CORPORATE TAX AVOIDANCE AND FIRM VALUE

Author(s): Mihir A. Desai and Dhammika Dharmapala

Source: The Review of Economics and Statistics, Vol. 91, No. 3 (August 2009), pp. 537-546
Published by: The MIT Press
Stable URL: http://www.jstor.org/stable/25651357
Accessed: 21-09-2015 13:02 UTC

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@jstor.org.

The MIT Press is collaborating with JSTOR to digitize, preserve and extend access to The Review of Economics and Statistics.

http://www.jstor.org

This content downloaded from 132.211.1.50 on Mon, 21 Sep 2015 13:02:52 UTC
All use subject to JSTOR Terms and Conditions
 CORPORATE TAX AVOIDANCE AND FIRM VALUE
Mihir A. Desai and Dhammika Dharmapala*

Abstract?Do corporate tax avoidance activities advance shareholder of tax avoidance as a transferof value to shareholders. First,
interests? This paper tests alternative theories of corporate tax avoidance
corporate expatriations?transactions where U.S. firms in
using unexplained differences between income reported to capital markets
and to tax authorities. OLS estimates indicate that the effect of tax vert their corporate structure so that a subsidiary in a tax
on firm value is a function of firm governance, as predicted by
significant cor
avoidance haven becomes the parent entity?provide
an agency perspective on corporate tax avoidance. Instrumental variables
estimates based on exogenous changes in tax regulations yield larger porate tax savings with limited, if any, operational changes.
overall effects and reinforce the basic result, as do several robustness However, markets do not react in a strongly positive fash
checks. The results suggest that the simple view of corporate tax avoid ion?and often react negatively?to U.S. firms' announcing
ance as a transfer of resources from the state to shareholders is incomplete
the agency relations.
such moves (e.g., Desai & Hines, 2002). Second, an event
given problems characterizing shareholder-manager
study of an episode of increased tax enforcement in Russia
indicates that these enforcement actions are associated with
I. Introduction
positive market reactions (Desai, Dyck, & Zingales, 2007).
tax consequences are a motivating factor in These small sample studies are provocative but leave open

WHILE many corporate decisions, managerial actions de questions about the nature of corporate tax avoidance ac
signed solely to minimize corporate tax obligations are tivity generally and in larger samples.
This paper investigates the degree towhich corporate tax
thought to be an increasingly important feature of U.S.
avoidance activity is valued by investors in a large sample
corporate activity.1 Do such activities advance shareholder
interests? If avoidance activities are costless to investors, of U.S. firms.While the traditional view of corporate tax
the question is trivial, as avoidance activity results in a avoidance suggests that shareholder value should increase
transfer of value from the state to shareholders. Indeed, this with tax avoidance activity, an agency perspective on cor
has been the presumption in the large literature on the porate tax avoidance provides a more nuanced prediction.
effects of taxes on financial decision making. Corporate tax Specifically, firmgovernance should be an important deter
avoidance activity, however, may be costly on several mar minant of the valuation of purported corporate tax savings.
from the direct costs of engaging in such While tax avoidance per se should increase the after-tax
gins. Aside
activities, managers have to ensure that these value of the firm,this effect is potentially offset, particularly
typically
actions are obscured from tax authorities. In the process, in poorly governed firms,by the increased opportunities for
such machinations may afford managers increased latitude rent diversion provided by tax shelters. Thus, the net effect
to pursue self-serving objectives. Can the latter effect be on firm value should be greater for firms with stronger

