamics inducing strategic complementarities
in corporate tax planning (untabulated F-statistic of 2.56, p-value of 0.12).
Columns (3) and (4) of Table 9 present the results from estimating Eq. (3) after splitting the Delaware-haven sample
based on whether the firm exhibits relatively less (Above Industry ETR = 1 in column (3)) or relatively more (Above Industry
ETR = 0 in column (4)) tax planning than its industry competitors. We find evidence that firms that are less aggressive than
their industry competitors—and therefore should have greater scope to learn and implement new tax planning strategies—
are sensitive to increases in their competitors’ tax planning (coefficient estimate of 0.014; t-stat of 2.31 in column (3)).
We find no evidence that firms that are more aggressive than their industry competitors—and therefore should be more
concerned about tax enforcement—are sensitive to changes in their industry competitors’ tax planning (coefficient estimate

34
The evidence in Dyreng et al. (2013) indicates that the direct effect of the Delaware PIC strategy was a 0.9 percentage point decline in competitors’
State ETRs. The larger effect we document is consistent with changes in tax avoidance policies causing spillovers on the tax behavior of firms that were
not directly targeted by the policy, biasing estimates of the effect downward if unaccounted for.

Please cite this article as: C.S. Armstrong, S. Glaeser and J.D. Kepler, Strategic reactions in corporate tax planning, Journal
of Accounting and Economics, https://doi.org/10.1016/j.jacceco.2019.03.003
JID: JAE
ARTICLE IN PRESS [m3Gsc;August 20, 2019;11:56]

C.S. Armstrong, S. Glaeser and J.D. Kepler / Journal of Accounting and Economics xxx (xxxx) xxx 23

Table 10
Falsification tests: placebo analysis.
This table reports OLS estimates of the following differences–in–differences specification:

Pretax Per f or mancei jt = α + β1 T reatedi j + β2 T reatedi j × Postt + Controls + εi jt

Where Pretax Performance is measured as Pretax Cash Flow, as defined in Appendix A, and i, j, and t index firms, industries,
and years, respectively, and replicates our Delaware-haven analysis using Pretax Performance as an outcome variable instead
of Tax Planning. Controls is a vector of time-varying firm, industry, and state-average characteristics as defined in Panel B of
Table 1. For parsimony we do not tabulate coefficients for the control variables and fixed effects. All variables are defined in
Appendix A. t-statistics are reported below coefficient estimates and are calculated based on robust standard errors clustered
by state of incorporation. ∗ , ∗ ∗ , ∗ ∗ ∗ indicate statistical significance (two-sided) at the 0.1, 0.05, and 0.01 levels, respectively.

Dependent variable: Pretax Cash Flowijt


(1)

Avg(Competitor DE PICijt × Post Lawist ) 0.006


(1.35)
Delaware-haven Own Treatment Variables Yes
Firm Controlsijt–1 Yes
Competitor Average Controls- ijt-1 Yes
State Controlsijt-1 Yes
Competitor State Average Controls- ijt-1 Yes
Year Fixed Effects Yes
Observations 8616
R2 0.584

of 0.001; t-stat of 1.11 in column (4)). A potential explanation for finding stronger evidence of enforcement herding in the
Irish-haven setting than in the Delaware-haven setting is that the IRS is the regulator in the former, while various state
enforcement agencies are the regulator in the latter. Because the IRS has broader scope and more resources, it is a more
salient enforcement authority than are its state-level counterparts.
Collectively, the results in Tables 4 and 9 provide no consistent evidence that smaller firms are more sensitive to changes
in their industry competitors’ tax planning than are their larger counterparts. This finding is inconsistent with leader-
follower dynamics partially causing strategic complementarities in corporate tax planning. We find some evidence that firms
that are engaged in less tax planning than their industry competitors are responsive to changes in their competitors’ tax
planning. This finding suggests that learning may be partially responsible for the strategic complementarities in corporate
tax planning. Finally, we find more consistent evidence that firms that engage in more tax planning than their competi-
tors are responsive to changes in their competitors’ tax planning, consistent with enforcement herding causing strategic
complementarities in corporate tax planning.

