How Quickly Do Firms Adjust to Target Levels of Tax Avoidance?
Jaewoo Kim
Simon School of Business
University of Rochester
Sean McGuire
Mays Business School
Texas A&M University
Steven Savoy
Tippie College of Business
University of Iowa
Ryan Wilson
Lundquist College of Business
University of Oregon
September 2015
Abstract: Prior research that examines firms’ tax avoidance activities focuses on the
determinants of tax avoidance. Because a firm’s tax environment is constantly evolving, studying
how quickly firms adjust their tax avoidance activities is necessary to develop a better
understanding of the tradeoffs managers face to achieve a given level of tax avoidance. We test
for evidence of targeting behavior by managers. Our results suggest that firms do have target
levels of tax avoidance and that the typical firm converges towards its target at a rate of almost
69 percent over a three-year period. We also find that the speed of adjustment varies with a
firm’s growth potential, presence of foreign operations, and governance.
We appreciate helpful comments from John Campbell, Katherine Drake, Nancy Du, Nathan Goldman, Jeff
Gramlich, Dave Guenther, Russ Hamilton, Shane Heitzman, Paul Hribar, John Jiang, Ed Outslay, Santhosh
Ramalingegowda, Casey Schwab, James Stekelberg, Bridget Stomberg, Erin Towery, Brian Williams, Julie Xiao,
and workshop participants at the UBCOW conference, Michigan State University, the University of Arizona, the
University of Georgia, the University of Oregon, and Washington State University.
I.
Introduction
Dyreng, Hanlon, and Maydew (2008) document substantial variation in the level of
firms’ tax avoidance. In an effort to explain the variation in tax avoidance, recent research
examines the determinants of the level of firms’ tax avoidance. While determinant studies
provide valuable insights, the full implications of their findings are not clear because they do not
consider firms’ tax planning objective. The tradeoff literature (e.g., Scholes, Wilson, and
Wolfson 1990) suggests that firms trade off tax benefits against nontax costs, which implies that
firms target a level of tax avoidance based on the costs and benefits. Despite the clarity of the
tradeoff literature’s prediction, there is limited evidence on whether firms target a specific level
of tax avoidance. The purpose of this study is to examine whether firms adjust to a target level of
tax avoidance and, if so, how quickly firms converge toward their predicted target.
In prior tax avoidance studies, researchers often regress a measure of tax avoidance on
potential determinants in the cross-section of firms, but the implications of a significant
coefficient in such a regression are unclear. A significant coefficient implies that the determinant
explains variation in the target level of tax avoidance across firms but less about deviations from
that target. Furthermore, a determinant may have a different effect on the level of tax avoidance
across the tax avoidance distribution. For example, Armstrong, Blouin, Jagolinzer, and Larcker
(2015) find that the impact of corporate governance on tax avoidance is most pronounced at the
extremes of the tax avoidance distribution. 1 For these reasons, examining the determinants of the
level of a firm’s tax avoidance does not provide a complete understanding of managers’ tax
planning activities. Studying whether firms adjust to a certain target level of tax avoidance and
1
Armstrong et al. (2015) provide evidence of the direction of tax avoidance in the cross-section of firms. Our study
complements and extends Armstrong et al. (2015) by examining the speed with which firms adjust their tax
avoidance activities in a dynamic setting.
1
the speed with which a firm converges to its target offers new insight into firms’ tax planning
activities. Further, examining how the convergence speed varies with firm characteristics
provides evidence on the tradeoffs that firms face between tax savings and the costs of
convergence.
The tradeoff literature argues that firms weigh tax benefits against nontax costs in making
investment and financing decisions. The nontax costs of tax avoidance include financial
reporting considerations, agency costs, and implementation costs (Shackelford and Shevlin
2001). For example, Scholes, Wilson, and Wolfson (1990) examine tradeoffs in the context of
banks’ security transactions. They find evidence that banks consider nontax costs related to asset
turnover, the level of income reported to shareholders, and regulatory capital concerns when
making the decision to sell investments in order to minimize taxes. Other studies document
tradeoffs related to LIFO adoptions (Cushing and LeClere 1992; Dopuch and Pincus 1988),
deferred compensation (Matsunaga 1992), and income shifting (Guenther 1994). In summarizing
the tradeoff literature, Shackelford and Shevlin (2001) identify some common themes from the
papers. Among their takeaways from this area of research they note that “taxes are not a cost that
taxpayers inevitably avoid.”
An important implication of the tradeoff literature is that the costs and benefits of tax
avoidance are different for each firm depending on its characteristics and the parties involved in
its transactions. As a result, each firm has a unique target level of tax avoidance. In a world
where firms are able to change an infinitesimal level of tax avoidance without logistic frictions,
tradeoff theory predicts a firm will always maintain its target level of tax avoidance and will
never deviate from the target. However, a firm’s tax environment is dynamic suggesting that a
2
firm’s target level of tax avoidance is constantly changing. 2 Further, implementing and
unwinding tax avoidance activities is likely associated with nontax costs. In such an
environment, the tradeoff literature implies that firms will adjust their level of tax avoidance
because either the firm has temporarily deviated from its target level of tax avoidance or the firm
identified a new target level of tax avoidance.
Because firms’ tax environments are dynamic in nature, it is important that our empirical
model allows firms’ target level of tax avoidance to vary over time and also recognizes that firms
potentially cannot converge to a target in a single period. Accordingly, we follow the capital
structure literature and estimate a partial adjustment model that allows incomplete adjustment
toward a target level of tax avoidance (e.g., Flannery and Rangan 2006). 3 Specifically, we
estimate a firm’s target level of avoidance as a function of operating, financing, and investing
characteristics that prior research finds are associated with the level of a firm’s tax avoidance
(e.g., Chen, Chen, Cheng, and Shevlin 2010). We use a three year cash effective tax rate (ETR)
to proxy for firms’ level of tax avoidance because it is a visible tax avoidance metric that is not
influenced by tax accrual management (Dyreng, Hanlon, and Maydew 2008).
Our results suggest that the typical firm converges toward its target level of tax avoidance
at a rate of almost 69 percent per year. This rapid adjustment speed is important for two reasons.
First, it suggests firms are able to adjust rapidly to the target level of tax avoidance despite the
costs and logistical difficulties of implementing changes in tax planning strategies. Second, prior
2
Changes to the target level of tax avoidance could stem from changes in tax law. Alternatively, changes in firm
characteristics such as profitability, capital structure, firm size, product mix, or geographic markets will cause a
firm’s target level of tax avoidance to change.
3
Although partial adjustment models are used in the finance literature to examine the speed with which firms adjust
to their optimal capital structure (e.g., Flannery and Rangan 2006) or optimal level of cash holdings (Dittmar and
Duchin 2010; Venkiteshwaran 2011), we acknowledge that it is not possible to measure a firm’s optimal level of tax
avoidance. Rather, it is our intent to measure a firm’s target level of tax avoidance before considering adjustment
costs.
3
capital structure research argues that a slow annual adjustment speed does not support the
presence of targeting behavior. 4 Thus, consistent with tradeoff theory, our estimated adjustment
speed suggests that firms actively adjust toward a specific tax avoidance target.
To provide additional insight into firms’ targeting behavior, we examine whether there is
cross-sectional variation in how quickly firms adjust to their target level of tax avoidance. First,
we examine if the speed of adjustment varies depending on whether a firm’s actual level of tax
avoidance is above or below the firm’s tax avoidance target. We expect to observe asymmetric
adjustment speeds to the extent that the cost of implementing a new tax strategy for firms below
their target level of tax avoidance differs from the cost of unwinding a current tax position for
firms above their target level of tax avoidance. Consistent with our expectations, we find that
firms that are above their target level of tax avoidance exhibit an adjustment speed of
approximately 89 percent while firms that are below their target have an adjustment speed of
approximately 47 percent. These results are consistent with the tradeoff theory’s prediction that
managers actively attempt to achieve a target level of tax avoidance regardless of whether their
initial level of tax avoidance is above or below their target. Further, our results suggest that firms
engaged in too little tax avoidance increase their tax avoidance more quickly than the rate at
which firms engaged in too much tax avoidance scale back their tax avoidance. One possible
explanation for this result is that the costs of failing to operate at the target level of tax avoidance
are higher when the firm is engaged in too little tax avoidance which motivates these firms to
adjust more quickly.