significant enough to change the simple answer that inves governance institutions.
tors fully capture the value of corporate tax avoidance The relative merits of these two views of tax avoidance
are evaluated using a data set with 4,492 observations on
activity?
Two small sample studies indicate that the valuation of 862 firms over the period 1993-2001. This panel is drawn
tax avoidance activities may not conform to the simple story from the Compustat and Execucomp databases, merged
with data on institutional ownership of firms from the
Received for publication October 24, 2005. Revision accepted for CDA/Spectrum database. Firm value is measured using
publication February 22, 2008. Tobin's q, and governance quality is proxied for by the level
* Desai:
Harvard University and NBER; Dharmapala: University of
Connecticut.
of institutional ownership, reflecting the ability of institu
We thank the editor (Daron Acemoglu), two anonymous referees, Ro tional owners to monitor managerial performance more
sanne Altshuler, Alan Auerbach, Amy Dunbar, Ray Fisman,
Michelle
Sanjay Gupta,
Jim Hines, Peter Katuscak, Mark Lang, Lillian Mills,
aggressively. Tax avoidance is measured by inferring the
Hanlon,
John Phillips, George Plesko, Dan Shaviro, Joel Slemrod, and JohnWald difference between income reported to capital markets and
for helpful discussions and comments. We also thank Sanjay Gupta and tax authorities?the book-tax gap?and controlling for ac
Jared Moore for kindly providing the data used in section III on firms cruals and other measures of earnings management. The
involved in tax shelter litigation. M.D. acknowledges the financial support
of the Division of Research of Harvard Business School, and D.D. analysis demonstrates that this measure takes on higher
acknowledges the financial support of the University of Connecticut values when a firm is involved in litigation relating to
Research Foundation.
1
See, for example, U.S. Department of the Treasury
aggressive tax sheltering activity.
(1999), Bankman
(2004), and Slemrod literature on the effects of OLS results indicate that controlling for a variety of other
(2004). The extensive
taxes on firms' behavior, as surveyed in Auerbach (2002) and Graham relevant factors including firm and year fixed effects, the
(2003), does not typically consider corporate tax avoidance. The literature effect of the tax avoidance measure on q is positive but not
on individual tax avoidance and evasion, as surveyed in Slemrod and
Yitzhaki (2002), is extensive. Despite the differences between the indi significantly different from 0. As predicted by the agency
vidual and corporate contexts stressed by Slemrod (2004), there has been
perspective on corporate tax avoidance, the effect is positive
relatively little theoretical modeling of tax compliance decisions by for those firm-years with high levels of institutional owner
corporations. Chen and Chu (2005) and Crocker and Slemrod (2005)
analyze the distinct question of the nature of the optimal incentive contract ship. The interpretation of these results, however, is com
when managers can engage in tax evasion on behalf of the firm.
plicated by the possibility ofmeasurement error in the proxy

The Review of Economics and Statistics, August 2009, 91(3): 537-546

? 2009 by the President and Fellows of Harvard College and theMassachusetts Institute of Technology

This content downloaded from 132.211.1.50 on Mon, 21 Sep 2015 13:02:52 UTC
All use subject to JSTOR Terms and Conditions
 538 THE REVIEW OF ECONOMICS AND STATISTICS

for tax avoidance and by the potential endogeneity of tax The paper proceeds as follows. Section II presents the
avoidance activity. Specifically, it is possible that firms that alternative views of corporate tax avoidance. Section III
are performing worse for exogenous reasons may be more describes the data and themeasure of corporate tax avoid
likely to engage in tax avoidance.2 Fortunately, a 1997 ance. Section IV presents the OLS results, while section V
regulatory change unintentionally and significantly changed describes the IV methodology and results. Section VI con
the costs of tax sheltering differentially across firms. This cludes.