5.2.3. Delaware-haven placebo analysis


Finally, Table 10 presents results from re-estimating Eq. (3) using Pretax Cash Flow as the dependent variable as a fal-
sification test. The results again provide no evidence of strategic reactions in firms’ pre-tax performance (t-stat of 1.35 for
the coefficient on Avg(Competitor DE PIC × Competitor Post Law)). Collectively, the results in Table 10 provide no evidence
of strategic complementarities in firms’ pre-tax performance, suggesting that our finding of strategic complementarities in
corporate tax planning is not an artefact of a spurious endogenous relation.35 , 36

6. Conclusion

Using a partially-overlapping groups difference-in-differences identification strategy in two different research settings, we
show that competitors’ tax planning is an important determinant of a firm’s own tax planning. We explore several possible
reasons why firms exhibit strategic reactions to their industry competitors’ tax planning, including leader-follower dynamics
among firms, rational herding, and learning. We find evidence that firms strategically respond to their competitors’ tax
planning by herding in order to avoid “standing out” among their industry competitors. We also find some evidence that
firms respond to their industry competitors’ tax planning because they learn from their industry competitors’ tax planning.
We contribute to the literature on the determinants of corporate tax planning by documenting evidence of strategic com-
plementarities in the tax planning of industry competitors. This finding has potential policy implications, as it suggests that

35
In untabulated analyses, we find no evidence of strategic reactions in the Delaware-haven setting when using our alternative industry classification. One
possible explanation for why our results are somewhat weaker when using this alternative classification of industry is that it does not closely correspond
to how tax authorities (e.g., the IRS) classify firms for purposes of identifying tax avoidance. In contrast, our primary industry classification using SIC codes
corresponds to how the IRS classifies industry peers. Accordingly, these results also provide evidence that rational herding by firms to avoid scrutiny from
tax authorities is a mechanism responsible for inducing strategic complementarities in firms’ tax planning. To the extent that the Hoberg and Phillips
industry definitions more closely correspond to how stakeholders implicitly or explicitly classify firms as peers, these results suggest that enforcement
herding is less likely to be driven by reputational concerns with stakeholders.
36
In untabulated analyses, we continue to find evidence of strategic complementarities in the Delaware-haven setting when using 3-year and 5-year State
ETRs as the dependent variable (t-stats of 3.13 and 1.94, respectively, for the coefficient on Competitor DE PIC × Competitor Post Law).

Please cite this article as: C.S. Armstrong, S. Glaeser and J.D. Kepler, Strategic reactions in corporate tax planning, Journal
of Accounting and Economics, https://doi.org/10.1016/j.jacceco.2019.03.003
JID: JAE
ARTICLE IN PRESS [m3Gsc;August 20, 2019;11:56]

24 C.S. Armstrong, S. Glaeser and J.D. Kepler / Journal of Accounting and Economics xxx (xxxx) xxx

policies that seek to limit or curtail the tax planning of a targeted subset of firms could inadvertently affect the tax planning
of their otherwise unaffected (i.e., untargeted) competitors through their strategic reactions. This finding also has implica-
tions for researchers, as it suggests that research designs and theoretical models that do not account for strategic reactions
in corporate tax planning may be incomplete and may fail to capture the full effect of firms’ tax planning or regulators’
enforcement efforts. We also contribute to the emerging literature on “peer” and “network” effects by demonstrating that
firms strategically respond to their peers’ tax planning.

Acknowledgments

We gratefully acknowledge financial support from the Wharton School of the University of Pennsylvania and the Kenan-
Flagler Business School at the University of North Carolina. Armstrong also thanks EY for its research support. We also
acknowledge helpful comments and suggestions from Anne Beatty, Wayne Guay, Shane Heitzman, Jeff Hoopes, Sudarshan
Jayaraman (referee), Urooj Khan, S.P. Kothari (editor), Aaron Nelson, Stephanie Sikes, Jake Thornock, David Tsui, and the
Wharton Accounting brownbag participants. We thank Scott Dyreng for providing data on firms’ Irish operations, the loca-
tions of corporate subsidiaries, and the dates of changes in state corporate tax laws. We thank Gerard Hoberg and Gordon
Phillips for providing data on text-based industry classifications.

Appendix A. Variable definitions

This table presents definitions of the variables in our empirical tests. We also compute industry-level averages for each of
these variables as the market value-weighted average of all firms within a three-digit SIC code industry-year combination,
excluding the ith observation.