4
The capital structure literature proposes several theories to explain the variation in firms’ leverage ratios (e.g.,
pecking order, market timing, and tradeoff theories). Prior to Flannery and Rangan (2006), estimates of firm’s
annual adjustment speed ranged from eight percent to 15 percent. Fama and French (2002) note that such slow
adjustment speeds do not provide support for firms pursuing a target level of capital structure. Although, the speed
of adjustment towards a tax avoidance target is much faster than the speed of adjustment toward a capital structure
target, we argue that Fama and French’s (2002) logic applies in the tax avoidance context as well.
4
Second, we examine whether the adjustment speed varies based on a firm’s growth
potential. It is costly for firms to acquire highly implicitly or explicitly taxed assets if they are
not in the correct tax clientele. Extending this line of thinking, we posit that firms more prone to
shifting between tax clienteles across time will find it more costly to commit to long-term tax
planning strategies for fear of experiencing a shift in profitability or operations and ending up in
the wrong clientele. Thus, we expect that firms with greater growth potential will exhibit slower
adjustment speeds relative to firms with less growth potential. We find firms with above the
sample median levels of growth potential exhibit significantly slower adjustment speeds to the
target level of tax avoidance relative to firms below the sample median. Consistent with the
tradeoff view, this result suggests that firms whose target level of tax avoidance is more likely to
change exhibit slower adjustment speeds because the costs of adjustment likely exceed the
benefits.
Third, we examine whether the adjustment speed varies based on the presence of foreign
operations. We expect that firms with foreign operations (MNCs) will exhibit faster adjustment
speeds relative to domestic-only firms because the presence of foreign operations provides
MNCs with more tax planning opportunities and, thus, lower costs of adjustment. Consistent
with our expectations, we find that MNCs exhibit significantly faster adjustment speeds relative
to domestic corporations. Finally, we examine whether the speed of adjustment varies with
corporate governance. Recent research finds that more independent boards have greater ability
and incentives to monitor managers’ tax avoidance decisions (Armstrong et al. 2015). We expect
that the adjustment speed is increasing in board independence because independent directors are
more likely to pressure managers to operate at their target level of tax avoidance. Consistent with
5
our expectations, we find that firms with greater board independence exhibit faster adjustment
speeds relative to firms with less independent boards.
This study contributes to the tax avoidance literature by providing a new method to
examine a firm’s tax avoidance decisions. Prior research investigates a wide variety of
determinants to help gain a better understanding of the substantial variation in the level of firms’
tax avoidance (Hanlon and Heitzman 2010). Because firms constantly adjust to their unique
target, our results suggest that the speed of adjustment to target levels of tax avoidance has been
an overlooked characteristic in the literature. If the construct of interest in a tax study is the
effectiveness of a firm’s tax function, the speed of adjustment may better capture the construct
relative to the level of tax avoidance. Furthermore, our findings suggest that the speed of
adjustment varies in the cross-section, providing indirect insight into the costs to firms of
adjusting their tax planning strategies. Finally, our study should be of interest to corporate
stakeholders. For tax managers, executives, investors, analysts, and members of the
compensation committee it can be more useful to evaluate the effectiveness of a firm’s tax
function by thinking in terms of the speed at which the firm adjusts to its target tax rate rather
than focusing on the level of tax avoidance.
The rest of the paper is organized as follows: Section II discusses prior literature. Section
III describes our research design. Section IV describes data and presents our results. Section V
presents our sensitivity analysis and Section VI concludes.
II.
Background and Prior Literature
Target Level of Tax Avoidance
The tradeoff literature in tax accounting suggests that effective tax planning considers all
of the potential taxes (both implicit and explicit) to a transaction, all of the parties in the
6
transaction, and all of the costs (both tax and non-tax) associated with a given transaction
(Scholes et al. 2014). Because each firm faces an idiosyncratic combination of potential taxes,
contracting parties, and non-tax costs, each firm has its own unique target level of tax avoidance.
One implication of the tradeoff view is that firms’ target level of tax avoidance is not going to be
equivalent to the maximum amount of tax that firms can avoid. Armstrong et al. (2015) argue
once firms exceed their optimal level of tax avoidance, the marginal costs of tax avoidance
exceed the marginal benefits. 5 Consistent with this notion, prior research provides evidence that
suggests high levels of tax avoidance is potentially costly. For example, Cook, Moser, and Omer
(2015) find that the cost of capital increases for firms with high levels of tax avoidance.
Likewise, Armstrong et al. (2015) find that better governance reduces tax avoidance, but only
among firms with extremely high levels of tax avoidance.
Another implication of the tradeoff literature is that a firm’s target level of tax avoidance
changes over time as the firm’s mix of potential taxes, parties, and costs changes. Changes to the
target level of tax avoidance could stem from changes in tax law, profitability, capital structure,
firm size, product mix, or geographic markets. For example, Gupta and Mills (2002) provide
evidence on how a firm’s tax planning opportunities change as they expand into new states.
Specifically, Gupta and Mills (2002) examine how firms exploit difference in state tax regimes
to lower their tax burden. Because each state has slightly different reporting requirements, firms
have the ability to locate their operations in states with favorable tax reporting requirements.
However, as firms expand to more states, they eventually must expand into states with less
favorable tax reporting requirements, which begins to eliminate their tax planning advantages.
5
Armstrong et al. (2015) note that the costs of tax avoidance include implementation costs, the inability to repatriate
and invest foreign earnings as well as potential political, regulatory, and reputational costs. Further, tax avoidance is
associated with reduced transparency (Balakrishnan, Blouin, and Guay 2012) as well as greater implicit taxes.
7
Gupta and Mills (2002) argue that a firm’s state tax planning opportunities increases when a firm
operates in more than one state, but fewer than all states. Consistent with their expectations, they
find that operating in 24 states yields the minimum state effective tax rate. Because firms’ target
level of tax avoidance changes over time due to such factors as expanding into new jurisdictions,
we expect that firms are constantly adjusting to their target level of tax avoidance.
It is possible firms will pursue the maximum amount of tax avoidance possible if the
marginal costs of excessive tax avoidance are not significant. Specifically, Jennings, Weaver,
and Mayew (2012) find that after the Tax Reform Act of 1986 (TRA86) implicit taxes are slow
to erode the after-tax benefits of new tax preferences. 6 They attribute the decline in implicit taxes
to an increased use of tax shelters and aggressive tax planning. They argue the proprietary nature
of these transactions makes it more difficult for competitive forces to equalize after-tax rates of
return. The lack of significant non-tax costs associated with aggressive tax planning strategies
has led to questions about why all firms do not use tax shelters (Weisbach 2002). Providing
further evidence that certain firms have the ability to avoid taxes without bearing meaningful
nontax costs, DeSimone et al. (2014) identify a group of U.S. MNCs whose operations allow
them numerous opportunities for low risk tax avoidance. DeSimone et al. (2014) note the
difficulty of taxing mobile capital, and that U.S. MNCs can locate valuable capital in low-tax
jurisdictions to avoid taxes with relatively low risk. Thus, to the extent that non-tax costs are not
significant, firms may not target a specific level of tax avoidance and instead attempt to
maximize their level of tax avoidance.
Tradeoffs of Adjusting to the Target Level of Tax Avoidance
8
The idea that managers will take steps to minimize the distance between their firm’s level
of tax avoidance and the target level of tax avoidance is similar to the tradeoff theory of capital
structure. The speed at which firms adjust to the target level of tax avoidance is a function of the
tradeoff between the costs of adjustment and the costs of failing to achieve their tax avoidance
target. If the costs of adjustment are zero then firms would never deviate from their target level
of tax avoidance. In contrast, if the costs of adjustment are infinite we would expect firms to
maintain a constant level of tax avoidance. The reality is between these two extreme cases,
suggesting that at any given point in time firms are moving toward their target level of
avoidance. Prior research on corporate tax avoidance fails to empirically investigate how quickly
firms converge to a target level of tax avoidance. 7 We fill this void.