source of exogenous variation permits the implementation

of an instrumental variables strategy that can be used to II. Theories of Corporate Tax Avoidance
address these concerns and to investigate the causal effect of
The purported growth in corporate tax avoidance activity
tax avoidance on firm value.
has given rise to two alternative perspectives on themoti
The check-the-box regulations were designed to enable
vations and effects of this activity. Several studies investi
small firms to choose their organizational form for tax
gate corporate tax avoidance as an extension of other
purposes. Altshuler and Grubert (2005) and various practi tax-favored activity, such as the use of debt. In particular,
tioners have observed that these regulations also had the
Graham and Tucker (2006) construct a sample of firms
unintended consequence of lowering the costs of tax avoid
involved in 44 corporate tax shelter cases over the period
ance for firms. Specifically, hybrid entities became increas
1975-2000. By comparing these firms with a matched
ingly common. These entities are classified as separately sample of firms not involved in such litigation, they identify
incorporated subsidiaries under the tax rules of one country characteristics (such as size and profitability) that are pos
while simultaneously being treated as unincorporated
itively associated with the use of tax shelters and argue that
branches under the tax rules of another country. This flex tax shelters serve as a substitute for interest deductions in
ibility in entity classification creates a sizable tax avoidance determining capital structure.This paper is representative of
opportunity for firms with incentives to capitalize on these the common view that corporate tax shelters are merely
regulatory changes. The central idea underlying the identi tax-saving devices without any other agency dimensions.
fication strategy is that for a given incentive to engage in tax An alternative theoretical approach emphasizes the inter
avoidance, a firmwill engage inmore actual tax avoidance action of these tax avoidance activities and the agency
after the check-the-box regulations were adopted. A crucial
problems inherent in publicly held firms.According to this
determinant of the incentives to engage in tax avoidance is view, obfuscatory tax avoidance activities can create a
the availability of tax shields. Thus, instruments for tax shield for managerial opportunism and the diversion of
avoidance are constructed by interacting a dummy variable rents. This perspective underlies several recent studies,
for the period after the check-the-box regulations with including Desai and Dharmapala (2006) and Desai et al.
variables (at the firm-year-level) that proxy for the avail (2007), and forms part of an emerging paradigm that em
ability of tax shields: net operating loss (NOL) carryfor phasizes the links between firms' governance arrangements
wards and two differentmeasures of debt. and their responses to taxes.3 In this view, corporate tax
Instrumental variables (IV) estimates using the instru avoidance not only entails distinct costs, but these costs may
ments described above lead to results that are in the same outweigh the benefits to shareholders, given the opportuni
direction as the OLS results but are considerably stronger. ties for diversion that these vehicles provide. Desai and
The interaction between institutional ownership and tax Dharmapala (2009) discuss examples of the interaction
avoidance is positive and significant, as predicted by the between tax shelters and various forms of managerial op
agency perspective on tax avoidance. This result is robust to portunism, illustrating that straightforward diversion and
the inclusion of various additional control variables and a subtle forms of earnings manipulation can be facilitated
when managers undertake tax avoidance activity.
variety of extensions to themodel. The exclusion restriction
While the traditional view of corporate tax avoidance
underlying the IV results may be invalid if the effect on firm
value of the tax shield variables changed over time for suggests that shareholder value should increase with tax
reasons unrelated to the check-the-box regulations. Reas avoidance activity, the alternative view provides a more
nuanced prediction. Specifically, firm governance should be
suringly, the basic result is robust to including interactions
between these variables and time trends in the model. an important determinant of the valuation of purported

Overall, the substantially larger effects found using the IV corporate tax savings. While the direct effect of tax avoid
results are significantly ance is to increase the after-tax value of the firm, these
approach suggest that the OLS
affected by attenuation bias due tomeasurement error in the
3For
tax avoidance proxy or by the endogeneity of tax avoidance. example, Chetty and Saez (2005) show that increases in dividend
payments in response to the 2003 dividend tax cut were most pronounced
among firms with high levels of managerial ownership, as well as those
2Unlike and Wei with high levels of institutional ownership. Managers with large stock
the study of tax evasion by Fisman (2004), where
prices of goods are observed more directly, the
measure employed here is option holdings, however, were less likely to respond to the tax change
indirect. While the validation check provides reassuring evidence, con (Brown, Liang, & Weisbenner, 2004). Each of these papers indicates that
of accruals may not sufficiently isolate tax tax incentives interact with and governance institutions in
trolling for various measures ownership
avoidance from earnings management. important ways.

This content downloaded from 132.211.1.50 on Mon, 21 Sep 2015 13:02:52 UTC
All use subject to JSTOR Terms and Conditions
 CORPORATE TAX AVOIDANCE AND FIRM VALUE 539