Tax planning measures

Cash ETR Total taxes paid during the current year scaled by pre-tax income
GAAP ETR Total tax expense during the current year scaled by pre-tax income
State ETR Total state tax expense during the current year scaled by pre-tax domestic income
3-Year Cash ETR Sum of total taxes paid during the current year and each of the following two years, scaled by the sum of total pre-tax
income during the current year and each of the following two years
3-Year GAAP ETR Sum of tax expense during the current year and each of the following two years, scaled by the sum of total pre-tax
income during the current year and each of the following two years
5-Year Cash ETR Sum of total taxes paid during the current year and each of the following four years, scaled by the sum of total pre-tax
income during the current year and each of the following four years
5-Year GAAP ETR Sum of tax expense during the current year and each of the following four years, scaled by the sum of total pre-tax
income during the current year and each of the following four years

Control Variables

Book Leverage Book value of total debt scaled by book value of total assets, measured at the end of the fiscal year
Market Leverage Book value of total debt scaled by market value of total assets, measured at the end of the fiscal year. Market value of
total assets is measured as the Compustat stock price as of fiscal year end multiplied by common shares outstanding, plus
debt in current liabilities, plus total long-term debt, plus preferred stock liquidating value, minus deferred taxes and
investment credits
ROA Income before extraordinary items scaled by beginning of the year total assets
Foreign Assets Indicator that equals one if a firm has foreign assets and zero otherwise
New Investment Sum of R&D, capital expenditures, and acquisition, minus sale of property and depreciation, all scaled by average total
assets. Average total assets is defined as beginning of year total assets plus end of year total assets, divided by two
Log(Market Value of Natural logarithm of Compustat stock price multiplied by common shares outstanding as of the end of the fiscal year
Equity)
Book-to-Market Book value of assets scaled by market value of assets, measured as the Compustat stock price as of fiscal year end
multiplied by common shares outstanding, plus debt in current liabilities, plus total long-term debt, plus preferred stock
liquidating value, minus deferred taxes and investment credits
Change in Goodwill Goodwill at the end of the fiscal year minus Goodwill at the beginning of the fiscal year, scaled by total assets as of the
end of the fiscal year
Net Operating Loss Indicator that equals one if the firm has had a tax loss carryforward during any of the previous three years and zero
otherwise

State Controls

GSP Growth The percentage change from the prior year’s state GDP, weighted by non-Delaware subsidiaries
Corporate State Tax Rate The top statutory tax rate on corporate income, weighted by non-Delaware subsidiaries
Personal State Tax Rate The top statutory rate on personal income, weighted by non-Delaware subsidiaries

Economic Mechanism Partitioning Variables

Large Firm Indicator that equals one if a firm’s size is greater than its market value-weighted industry (based on three-digit SIC code)
average market value of equity and zero otherwise
(continued on next page)

Please cite this article as: C.S. Armstrong, S. Glaeser and J.D. Kepler, Strategic reactions in corporate tax planning, Journal
of Accounting and Economics, https://doi.org/10.1016/j.jacceco.2019.03.003
JID: JAE
ARTICLE IN PRESS [m3Gsc;August 20, 2019;11:56]

C.S. Armstrong, S. Glaeser and J.D. Kepler / Journal of Accounting and Economics xxx (xxxx) xxx 25

Tax planning measures

Above Industry ETR Indicator equal to one if a firm’s own effective tax rate is higher than its competitor industry average (based on
three-digit SIC code) during the year and zero otherwise

Irish-haven Partially-Overlapping Industry Membership Variables

Post Rate Cut Indicator equal to one after 1998 and zero otherwise
Treated Irish Exposure Indicator equal to one if a firm has industry-competitors with operations in Ireland and zero otherwise

Delaware-haven Competitor Treatment Variables

Avg(Competitor Post Indicator equal to one if the state requires combined reporting, follows the economic nexus doctrine, or both, and zero
Law) otherwise, weighted by non-Delaware subsidiaries
Avg(Competitor DE PIC) Indicator equal to one if the subsidiary has an in-state industry competitor in the top tercile of Delaware subsidiaries and
above-median market-to-book ratios and zero otherwise, weighted by non-Delaware subsidiaries
Avg(Competitor DE Non-Delaware subsidiary weighted average of Competitor Post Law multiplied by Competitor DE PIC
PIC × Post Law)