As discussed above, the speed of adjustment is likely a function of the costs of
adjustment. Therefore, firms will likely adjust their tax avoidance activities more slowly as the
costs of adjustment, including logistic frictions, increase. To shed light on the costs of
adjustment and tradeoffs that firms make, we compare the speed of adjustment for firms with
actual levels of tax avoidance above and below their target. We expect to observe asymmetric
adjustment speeds to the extent that the cost of implementing a new tax strategy for firms below
their target level of tax avoidance differs from the cost of unwinding a current tax position for
firms above their target level of tax avoidance.
Because each firm faces a unique tradeoff when adjusting to its target tax avoidance
level, the tradeoff theory predicts that the speed of adjustment varies in the cross-section.
Accordingly, we examine two cases where we expect the costs of adjustment to be salient. First,
7
One exception is Hoopes, Mescall, and Pittman (2012). The authors conduct a survey of 25 tax executives and find
that 69.2 percent of all tax plans or positions can be changed within one year while approximately 90 percent of all
tax positions can be changed within two to three years. Our finding that the typical firm converges towards its target
at a rate of approximately 69 percent per year is consistent with this survey evidence.
9
we examine whether the speed of adjustment varies as a function of a firm’s growth potential.
Firms with high growth potential are more likely to enter new markets and introduce new
product lines relative to firms with lower growth potential. As firms with high growth potential
expand, they likely encounter shifts in the tax clientele that they belong to and, consequently,
significant changes in their target level of tax avoidance. Scholes et al. (2014) note that it is
costly for investors to acquire highly implicitly or explicitly taxed assets if they are not in the
correct tax clientele. We argue firms that are more likely to shift between tax clienteles across
time will find it more costly to commit to long-term tax planning strategies.
The 1981 Research and Development (R&D) tax credit provides an example of the costs
of operating in the wrong tax clientele. Berger (1993) examines the stock price reaction to the
R&D tax credit’s enactment for firms that compete for R&D factor inputs and customers, but
that are unable to receive tax credits because of low marginal tax rates. He notes that because the
pretax returns to R&D investments are bid down by the competition for the explicit tax benefits
offered by the tax credit that the tax subsidy created by the R&D tax credit only benefits firms
that can use the credit. Berger finds a significant negative market reaction to the credit’s
enactment for the low marginal tax rate firms and concludes the implicit tax costs for the firms
unable to use the credit are substantial. Thus, the costs of adjustment for high growth firms
includes the cost of engaging in additional tax planning as well as the present value of the
expected future costs associated with being in the wrong tax clientele. Because high growth
firms are likely to change tax clienteles frequently, their costs of adjustment likely exceed the
cost of failing to operate at their target level of tax avoidance. In contrast, because low growth
firms are better able to predict their tax clientele, the costs of not achieving their tax avoidance
10
target likely exceed the costs of adjustment. For this reason, we expect that high growth firms
will adjust to their target level of tax avoidance at a slower rate relative to low growth firms.
Second, we examine whether the speed of adjustment varies based on whether a firm has
foreign operations. Relative to a domestic-only corporation, multinational corporations have
substantial opportunities to avoid income tax. For example, MNCs have the opportunity to locate
operations in low-tax jurisdictions, shift income from high-tax to low-tax jurisdictions, exploit
differences between tax rules in different countries, and take advantage of tax subsidy
agreements with certain countries (Rego 2003). Consistent with MNCs having more
opportunities to avoid taxes, Rego (2003) finds that MNCs have lower worldwide effective tax
rates relative to domestic-only corporations. Rego’s (2003) finding suggest that MNCs enjoy
economies of scale with respect to tax planning and that the marginal cost of implementing
additional tax planning or reducing current tax avoidance is likely lower for MNCs relative to
domestic-only firms. Consequently, we predict that MNCs will adjust to their target level of tax
avoidance more quickly than domestic-only firms.
Finally, we examine whether the speed of adjustment varies with the quality of a firm’s’
corporate governance. Although corporate governance has multiple dimensions, we focus on the
independence of the board of directors because prior research suggests that it is an effective
monitor of firm operations. For example, Klein (2002) finds that firms with more independent
boards engage in less earnings management. Further, Armstrong et al. (2015) find that board
independence constrains additional tax avoidance for firms that are already engaging in
excessive tax avoidance and encourages additional tax avoidance for firms that are not engaging
in large amounts of tax avoidance. As discussed earlier, it is likely that the marginal costs exceed
the marginal benefits when firms do not operate at their target level of tax avoidance. To the
11
extent that managers and shareholders have different preferences for tax avoidance, corporate
governance mechanisms likely discipline managers to operate at the firm’s target level of tax
avoidance. Thus, we expect that firms with greater board independence will adjust more quickly
to their target level of tax avoidance.
III.
Research Design
To examine the tradeoff between the tax benefits and the nontax costs of operating at a
target level of tax avoidance, we employ a regression specification that allows each firm’s target
level of tax avoidance to vary over time. Our primary measure of tax avoidance is a firm’s cash
effective tax rate (CASHETR),
𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑡𝑡 =
∑𝑡𝑡𝑡𝑡−2 𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃
(1)
We use cash ETR as opposed to GAAP ETR for two reasons. First, GAAP ETRs do not
reflect tax avoidance activities that create temporary book-tax differences by deferring payment
of taxes to future periods (Hanlon and Heitzman 2010). Given that deferring taxes is a strategy
used by many firms (Dyreng et al. 2008), GAAP ETRs understate firms’ actual tax avoidance.
Second, financial accounting accruals such as the valuation allowance and tax contingency
reserves (or unrecognized tax benefits) also affect GAAP ETRs (Dyreng et al. 2008).
Consequently, changes to a firm’s GAAP ETR could be due to the reversal of accruals as
opposed to firms adjusting to their target level of tax avoidance.
We calculate a firm’s cash ETR over a three-year period to avoid measurement error
when it is calculated over a short time period (Dyreng et al. 2008). Specifically, the timing of tax
payments can occur in a different year than the corresponding pre-tax income and cause cash
ETR to be a misleading measure of tax avoidance when calculated over a single year.
12
Calculating cash ETR over a three-year period reduces measurement error related to the
mismatch between tax payments and pre-tax income. Furthermore, using a tax avoidance
measure calculated over a three-year period also reduces the possibility that our results are
simply due to mean reversion as opposed to firms strategically converging to a target level of tax
avoidance.
To estimate the average speed of adjustment to a target level of tax avoidance, we modify
Flannery and Rangan’s (2006) model that estimates how quickly firms adjust to their target
capital structure. Specifically, we first model a firm’s target level of tax avoidance as a function
of the firm’s operating, financing, and investment characteristics such that:
∗
𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖,𝑡𝑡+3
= 𝛽𝛽𝑿𝑿𝑖𝑖,𝑡𝑡
(2)
∗
is firm i’s desired level of tax avoidance over the period t + 1 to t + 3, 𝑿𝑿𝑖𝑖,𝑡𝑡
where 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐸𝐸𝐸𝐸𝐸𝐸𝑖𝑖,𝑡𝑡+3
is a vector of firm characteristics that have been shown to be determinants of tax avoidance, and
𝛽𝛽is a coefficient vector. Prior research suggests that economies of scale and firm complexity are
associated with additional tax planning opportunities (e.g., Mills, Erickson, and Maydew 1998;
Rego 2003). Accordingly, the vector of determinants includes firm size (SIZE), income from
foreign operations (FORINC), leverage (LEV), capital intensity (CAPINT), and research and
development activities (R&D).
In addition, the need to avoid income taxes varies with firm profitability (Chen, Chen,
Cheng, and Shevlin 2010; Rego 2003). Consequently, we include firm profitability (ROA), net
operating loss carryforwards (NOL and CNOL), and the number of pre-tax losses that a firm
experienced over the previous four years (LOSSINT). We include income related to the equity
method of accounting (EQINC) to control for differences in financial and tax accounting
treatment that influence our measures of tax avoidance (Chen et al. 2010; Frank, Lynch, and
13
Rego 2009). The vector also includes firms’ growth opportunities (MTB) because Chen et al.