_Table
1.?Summary
Statistics_ correlation between the dependent variable and themeasure
Standard Number of of tax avoidance.4 While q is the primary dependent vari
Mean Deviation Observations
able used in the analysis, alternative measures of firmvalue
Tobin's q (excluding deferred tax lead to consistent results, as discussed in section V.
expense) 2.3537 2.2885 4,492 In addition to drawing on financial statement data, the
Tobin's q (including deferred tax
expense) 2.3123 2.0011 4,062 analysis here requires a measure of firm governance. The
Market value (scaled) 1.7825 2.2881 4,470 primary measure of governance used in testing the paper's
Book-tax gap (scaled) -0.0074 0.1077 4,492 main hypothesis is the fraction of the firm's shares owned
Total accruals (scaled) -0.0432 0.0787 4,492
Ratio of value of stock option grants by institutional investors (from the CDA Spectrum data
to total compensation for top five base, based on Schedule 13F filingswith the SEC by large
executives 0.4066 0.2542 4,492 institutional investors). This fraction (which is reported
Value of stock option exercises by
top five executives (scaled) 0.0061 0.0257 4,492 quarterly) is averaged over each firm-yearand is denoted by
Sales (millions of $) 3.5831 8.3074 4,492 Iit E [0, 1] for firm i in year t. The basic motivation
Volatility (Black-Scholes measure) 0.4021 0.1864 4,492
NOL carryforwards (scaled) 0.0351 0.2074 4,492
underlying this proxy is that institutional investors have
Foreign income or loss (absolute greater incentives and capacity tomonitor managerial per
value; scaled) 0.0327 0.0401 4,492 formance. Thus, the higher is Iit, the greater the degree of
R&D expenditures (scaled) 0.0475 0.0666 4,492
Long-term debt (scaled) 0.1748 0.1537 4,492 scrutiny towhich managerial actions are subjected, and the
Debt in current liabilities (scaled) 0.0387 0.0608 4,492 less important are agency problems between managers and
Deferred tax expense (scaled) 0.0256 0.0345 4,062 shareholders. In addition, a differentmeasure?the index of
Future sales growth (fraction) 0.1077 0.2984 4,328
0.0674 0.0505
anti-takeover provisions constructed by Gompers et al.
Capital expenditures (scaled) 4,433
Pretax income (scaled) 0.0544 0.1363 4,492 (2003)?is used in robustness checks. While this captures a
Governance Index, 1998 9.2650 2.8387
Institutional ownership (fraction) 0.5853 0.1768
3,906
quite different aspect of governance than does Iit (namely,
4,492
managerial entrenchment rather than the quality of moni
Note:These variablesaredefinedas inthetext.Scaled variablesaredeflated
by thecontemporaneous
book valueof assets. toring), its use leads to highly consistent results, as dis
cussed in section V.
Given the efforts undertaken to obscure such activities,
effects are potentially offset, particularly in poorly governed tax avoidance is difficult to measure. The analysis in this
firms, by the increased opportunities for managerial rent paper adopts an indirect approach, constructing a measure
diversion. Thus, the net effect on firm value should be of corporate tax avoidance that takes as its starting point the
greater for firmswith stronger governance institutions. gap between financial and taxable income. The difference
between income reported to capital markets, using Gener
III. Firm Value, Governance, and Corporate
Measuring ally Accepted Accounting Principles (GAAP) and to the tax
Tax Avoidance authorities?the so-called book-tax gap?has attracted con
siderable interest in recent years and has been related to
The data used to test the hypothesis described above are
measures of corporate tax avoidance (Manzon & Plesko,
drawn from three sources. Financial accounting data are
drawn from Standard and Poor's Compustat database, ex 2002; Desai, 2003, 2005). Given that tax returns are confi
ecutive compensation data (and certain other control vari dential, income reported to tax authorities cannot be ob
served directly and must be inferred using financial account
ables) from Standard and Poor's Execucomp database, and
data on institutional ownership of firms from the CDA/ ing data, as described inManzon and Plesko (2002) and
implemented in Desai and Dharmapala (2006). This ap
Spectrum database. Merging these variables leads to a data
proach uses firms' reported current federal tax expense and
set with 4,492 observations at the firm-year level on 862
"grosses up" this tax liability by the U.S. federal corporate
firms over the period 1993-2001. The variables are de
tax rate.5 For firmswith positive current federal tax expense,
scribed in detail below; summary statistics are reported in
table 1.
4
the Compustat data item numbers, qit for firm i in year t is
In emphasizing the value implications of corporate tax Using
defined as follows:
avoidance, this paper builds on the extensive literature in
corporate finance on the determinants of firm value. Within
qit_(#6)<f+((#24)*(#25)ft)-(#60)ft
this literature, it has become standard since Demsetz and (#6),
Lehn (1985) to use Tobin's q to measure firm value. The Note, however, that using the standard definition (Kaplan &
Zingales,
definition of q used in Kaplan and Zingales (1997) and 1997) that includes deferred taxes leads to consistent results, as described
in section V.
Gompers, Ishii, and Metrick (2003) is employed in the 5
The exclusion of foreign taxes and income from this calculation avoids
analysis below, with one modification: deferred tax expense problems associated with inferring the applicable foreign tax rates. How
is not included in the definition of q used in the basic results ever, foreign activities can affect U.S. tax liability under the worldwide
system of taxation that applies to U.S. corporations. Thus, the regression
below, as current tax avoidance activity may result in analysis includes a control variable that proxies for foreign activity (the
changes to future tax liabilities and thus create a mechanical absolute value of foreign income or loss, as described below).