Delaware-haven Own Treatment Variables

Avg(Own DE PIC) Indicator equal to one if the firm’s own subsidiary is in the top tercile of Delaware subsidiaries and above-median
market-to-book ratios of our sample and zero otherwise, weighted by non-Delaware subsidiaries
Avg(Own Post Law) Indicator equal to one if the state requires combined reporting, follows the economic nexus doctrine, or both, and zero
otherwise, weighted by non-Delaware subsidiaries
Avg(Own DE PIC × Post Non-Delaware subsidiary weighted average of Own Post Law multiplied by Own DE PIC
Law)

Placebo Test Outcome

Pretax Cash Flow Pre-tax cash flows from operations during the fiscal year, scaled by beginning of the year total assets

References

Angrist, J., 2014. The perils of peer effects. Labour Econ. 30, 98–108.
Armstrong, C.S., Kepler, J.D., 2018. Theory, research design assumptions, and causal inferences. J. Acc. Econ. 66 (2-3), 366–373.
Armstrong, C., Blouin, J., Larcker, D., 2012. The incentives for tax planning. J. Account. Econ. 53.1 (1-2), 391–411.
Armstrong, C.S., Glaeser, S., Huang, S., 2018a. Controllability of risk and the design of incentive-compensation contracts. Working paper.
Armstrong, C.S., Glaeser, S., Huang, S., Taylor, D.J., 2018b. The economics of managerial taxes and corporate risk-taking. Account. Rev.
Ayers, B., Seidman, J., Towery, E., 2015. Taxpayer behavior under audit certainty. Working paper.
Beck, T., Lin, C., Ma, Y., 2014. Why do firms evade taxes? The role of information sharing and financial sector outreach. J. Finance 69.2 (2), 763–817.
Bird, A., Edwards, A., Ruchti, T., 2018. Taxes and peer effects. Account. Rev. 95 (5), 97–117.
Blank, J., 2014. Reconsidering corporate tax privacy. NYUJL Bus. 11, 31.
Blume, L., Durlauf, S., 2005. Identifying social interactions: a review. Madison: Social Systems Research Institute. University of Wisconsin.
Boone, J., 20 0 0. Competitive pressure: the effects on investments in product and process innovation. RAND J. Econ. 31.3 (3), 549–569.
Bozanic, Z., Hoopes, J., Thornock, J., Williams, B., 2016. IRS attention. J. Account. Res. 55 (1), 79–114.
Blume, L., Brock, W., Durlauf, S., Jayaraman, R., 2015. Linear social interactions models. J. Political Econ. 123.2, 444–496.
Blume, L., Brock, W., Durlauf, S., Ioannides, Y, 2011. Identification of social interactions. In: Handbook of Social Economics, 1, pp. 853–964. North-Holland.
Boning, W.C., Guyton, J., Hodge, R.H., Slemrod, J., Troiano, U., 2018. Heard it Through the Grapevine: Direct and Network Effects of a Tax Enforcement Field
Experiment. National Bureau of Economic Research No. w24305.
Bramoullé, Y., Djebbari, H., Fortin, B., 2009. Identification of peer effects through social networks. J. Econ. 150.1 (1), 41–55.