(2010) note that growing firms potentially invest in more tax-favored assets. Please see
Appendix A for a detailed definition of the above variables. To the extent that the tradeoff
∗
literature is descriptive, we expect that 𝛽𝛽 ≠ 0 and that 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖,𝑡𝑡+3
will vary across firms and
over time.
Base Partial Adjustment Model
In a frictionless world, a firm will always maintain its target level of tax avoidance and
will never deviate from the target. However, the costs of adjusting a firm’s level of tax avoidance
are expected to be non-zero, and firms must trade off adjustment costs against the costs of failing
to operate at their target level of tax avoidance. Furthermore, given the dynamic components of
firms’ tax environments, it is possible that firms will constantly make adjustments to their level
of tax avoidance either because the firm has temporarily moved away from its target level of tax
avoidance or because the firm identified a new level of target tax avoidance. Therefore, we
estimate a model that allows partial adjustment of the firm’s level of tax avoidance toward its
target each fiscal year, which allows us to estimate the average adjustment speed and examine
whether firms exhibit targeting behavior.
The partial adjustment model is as follows:
∗
− 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖,𝑡𝑡 � + 𝜀𝜀𝑖𝑖,𝑡𝑡+3
𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖,𝑡𝑡+3 − 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖,𝑡𝑡 = 𝜃𝜃�𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐸𝐸𝐸𝐸𝐸𝐸𝑖𝑖,𝑡𝑡+3
(3)
Each fiscal year, the typical firm closes a proportion, 𝜃𝜃, of the gap between its actual and its
target level of tax avoidance. After substituting Eq. (2) into Eq. (3), rearranging gives the
estimable model:
𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖,𝑡𝑡+3 = (𝜃𝜃𝜃𝜃)𝑿𝑿𝑖𝑖,𝑡𝑡 + (1 − 𝜃𝜃)𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖,𝑡𝑡 + 𝜀𝜀𝑖𝑖,𝑡𝑡+3
(4)
14
Equation (4) implies that firms’ actual level of tax avoidance eventually converges to its target at
a given adjustment speed. Note that this specification implies all firms have the same adjustment
speed (𝜃𝜃). We interpret 𝜃𝜃 as the average adjustment speed for a typical firm, but we later
examine whether the adjustment speed varies in the cross-section.
The partial adjustment model is subject to two different criticisms. First, in a capital
structure context, Chang and Dasgupta (2009) examine the partial adjustment model on
simulated data that is designed not to exhibit targeting behavior. The authors find that the partial
adjustment model yields reasonable estimates of firms’ speed of adjustment on the simulated
data, calling into question the power of the model. Second, prior research argues that the
estimates produced by the partial adjustment model are due to mean reversion (Chen and Zhao,
2007; Chang and Dasgupta, 2009; Shyam-Sunder and Myers, 1999). Specifically, this stream of
research argues that because debt ratios are bounded at zero and one, firms with extreme debt
ratios are forced to revert to the mean because the debt ratio cannot become negative or exceed
one. However, Öztekin and Flannery (2012) note that if the speed of adjustment varies
predictably in the cross-section, it provides evidence that the partial adjustment model is properly
specified and not due to mean reversion. Accordingly, our tests that examine the cross-sectional
variation in the speed of adjustment provide additional insight into firms’ tax planning function
as well as validating our partial adjustment model. Further, as discussed below, we conduct
additional analysis to provide evidence that our results are not primarily attributable to mean
reversion. 8
8
We perform an analysis on the middle fifty percent of the CASHETR distribution which represents effective tax
rates ranging between 20.0% and 36.6%. We find similar adjustment speeds for this sample which helps rule out a
mean reversion explanation. We also note that the proportion of firm-year observations with a CASHETR of zero or
one is quite small. Specifically, we find that one percent of our firm-year observations have an CASHETR equal to
zero while none of our firm-year observations have an ETR equal to one. Furthermore, the fifth and ninety-fifth
percentile of the CASHETR distribution are 5.2% and 50.2% respectively indicating that extreme observations are
unlikely to be causing mean reversion in our sample.
15
Asymmetric Adjustment Speed
To examine the possibility of an asymmetric adjustment speed for firms above and below
their target level of tax avoidance, we first perform profile analyses. Specifically, we calculate
the proportion of the gap between the firm’s target and actual ETR that is closed within the next
𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶
−𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶
𝑖𝑖,𝑡𝑡
year: 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖,𝑡𝑡+3
. For example, consider a firm with a target ETR of 33% and an
∗
−𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶
𝑖𝑖,𝑡𝑡+3
𝑖𝑖,𝑡𝑡
actual ETR of 35%. If the firm adjusts its ETR to 34% in the following year, it closed half of the
gap between its target and actual ETR. We plot the median gap that is closed for each quartiles
based on the distance between actual and target ETRs.
Next, we conduct regression analyses to investigate whether the adjustment speed is
significantly different for firms above and below their target level of tax avoidance. We classify
firms based on whether the distance between their actual and target ETRs is positive or negative.
∗
If 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖,𝑡𝑡+3
− 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖,𝑡𝑡 is positive (negative), the firm has a level of tax avoidance that
is greater (less) than its target level. 9 We re-estimate Eq. (4) separately for firms above and
below their target level of tax avoidance. We then compare the adjustment speed coefficient, 𝜃𝜃,
across the two regressions and test for a significant difference.
Cross-sectional Variation in Adjustment Speed
We next examine whether a firm’s adjustment speed varies as a function of its growth
options, governance, and whether it is a multinational corporation. We use sales growth
(SALESGROWTH) to proxy for a firm’s growth options. SALESGROWTH is calculated by taking
the revenue growth over the prior three years. Firms are classified as multinational if they have
any foreign income over the three-year period. Following Armstrong et al. (2015), we use the
9
Byoun (2008) investigates whether firms above and below target debt ratios have different speeds of adjustment
toward their target debt ratios. The methodology we use to classify firms as above or below target is similar to the
methodology in Byoun (2008).
16
ratio of independent directors to proxy for governance (IND_RATIO). Firm-years are sorted by
SALESGROWTH and IND_RATIO within each fiscal year. We estimate equation (4) separately
for firms with above and below median SALESGROWTH and IND_RATIO as well as for
domestic and multinational firms. Finally, we compare the adjustment speed coefficient, 𝜃𝜃,
across the various regressions and test for significant differences.
IV.
Results
Sample Selection
Our sample includes all firms with available data in the COMPUSTAT industrial annual
files from 1990 through 2011. The sample begins in 1990 because three years of Statement of
Cash Flows data is required to calculate CASHETR. The sample ends in 2011 to ensure that data
is available to calculate three-year ahead CASHETR. We exclude financial institutions (SIC
codes 6000–6999), utilities (SIC codes 4900–4999) because they operate in regulated industries
and face different tax planning incentives. Firms incorporated outside the United States are also
eliminated as they face a different set of tax and/or financial reporting rules. We delete any
observations without the available data to calculate the variables described in Appendix A,
including observations with non-positive values of book equity. We also delete firm-year
observations with a negative denominator in the calculation of CASHETR as it is unclear if loss
firms have a target level of tax avoidance. Observations with beginning total assets below $10
million are also deleted to avoid scaling issues with extremely small firms. Finally, we delete
observations with an CASHETR outside of the [0,1] range to avoid the influence of outliers. To
further avoid influence of outliers, we winsorize all continuous variables other than CASHETR at
the 1st and 99th percentile. The sample selection criteria produces a final sample of 24,762 firmyears.
17
Descriptive Statistics
Table 1 provides the descriptive statistics for our sample of firm-year observations. The
mean (median) CASHETR is 0.288 (0.294) which is comparable to three-year CASHETR
measures in prior studies. Note that the control variables are measured over the same three-year
window as CASHETR. Income statement variables are accumulated over the three years while
balance sheet variables are averaged. For example, ROA is calculated by summing three years of
income statement data and dividing by average total assets over the period.
[INSERT TABLE 1 HERE]
Partial Adjustment to Target Level of Tax Avoidance
Table 2 presents the estimation of the partial adjustment model. In addition to the
common determinants of tax avoidance discussed previously, the regressions in Columns 1-2
include industry and year fixed-effects. The coefficient on CASHETR is our primary interest
because it is the estimation of (1 − 𝜃𝜃), with 𝜃𝜃 representing the average adjustment speed for a
typical firm. The base model in Column 1 shows a coefficient on CASHETR of 0.306 implying
an adjustment speed of 69 percent. Stated differently, a firm closes 69 percent of the gap between
its actual and target levels of tax avoidance in the following three-year period. This adjustment
speed suggests that firms are able to rapidly adjust to their target level of tax avoidance despite
the adjustment costs associated with altering tax planning strategies. In addition, the adjustment
speed for tax avoidance is significantly higher than the adjustment speed documented in the
capital structure literature, which suggests that the costs of adjusting towards a tax avoidance
target are significantly less than the costs of adjusting towards a capital structure target. 10 The
magnitude of the estimated adjustment speed is also significant because it indicates that firms
10
For example, Flannery and Rangan (2006) estimate that the firms adjust toward their target capital structure at a
rate of 34 percent per year.
18
actively attempt to adjust their level of tax avoidance to achieve a specific target. In the context
of capital structure adjustments, Fama and French (2002) note that slow adjustment speeds do
not provide support for firms pursuing a target level of capital structure. Although, the speed of
adjustment towards a tax avoidance target is significantly different than the speed of adjustment
towards a capital structure target, Fama and French’s assertion applies in the tax avoidance
context.
It is possible that the estimated speed of adjustment is driven by mean reversion.
Specifically, observations that are farther from their target are more likely to have relatively low
or high levels of tax avoidance, and firms with a relatively low or high level of tax avoidance
may have a tendency to move back toward the mean level of tax avoidance. To investigate
whether targeting behavior is apparent in the firms with less extreme levels of tax avoidance, we
follow Flannery and Rangan (2006) and rank observations by CASHETR for each fiscal year. In
Column 2, we present the results of estimating equation (4) on only the middle 50% of the
CASHETR distribution, quartiles 2 and 3. The coefficient on CASHETR is 0.377 which translates
into an adjustment speed of approximately 62%. Though a tendency for extreme observations to
mean revert may partially explain the rapid adjustment speed in Column 1, the fact that the
adjustment speed is 62% for the middle quartiles provides strong evidence of targeting behavior.
[INSERT TABLE 2 HERE]
The partial adjustment model implies that a firm’s actual level of tax avoidance
eventually converges to its target level of tax avoidance. If our model estimates meaningful
targets, we should find that firms begin to adjust to these targets the following year. To
investigate whether the sample firms adjust their level of tax avoidance towards our estimated
targets, we first split firm-years into two groups based on whether the distance between their
19
∗
actual and target levels of tax avoidance, �𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖,𝑡𝑡 −𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖,𝑡𝑡+3
� is positive or negative.
For each group, we rank firm-years as above or below the median based on the distance between
their actual and target levels of tax avoidance. We next plot the mean and median year-over-year
change in CASHETR for the following four groups: Far Below Target, Below Target, Above
Target, Far Above Target. Figure 1 provides further evidence of targeting behavior. Firms with
below target CASHETR increase their CASHETR in the following year while the opposite is true
for firms with above target CASHETR.
[INSERT FIGURE 1 HERE]
Figure 1 also shows that the mean change in CASHETR is substantially greater than the
median change. Also of note, firms that are farther away from their target (far left and far right
bars) appear to close a greater proportion of the gap between their actual and target CASHETR
in the next year then firms that begin closer to their target (middle two bars). A faster adjustment
speed for the extreme observations suggests the costs of a firm failing to achieve its target
CASHETR are more likely to exceed the adjustment costs for firms that are further away from
their target. When a firm is already near its target, adjustment costs could be more likely to
exceed the costs of not achieving its target CASHETR because it is difficult to identify costeffective tax strategies that create small changes in the CASHETR .
Asymmetric Adjustment Speed Analyses
Figure 2 shows the proportion of the gap between a firm’s target and actual CASHETR
𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐸𝐸𝐸𝐸𝐸𝐸
−𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶
𝑖𝑖,𝑡𝑡
, for the median firm in each of the
that is closed within the next period, 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖,𝑡𝑡+3
∗
−𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶
𝑖𝑖,𝑡𝑡+3
𝑖𝑖,𝑡𝑡
groups from Figure 1. Firms that are above their target CASHETR appear to close a larger
proportion of the gap between their target and actual level of tax avoidance in the next period.
The Far Above Target group closes 75.6% of the gap in the following year while the Far Below
20
Target group closes 45.7%. Overall, Figure 2 provides initial visual evidence that the adjustment
speeds may be asymmetric for firms with above and below target CASHETRs.
[INSERT FIGURE 2 HERE]
Table 3 presents the results of estimating equation (4) separately for firms with below
target CASHETRs and firms with above target CASHETRs. Recall, the tradeoff literature predicts
that the adjustment speed will be between zero and one for firms whose actual CASHETR is
below the target. Column 1 shows that firms with below target CASHETRs have an adjustment
speed of approximately 47% (1 – 0.532) whereas Column 2 shows that firms with above target
CASHETRs have an adjustment speed of approximately 89% (1 – 0.107). In combination, these
results are consistent with the tradeoff theory’s prediction that managers actively attempt to
achieve a target level of tax avoidance regardless of whether their initial level of tax avoidance is
above or below their target. In addition, the difference in adjustment speeds between the two
columns is significant at the 1% level. A faster adjustment speed for firms with above target
CASHETRs suggests that firms engaged in too little tax avoidance increase their tax avoidance
more quickly than the rate at which firms engaged in too much tax avoidance scale back their tax
avoidance.
[INSERT TABLE 3 HERE]
Cross-sectional Variation in Adjustment Speed Analyses
Table 4 presents the results of estimating equation (4) for subsamples partitioned on sales
growth. Consistent with our expectations, firms with high sales growth have a slower adjustment
speed than firms with low sales growth. High sales growth firms have an adjustment speed of
63%. In contrast, low sales growth firms have an adjustment speed of 74%. The difference in
adjustment speeds between Columns 1 and 2 is significant at the 1% level. A slower adjustment
21
speed suggests that the costs of failing to meet the tax avoidance target are less likely to
outweigh the adjustment costs for high sales growth firms. This result is consistent with firms
with high sales growth being more likely to change tax clienteles in the future, which reduces the
net present value of the benefits to operating at its target level of tax avoidance. In other words, a
high sales growth firm does not see the benefits in quickly adjusting its target level of tax
avoidance because its target is likely to change in the near future as a result of its growth.
[INSERT TABLE 4 HERE]
Table 5 presents the results of estimating equation (4) separately for domestic and
multinational corporations (MNCs). Consistent with our expectations, MNCs have a faster
adjustment speed than domestic firms. MNCs have an adjustment speed of 73% compared to
69% for domestic firms. The difference in adjustment speeds is significant at the 10% level. This
result is consistent with MNCs enjoying economies of scale with respect to tax planning and
having lower costs of adjusting to a target level of tax avoidance.
[INSERT TABLE 5 HERE]
Table 6 presents the results of estimating equation (4) for subsamples partitioned on
board independence. Consistent with our expectations, firms with a higher ratio of independent
directors have a faster adjustment speed than firms with less independent boards. Firms with an
above median ratio of independent directors have an adjustment speed of 73%. In contrast, firms
with a below median ratio of independent directors have an adjustment speed of 69%. The
difference in adjustment speeds between the subsamples is significant at the 10% level. This
result is consistent with the finding in Armstrong et al. (2015) that more independent boards have
greater ability and incentives to monitor managers’ tax avoidance decisions.
22
[INSERT TABLE 6 HERE]
V.