Brock, W., Durlauf, S., 2001a. Interaction-based models. In: Handbook of Econometrics, 5, pp. 3297–3380. North-Holland.
Brock, W., Durlauf, S., 2001b. Growth empirics and reality. World Bank Econ. Rev. 15.2, 229–272.
Brown, J., 2011. The spread of aggressive corporate tax reporting: a detailed examination of the corporate-owned life insurance shelter. Account. Rev. 86.1
(1), 23–57.
Brown, J., Drake, K., 2013. Network ties among low-tax firms. Account. Rev. 89.2 (2), 483–510.
Bulow, J., Geanakoplos, J., Klemperer, P., 1985. Multimarket oligopoly: strategic substitutes and complements. J. Political Econ. 93.3 (3), 488–511.
Cohen-Cole, E., Fletcher, J., 2008. Is obesity contagious? Social networks vs. environmental factors in the obesity epidemic. J. Health Econ. 27.5 (5),
1382–1387.
Cen, L., Maydew, E.L., Zhang, L., Zuo, L., 2017. Customer-supplier relationships and corporate tax avoidance. J. Finan. Econ. 123 (2), 377–394.
Cen, L., Maydew, E.L., Zhang, L., Zuo, L., 2018. Tax Planning Diffusion Along the Supply Chain.
Chen, S., Chen, X., Cheng, Q., Shevlin, T, 2010. Are family firms more tax aggressive than non-family firms. J. Financ. Econ. 95.1 (1), 41–61.
De Giorgi, G., Pellizzari, M., Redaelli, S., 2010. Identification of social interactions through partially overlapping peer groups. Am. Econ. J.: Appl. Econ. 2.2
(2), 241–275.
Desai, M., Dyck, A., Zingales, L., 2007. Theft and taxes. J. Financ. Econ. 84.3 (3), 591–623.
Diamond, D., Dybvig, P., 1983. Bank runs, deposit insurance, and liquidity. J. Political Econ. 91, 3 , 401–419.
Donohoe, M., Lisowsky, P., Mayberry, M., 2015. Taxes, competition, and organizational form. Working paper.
Durlauf, S., Ioannides, Y., 2010. Social interactions. Annu. Rev. Econ. 2.1 (1), 451–478.
Dyreng, S., Lindsey, B., 2009. Using financial accounting data to examine the effect of foreign operations located in tax havens and other countries on US
multinational firms’ tax rates. J. Account. Res. 47.5, 1283–1316.
Dyreng, S., Lindsey, B., Thornock, J., 2013. Exploring the role Delaware plays as a domestic tax haven. J. Financ. Econ. 108.3 (3), 751–772.
Dyreng, S., Hanlon, M., Maydew, E., 2008. Long-run corporate tax planning. Account. Rev. 83.1, 61–82.
Dyreng, S., Hanlon, M., Maydew, E., Thornock, J., 2017. Changes in corporate effective tax rates over the past 25 years. J. Financ. Econ. 124 (3), 441–463.
Dyreng, S., Hoopes, J., Wilde, J., 2016. Public pressure and corporate tax behavior. J. Account. Res. 54 (1), 147–186.
Fudenberg, D., Tirole, J., 1984. The fat-cat effect, the puppy-dog ploy, and the lean and hungry look. Am. Econ. Rev. 74.2, 361–366.
Gallemore, J., Maydew, E., Thornock, J., 2014. The reputational costs of tax planning. Contemp. Account. Res. 31.4, 1103–1133.
Glaeser, E., Sacerdote, B., Scheinkman, J., 1996. Crime and social interactions. Q. J. Econ. 111.2, 507–548.

Please cite this article as: C.S. Armstrong, S. Glaeser and J.D. Kepler, Strategic reactions in corporate tax planning, Journal
of Accounting and Economics, https://doi.org/10.1016/j.jacceco.2019.03.003
JID: JAE
ARTICLE IN PRESS [m3Gsc;August 20, 2019;11:56]

26 C.S. Armstrong, S. Glaeser and J.D. Kepler / Journal of Accounting and Economics xxx (xxxx) xxx

Glaeser, E., Sacerdote, B., Scheinkman, J., 2003. The social multiplier. J. Eur. Econ. Assoc. 1.2-3, 345–353.
Glaeser, S., Guay, W.R., 2017. Identification and generalizability in accounting research: a discussion of Christensen, Floyd, Liu, and Maffett (2017). J. Account.
Econ. 64 (2-3), 305–312.
Gormley, T., Matsa, D., 2014. Common errors: how to (and not to) control for unobserved heterogeneity. Rev. Financ. Stud. 27.2 (2), 617–661.
Graham, B, 2008. Identifying social interactions through conditional variance restrictions. Econometrica 76.3 (3), 643–660.
Graham, J., Hanlon, M., Shevlin, T., Shroff, N., 2014. Incentives for tax planning and avoidance: evidence from the field. Account. Rev. 89.3 (3), 991–1023.
Grinblatt, M, Keloharju, M., Ikäheimo, S., 2008. Social influence and consumption: evidence from the automobile purchases of neighbors. Rev. Econ. Stat.
90.4 (4), 735–753.
Hanlon, M., Heitzman, S., 2010. A review of tax research. J. Account. Econ. 50.2 (2-3), 127–178.
Hasan, I., Keung, C., Hoi, S., Wu, Q., Zhang, H., 2014. Beauty is in the eye of the beholder: the effect of corporate tax planning on the cost of bank loans. J.
Financ. Econ. 113.1, 109–130.
Heckman, J., LaLonde, R., Smith, J., 1999. The economics and econometrics of active labor market programs. In: Handbook of Labor Economics, 3,
pp. 1865–2097.
Heitzman, S., Ogneva, M., 2018. Corporate Tax Planning and Stock Returns. Working paper.
Hirshleifer, D., Teoh, S.H., 2003. Herd behaviour and cascading in capital markets: a review and synthesis. Eur. Financ. Manag. 9.1 (1), 25–66.
Hoberg, G., Phillips, G., 2010. Product market synergies and competition in mergers and acquisitions: a text-based analysis. Rev. Financ. Stud. 23.10 (10),
3773–3811.
Hoberg, G., Phillips, G., 2016. Text-based network industries and endogenous product differentiation. J. Political Econ. 124.5, 1423–1465.
Horowitz, J., Manski, C., 1995. Identification and robustness with contaminated and corrupted data. Econometrica: J. Econ. Soc. 63.2 (2), 281–302.
Ioannides, Y., 2006. Topologies of social interactions. Econ. Theory 28.3 (3), 559–584.
Kubick, T., Lynch, D., Mayberry, M., Omer, T., 2014. Product market power and tax planning: market leaders, mimicking strategies, and stock returns. Account.
Rev. 90.2, 675–702.
Leary, M., Roberts, M., 2014. Do peer firms affect corporate financial policy? J. Finance 69.1 (1), 139–178.
Lisowsky, P, Robinson, L., Schmidt, A., 2013. Do publicly disclosed tax reserves tell us about privately disclosed tax shelter activity? J. Account. Res. 51.3 (3),
583–629.
Manski, C.F., 1993. Identification of endogenous social effects: the reflection problem. Rev. Econ. Stud. 60 (3), 531–542.
Manski, C, 2011. Policy analysis with incredible certitude. Econ. J. 121, F261–F289.
Mills, L., Erickson, M., Maydew, E., 1998. Investments in tax planning. J. Am. Tax. Assoc. 20.1, 1.
Morris, S., Shin, H.S., 1998. Unique equilibrium in a model of self-fulfilling currency attacks. Am. Econ. Rev. 587–597.
Murnane, R., Willett, J., 2010. Methods matter: Improving Causal Inference in Educational and Social Science Research. Oxford University Press.
Nofsinger, J., Sias, R., 1999. Herding and feedback trading by institutional and individual investors. J. Finance 54.6 (6), 2263–2295.
Rajan, R., Zingales, L., 1995. What do we know about capital structure? Some evidence from international data. J. Finance 50.5, 1421–1460.
Rego, S., 2003. Tax-avoidance activities of US multinational corporations. Contemp. Account. Res. 20.4, 805–833.
Roberts, M., Whited, T., 2013. Endogeneity in empirical corporate finance. In: Handbook of the Economics of Finance, 2, pp. 493–572.
Rubin, D., 1986. Comment: which ifs have causal answers. J. Am. Stat. Assoc. 81 (396), 961–962.
Slemrod, J., 1993. Why People Pay taxes: Tax Compliance and Enforcement. University of Michigan Press.
Sobel, M., 1996. An introduction to causal inference. Sociol. Methods Res. 76 (3), 353–379.
Wilde, J.H., Wilson, R.J., 2018. Perspectives on corporate tax planning: observations from the past decade. J. Am. Tax. Assoc. 40 (2), 63–81.
Xiao, J., 2017. Diffusion of Corporate Governance from Influential Adopters. Working paper.
Zimmerman, J., 1983. Taxes and firm size. J. Account. Econ. 5, 119–149.

Please cite this article as: C.S. Armstrong, S. Glaeser and J.D. Kepler, Strategic reactions in corporate tax planning, Journal
of Accounting and Economics, https://doi.org/10.1016/j.jacceco.2019.03.003

You might also like