Sensitivity Analysis
Alternative Measure of Tax Avoidance
As discussed above, we proxy for tax avoidance using the cash effective tax rate
measured over a three-year period (CASHETR). We chose CASHETR as our proxy to capture
temporary tax avoidance strategies and to avoid the influence of tax accruals such as reserves for
uncertain tax positions and valuation allowances. However, prior research finds that GAAP ETR
is the tax avoidance measure that is most important to public firms (Graham et al. 2014). To
confirm our results are robust to using GAAP ETR, we replicate our analysis in Table 2 using
GAAP ETR (GAAPETR) as our proxy for tax avoidance. We still measure GAAPETR over a
three-year period in an effort to mitigate the effect of extreme observations which could cause
mechanical mean reversion. Specifically, we modify equation (4) to estimate the following:
𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝑖𝑖,𝑡𝑡+3 = (𝜃𝜃𝜃𝜃)𝑿𝑿𝑖𝑖,𝑡𝑡 + (1 − 𝜃𝜃)𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐸𝐸𝐸𝐸𝐸𝐸𝑖𝑖,𝑡𝑡 + 𝜀𝜀𝑖𝑖,𝑡𝑡+3
(5)
where GAAPETRt+3 is measured over years t+1 to t+3 and GAAPETRt is measured over years t-2
to t. All other control variables are measured as defined in Appendix A.
Table 7 presents the results of this sensitivity analysis. We find that the estimated
adjustment speed is similar when we use the GAAPETR as our measure of tax avoidance. The
convergence speed is approximately 73 percent when GAAPETR serves as our proxy for tax
avoidance. For comparison purposes, we find that the convergence speed is approximately 69
percent when CASHETR serves as our proxy for tax avoidance. This result suggests that the
convergence speeds we document are robust to using GAAP ETR as the measure of tax
avoidance.
23
VI.
Conclusion
In an effort to explain the variation in tax avoidance, recent research examines the
determinants of the level of firms’ tax avoidance (e.g., Chen et al. 2010; Cheng et al. 2012;
Dyreng et al. 2010; McGuire et al. 2014). While determinant studies provide valuable insight, the
full implications of their findings are not clear because they do not consider firms’ tax planning
objectives. The tradeoff literature (e.g., Scholes, Wilson, and Wolfson 1990) implies that firms
select target levels of tax avoidance based on the costs and benefits. Despite the clarity of the
tradeoff literature’s prediction, there is limited evidence on whether firms target a specific level
of tax avoidance. The purpose of this study is to examine whether firms adjust to a target level of
tax avoidance and, if so, how quickly firms converge toward their predicted target.
Our results suggest firms do target a specific level of tax avoidance and that the typical
firm converges toward its target level of tax avoidance at a rate of almost 69 percent per year. In
addition, we find that firms whose actual cash ETR is below their target exhibit an adjustment
speed of approximately 47 percent per year while the average speed of convergence for firms
whose actual cash ETR is above their target is approximately 89 percent per year. In
combination, these results are consistent with the tradeoff theory’s prediction that managers
actively attempt to achieve a target level of tax avoidance regardless of whether their initial level
of tax avoidance is above or below their target. Further, our results suggest firms that are
engaged in too little tax avoidance increase their tax avoidance more quickly than the rate at
which firms that are engaged in too much tax avoidance scale back their tax avoidance.
Finally, we investigate whether there is cross-sectional variation in the speed of
adjustment. We examine whether the adjustment speed varies based on a firm’s growth options
and find that firms with low sales growth exhibit significantly faster adjustment speeds to their
24
target level of tax avoidance. We next examine whether the speed of adjustment varies with the
presence of foreign operations and find that MNCs experience faster adjustment speeds than
domestic firms. We also find the speed of adjustment varies with board independence such that
firms with more independent boards have faster adjustment speeds. Together, these results
suggest firms that are less likely to shift tax clienteles, firms with more tax planning
opportunities, and firms with better governance adjust to their target level of tax avoidance more
quickly.
In summary, our results imply that managers attempt to minimize deviations between
actual and target levels of tax avoidance rather than minimize taxes. The finding that firms move
toward target levels of tax avoidance supports the tradeoff view of tax planning. Given that each
firm faces a unique mix of costs associated with any tax strategy, the optimal level of tax
avoidance differs for each firm. As such, it is not possible to simply examine the level of firms’
effective tax rates to evaluate the effectiveness of the firm’s tax function. Because firms
constantly adjust to their unique target, our results suggest that the speed of adjustment to target
levels of tax avoidance has been an overlooked characteristic in the literature. If the construct of
interest in a tax study is the effectiveness of a firm’s tax function, the speed of adjustment may
better capture the construct relative to the level of tax avoidance. For example, if a researcher is
interested in how executive characteristics impact the effectiveness of a firm’s tax function,
examining adjustment speeds may be more appropriate than examining levels of tax avoidance.
Furthermore, our findings suggest that the speed of adjustment varies in the cross-section,
providing indirect insight into the costs to firms of adjusting their tax planning strategies. Finally,
our study should be of interest to corporate stakeholders. For tax managers, executives, investors,
analysts, and members of the compensation committee it can be more useful to evaluate the
25
effectiveness of a firm’s tax planning by thinking in terms of the speed at which the firm adjusts
to its target tax rate rather than focusing on the level of tax avoidance.
26
References
Armstrong, C., Blouin, J., Jagolinzer, A., and Larcker, D. (2015). “Corporate governance,
incentives, and tax avoidance.” Journal of Accounting and Economics, 60 (1) (2015): 1-17.
Balakrishnan, K., J. Blouin, and W. Guay. “Does Tax Aggressiveness Reduce Corporate
Transparency?” Working paper (2012), London Business School and University of
Pennsylvania.
Berger, P. “Explicit and implicit tax effects of the R & D tax credit.” Journal of Accounting
Research (1993): 131-171.
Byoun, S. “How and when do firms adjust their capital structures toward targets?.” The Journal
of Finance 63.6 (2008): 3069-3096.
Chang, X., and S. Dasgupta. “Target behavior and financing: How conclusive is the evidence?”
Journal of Finance 64.4 (2009): 1767-1796.
Chen, L., and X. Zhao. “Mechanical mean reversion of leverage ratios.” Economics Letters 95
(2007): 223-229.
Chen, S., X. Chen, Q. Cheng, and T. Shevlin. “Are family firms more tax aggressive than nonfamily firms?.” Journal of Financial Economics 95.1 (2010): 41-61.
Cook, K., W. Moser, and T. Omer. “Towards an optimal level of tax avoidance.” Working paper
(2015), Texas Tech University.
Dittmar, A., and R. Duchin. “The dynamics of cash.” Working paper (2010), University of
Michigan.
Dyreng, S., M. Hanlon, and E. Maydew. “Long-run corporate tax avoidance.” The Accounting
Review 83.1 (2008): 61-82.
Fama, E., and K. French. “Testing trade-off and pecking order predictions about dividends and
debt.” Review of Financial Studies 15.1 (2002): 1-34.
Flannery, M., and K. Rangan. “Partial adjustment toward target capital structures.” Journal of
Financial Economics 79.3 (2006): 469-506.
Frank, M., L. Lynch, and S. Rego. “Tax reporting aggressiveness and its relation to aggressive
financial reporting.” The Accounting Review 84.2 (2009): 467-496.
Graham, J., M. Hanlon, and T. Shevlin. “Real effects of accounting rules: Evidence from
multinational firms’ investment location and profit repatriation decisions.” Journal of
Accounting Research 49.1 (2011): 137-185.
27
Graham, J., M. Hanlon, T. Shevlin, and N. Shroff. “Incentives for tax planning and avoidance:
Evidence from the field.” The Accounting Review 89.3 (2014): 991-1023.
Gupta, S., and L. Mills. "Corporate multistate tax planning: benefits of multiple jurisdictions."
Journal of Accounting and Economics 33.1 (2002): 117-139.
Hanlon, M., and S. Heitzman. “A review of tax research.” Journal of Accounting and Economics
50.2 (2010): 127-178.
Hoopes, J., D. Mescall, and J. Pittman. “Do IRS audits deter corporate tax avoidance?” The
Accounting Review 87.5 (2012): 1603-1639.
Mills, L., M. Erickson, and E. Maydew. “Investments in tax planning.” Journal of the American
Taxation Association 20.1 (1998): 1-20.
Öztekin, Ö., and M. Flannery. “Institutional determinants of capital structure adjustment speeds.”
Journal of Financial Economics 103.1 (2012): 88-112.
Rego, S. “Tax‐Avoidance Activities of US Multinational Corporations.” Contemporary
Accounting Research 20.4 (2003): 805-833.
Scholes, M., M. Wolfson, M. Erickson, M. Hanlon, E. Maydew, and T. Shevlin. “Taxes and
Business Strategy: A Planning Approach. (2014).”
Shackelford, D., and T. Shevlin. “Empirical tax research in accounting.” Journal of Accounting
and Economics 31.1 (2001): 321-387.
Shyam-Sunder, L., and S. Myers. “Testing static tradeoff against pecking order models of capital
structure.” Journal of Financial Economics 51 (1999): 219-244.
Tax Executives Institute, Inc. “Corporate tax department survey.” Tax Executives Institute, Inc.
(2004-2005). Washington, D.C.
Venkiteshwaran, V. “Partial adjustment toward optimal cash holding levels.” Review of
Financial Economics 20.3 (2011): 113-121.
28
Appendix A: Variable Definitions
Variable
Defintion
Dependent Variable 1
The sum of cash taxes paid from year t-2 to year t divided by the sum of pre-tax book income less special items over
CASHETR
∑𝑡𝑡𝑡𝑡−2 𝐶𝐶𝑋𝐺𝐺𝐷
the same period:
GAAPETR
∑𝑡𝑡𝑡𝑡−2 𝐺𝐺𝐼 − 𝐶𝐶𝐺𝐺𝐼
The sum of cash taxes paid from year t-2 to year t divided by the sum of pre-tax book income less special items over
the same period:
∑𝑡𝑡𝑡𝑡−2 𝐶𝐶𝑋𝐶𝐶
2
∑𝑡𝑡𝑡𝑡−2 𝐺𝐺𝐼 − 𝐶𝐶𝐺𝐺𝐼
Tax Avoidance Determinants
Sum of pre-tax income (COMPUSTAT PI) less extraordinary items (COMPUSTAT XI) over the three-year period
ROA
divided by average total assets (COMPUSTAT AT).
Average long-term debt over the three-year period (COMPUSTAT DLTT) scaled by average total assets
LEV
(COMPUSTAT AT).
Indicator variable equal to one if the firm has a tax loss carryforward (COMPUSTAT TLCF is positive) at anytime
NOL
during years t-2 to t ; zero otherwise.
Change in tax-loss carryforward (COMPUSTAT TLCF) from beginning of year t-2 to end of year t scaled by average
CNOL
total assets.
Sum of pre-tax foreign income over the three-year period (COMPUSTAT PIFO) scaled by average total assets
FORINC
(COMPUSTAT AT).
Average net PPE over the three-year period (COMPUSTAT PPENT) scaled by average total assets (COMPUSTAT
CAPINT
AT).
Sum of equity income over the three-year period (COMPUSTAT ESUB) scaled by average total assets
EQINC
(COMPUSTAT AT).
Sum of research and development expense over the three-year period (COMPUSTAT XRD) scaled by average total
RD
assets (COMPUSTAT AT).
MTB
SIZE
LOSSINT
The average of the market-to-book ratio at the beginning of year t-2 and at the end of year t . The market-to-book ratio
is measured as market value of equity (COMPUSTAT PRCC_F x CSHO) divided by book value of equity
(COMPUSTAT CEQ).
Natural log of average total assets over the three-year period (COMPUSTAT AT).
Loss intensity over the previous four year period defined as the number of years a firm has negative pre-tax book income
(COMPUSTAT PI) from year t-4 to year t-1 scaled to range from [0,1].
Cross-Sectional Variables
SALESGROWTH Sales growth from year t-3 to year t (COMPUSTAT REVT).
IND_RATIO
The ratio of independent directors as of the meeting that occurred in the fiscal year t+3 .
1
Observations with a negative denominator in the calculation of CASHETR are deleted. Observations with an CASHETR outside of the
[0,1] range are also deleted.
2
The determinants, with the exception of NOL and LOSSINT , are winsorized at the 1st and 99th percentile. Missing values of DLTT,
ESUB, PIFO, PPENT, SPI, TLCF, XI, and XRD are set to zero. Balance sheet variables are the average of the data item at the beginning
of year t-2 and at the end of year t . Income statement variables are summed over years t-2 through t .
29
Table 1
Descriptive Statistics
Variable
CASHETR
ROA
LEV
NOL
CNOL
FORINC
CAPINT
EQINC
RD
MTB
SIZE
LOSSINT
SALESGROWTH
IND_RATIO
N
24,762
24,762
24,762
24,762
24,762
24,762
24,762
24,762
24,762
24,762
24,762
24,762
24,762
12,056
Mean
Std Dev
0.288
0.140
0.335
0.224
0.166
0.140
0.315
0.465
0.001
0.049
0.047
0.089
0.300
0.212
0.002
0.010
0.070
0.123
2.871
2.266
6.299
1.827
0.071
0.154
0.469
0.651
0.700
0.166
Q1
0.200
0.174
0.039
0.000
0.000
0.000
0.139
0.000
0.000
1.505
4.953
0.000
0.095
0.600
Median
0.294
0.294
0.148
0.000
0.000
0.000
0.250
0.000
0.000
2.215
6.186
0.000
0.302
0.727
Q3
0.366
0.451
0.257
1.000
0.000
0.060
0.413
0.000
0.086
3.384
7.506
0.000
0.628
0.833
st
th
*We winsorize all continuous variables other than CASHETR at the 1 and 99 percentile. CASHETR is
truncated at [0,1]. Please see Appendix A for full variable definitions.
30
Table 2
Partial Adjustment Model to Target Level of Tax Avoidance
𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐸𝐸𝐸𝐸𝐸𝐸𝑖𝑖,𝑡𝑡+3 = (𝜃𝜃𝜃𝜃)𝑿𝑿𝑖𝑖,𝑡𝑡 + (1 − 𝜃𝜃)𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐸𝐸𝐸𝐸𝐸𝐸𝑖𝑖,𝑡𝑡 + 𝜀𝜀𝑖𝑖,𝑡𝑡+3
(4)
Where 𝑿𝑿 includes an intercept term, ROA, LEV, NOL, CNOL, FORINC, CAPINT, EQINC, RD,
MTB, SIZE, and LOSSINT.
CASHETR t+3
CASHETR t+3
(1) Base Model
(2) Middle 50%
Coefficient
(Std Error)
Coefficient
(Std Error)
0.306***
(0.011)
0.057***
(0.007)
-0.030***
(0.011)
-0.006**
(0.003)
-0.043*
(0.023)
0.025
(0.017)
-0.059***
(0.009)
0.175
(0.129)
-0.102***
(0.013)
-0.004***
(0.001)
-0.006***
(0.001)
-0.010
(0.009)
0.377***
(0.027)
0.033***
(0.008)
-0.011
(0.013)
-0.003
(0.003)
-0.048
(0.038)
0.020
(0.018)
-0.038***
(0.010)
0.228*
(0.136)
-0.091***
(0.018)
-0.002***
(0.001)
-0.006***
(0.001)
0.022
(0.014)
Adjustment Speed
69.4%
62.3%
Number of Observations
24,762
12,393
0.243
Yes
Yes
0.150
Yes
Yes
Variable
CASHETR t
ROA
LEV
NOL
CNOL
FORINC
CAPINT
EQINC
RD
MTB
SIZE
LOSSINT
2
R
Industry and Year Fixed Effects
Standard Errors Clustered by Firm
*, **, and *** denote significance at the p < 0.10, 0.05, and 0.01 levels, respectively. The adjustment
speed is calculated as one minus the coefficient on CASHETR . Please see Appendix A for full variable
definitions.
31
Table 3
Asymmetric Adjustment to Target Level of Tax Avoidance
Variable
CASHETR t
ROA
LEV
NOL
CNOL
FORINC
CAPINT
EQINC
RD
MTB
SIZE
LOSSINT
Adjustment Speed
CASHETR t+3
(1) Below Target
Coefficient
(Std Error)
CASHETR t+3
(2) Above Target
Coefficient
(Std Error)
0.532***
(0.023)
0.007
(0.009)
-0.038***
(0.013)
-0.000
(0.003)
-0.097***
(0.023)
-0.014
(0.019)
-0.057***
(0.011)
0.170
(0.181)
-0.061***
(0.017)
-0.002**
(0.001)
-0.008***
(0.001)
0.022**
(0.011)
0.107***
(0.019)
0.053***
(0.009)
-0.013
(0.015)
-0.012***
(0.004)
0.054
(0.043)
0.069***
(0.025)
-0.053***
(0.012)
0.175
(0.164)
-0.128***
(0.019)
-0.005***
(0.001)
-0.006***
(0.001)
-0.017
(0.013)
46.8%
89.3%
Test of difference in CASHETR coefficient
Number of Observations
2
R
Industry and Year Fixed Effects
Standard Errors Clustered by Firm
chi2(1) = 91.32
Prob > chi2 = 0.0000 ***
12,518
12,244
0.274
Yes
Yes
0.147
Yes
Yes
*, **, and *** denote significance at the p < 0.10, 0.05, and 0.01 levels, respectively. The
adjustment speed is calculated as one minus the coefficient on CASHETR . Please see Appendix A
for full variable definitions.
32
Table 4
Sales Growth and Cross-Sectional Variation in Adjustment Speeds
Variable
CASHETR t
ROA
LEV
NOL
CNOL
FORINC
CAPINT
EQINC
RD
MTB
SIZE
LOSSINT
Adjustment Speed
CASHETR t+3
CASHETR t+3
(1) Below Median
SALESGROWTH
(2) Above Median
SALESGROWTH
Coefficient
(Std Error)
Coefficient
(Std Error)
0.257***
(0.014)
0.064***
(0.009)
-0.029**
(0.014)
-0.010***
(0.004)
-0.029
(0.036)
0.050**
(0.024)
-0.037***
(0.012)
0.008
(0.167)
-0.100***
(0.019)
-0.004***
(0.001)
-0.007***
(0.001)
-0.025**
(0.012)
0.374***
(0.017)
0.045***
(0.009)
-0.032**
(0.014)
-0.002
(0.003)
-0.056**
(0.027)
0.007
(0.020)
-0.074***
(0.012)
0.401**
(0.171)
-0.099***
(0.016)
-0.003***
(0.001)
-0.005***
(0.001)
0.008
(0.012)
74.3%
62.6%
Test of difference in CASHETR coefficient
Number of Observations
2
R
Industry and Year Fixed Effects
Standard Errors Clustered by Firm
chi2(1) = 40.86
Prob > chi2 = 0.0000 ***
12,376
12,386
0.228
Yes
Yes
0.276
Yes
Yes
*, **, and *** denote significance at the p < 0.10, 0.05, and 0.01 levels, respectively. The
adjustment speed is calculated as one minus the coefficient on CASHETR . Please see Appendix A
for full variable definitions.
Table 5
Foreign Operations and Cross-Sectional Variation in Adjustment Speeds
Variable
CASHETR t
ROA
LEV
NOL
CNOL
CASHETR t+3
(1) Domestic
Coefficient
(Std Error)
CASHETR t+3
(2) Multinational
Coefficient
(Std Error)
0.313***
(0.008)
0.052***
(0.006)
-0.047***
(0.009)
-0.009***
(0.003)
-0.064***
(0.024)
-0.051***
(0.007)
0.173
(0.116)
-0.050***
(0.014)
-0.003***
(0.001)
-0.005***
(0.001)
-0.009
(0.008)
0.273***
(0.009)
0.062***
(0.007)
-0.006
(0.011)
-0.005**
(0.002)
-0.013
(0.021)
0.016
(0.012)
-0.065***
(0.009)
0.125
(0.108)
-0.129***
(0.010)
-0.004***
(0.001)
-0.008***
(0.001)
-0.008
(0.008)
68.7%
72.7%
FORINC
CAPINT
EQINC
RD
MTB
SIZE
LOSSINT
Adjustment Speed
Test of difference in CASHETR coefficient
Number of Observations
2
R
Industry and Year Fixed Effects
Standard Errors Clustered by Firm
chi2(1) = 3.23
Prob > chi2 = 0.0722 *
12,789
11,973
0.283
Yes
Yes
0.212
Yes
Yes
*, **, and *** denote significance at the p < 0.10, 0.05, and 0.01 levels, respectively. The
adjustment speed is calculated as one minus the coefficient on CASHETR . Please see Appendix
A for full variable definitions.
34
Table 6
Board Independence and Cross-Sectional Variation in Adjustment Speeds
Variable
CASHETR t
ROA
LEV
NOL
CNOL
FORINC
CAPINT
EQINC
RD
MTB
SIZE
LOSSINT
Adjustment Speed
CASHETR t+3
CASHETR t+3
(1) Below Median
IND_RATIO
(2) Above Median
IND_RATIO
Coefficient
(Std Error)
Coefficient
(Std Error)
0.315***
(0.012)
0.062***
(0.009)
-0.036***
(0.014)
0.002
(0.003)
-0.003
(0.032)
-0.011
(0.018)
-0.058***
(0.011)
0.318**
(0.160)
-0.079***
(0.015)
-0.003***
(0.001)
-0.003***
(0.001)
-0.036***
(0.012)
0.266***
(0.012)
0.070***
(0.009)
-0.005
(0.014)
-0.004
(0.003)
-0.083***
(0.030)
-0.008
(0.017)
-0.083***
(0.011)
0.532***
(0.128)
-0.132***
(0.014)
-0.003***
(0.001)
-0.004***
(0.001)
0.007
(0.011)
68.5%
73.4%
Test of difference in CASHETR coefficient
Number of Observations
2
R
Industry and Year Fixed Effects
Standard Errors Clustered by Firm
chi2(1) = 2.82
Prob > chi2 = 0.0929 *
5,997
6,059
0.271
Yes
Yes
0.227
Yes
Yes
*, **, and *** denote significance at the p < 0.10, 0.05, and 0.01 levels, respectively. The
adjustment speed is calculated as one minus the coefficient on CASHETR . Please see Appendix
A for full variable definitions.
35
Table 7
Partial Adjustment Model Using an Alternative Measure of Tax Avoidance
𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐸𝐸𝐸𝐸𝐸𝐸𝑡𝑡+1,𝑡𝑡+3 = (𝜃𝜃𝜃𝜃)𝑿𝑿𝑖𝑖,𝑡𝑡 + (1 − 𝜃𝜃)𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐸𝐸𝐸𝐸𝐸𝐸𝑡𝑡−2,𝑡𝑡 + 𝜀𝜀𝑖𝑖
(5)
Where 𝑿𝑿 includes an intercept term, ROA, LEV, NOL, CNOL, FORINC, CAPINT, EQINC, RD,
MTB, SIZE, and LOSSINT.
Variable
GAAPETR t
ROA
LEV
NOL
CNOL
FORINC
CAPINT
EQINC
RD
MTB
SIZE
LOSSINT
GAAPETR t+3
Coefficient
(Std Error)
0.270***
(0.016)
0.020***
(0.006)
-0.006
(0.009)
-0.003
(0.002)
-0.027
(0.020)
-0.072***
(0.015)
-0.017**
(0.007)
-0.025
(0.115)
-0.073***
(0.011)
-0.001
(0.000)
-0.004***
(0.001)
0.001
(0.007)
Adjustment Speed
73.0%
Number of Observations
24,762
2
Adj R
Industry and Year Fixed Effects
Standard Errors Clustered by Firm
0.204
Yes
Yes
*, **, and *** denote significance at the p < 0.10, 0.05, and 0.01 levels,
respectively. The adjustment speed is calculated as one minus the coefficient on
GAAPETR . Please see Appendix A for full variable definitions.
36
Figure 1
Cash ETR Changes in Following Period
Figure 2
Percentage of Distance from Target Closed in Following Year
38