This content downloaded from 132.211.1.50 on Mon, 21 Sep 2015 13:02:52 UTC
All use subject to JSTOR Terms and Conditions
 540 THE REVIEW OF ECONOMICS AND STATISTICS

the graduated structure of corporate tax rates is used in this cruals proxy isolates the component of the gap that is due to
calculation. For firms with negative current federal tax tax avoidance. In the regressions reported below, BTit is
expense, the top statutory rate of 35% is used. used as a proxy for tax avoidance activity, while earnings
Given this inferred value of the firm's taxable income, the management is controlled for by including a measure of
book-tax gap can be estimated by simply subtracting in total accruals (denoted TAit for firm / in year t) as a control
ferred taxable income from the firm's reported pretax (do variable.9

mestic U.S.) financial income.6 To control for differences in Given the confidentiality of tax returns, this procedure
firm scale and because the dependent variable is deflated by yields the best measure of corporate tax avoidance that can
the book value of assets, the inferred book-tax gap is also be obtained using publicly available data. However, in view
scaled by the book value of assets. This yields themeasure of the limitations associated with inferring taxable income,
and as there are alternative explanations for book-tax gaps,
of the book-tax gap used in the analysis below (denoted BTit
it is useful to implement a validation check of the book-tax
for firm / in year t).1
The book-tax gap does not necessarily reflect corporate gap as a measure of corporate tax sheltering activity before
proceeding to determine its valuation effects.
tax avoidance activity, so any measure of tax avoidance
Graham and Tucker (2006) construct a sample of firms
must control for other factors. In particular, the overreport
involved in 44 cases of tax shelter litigation over the period
ing of financial income (known in the accounting literature
1975-2000, using publicly available court records and press
as earnings management) may contribute to the measured
articles. The validation check undertaken here uses a data
book-tax gap.8 Studies of earnings management (e.g., Healy,
set compiled using a similar methodology. This information
1985) have argued that such manipulation ismost likely to can be used to construct a variable that indicates whether tax
occur through the exercise of managerial discretion in
sheltering activity was alleged in any given firm-year. Spe
determining accounting accruals (i.e., adjustments to real
cifically, let the indicator variable Lit be equal to 1 if firm /
ized cash flows that are used in calculating the firm's net was alleged to have used a tax shelter in year r, and 0
income). The basic intuition underlying themeasure of tax otherwise. This variable is merged with data on the
avoidance used here is thatbook-tax gaps are attributable to book-tax gap and a set of control variables from the
either earnings management or tax avoidance activity. Ac data set in order to ex
merged Compustat-Execucomp
an ac
cordingly, adjusting for earnings management with amine the relationship between involvement in tax shelter
litigation and book-tax gaps. The regression specification
6One A
particular concern with this approach may be the following.
at the
(for the case of a linear probability model) is
firm's taxable income is reduced by the value of the compensation
time that employees exercise stock options. However, under the applicable
= + $2TAit + X,yy + e, + vit9
accounting rules, the reported tax expense is unaffected. It should be Lit ^BTit (1)
not reduced by
remembered, though, that reported financial income is
stock option exercises either. Thus, the exclusion of stock
employees' where et is a year fixed effect and vit is the error term.Xit
income does not bias
option exercises from both tax expense and financial
the measure of the book-tax gap (see Manzon & Plesko, 2002, or Desai is a vector of control variables that includes measures of
and Dharmapala, 2006, for more details). Nonetheless, some concerns firm size (assets, sales, and market value) and the structure
tax
may remain that the valuation of tax avoidance may be affected by the of executive compensation.
shield available to a firm from stock option exercises. This issue is
addressed by including the value of observed stock option exercises as a Estimating this relationship using a logitmodel results in
control in robustness checks. a positive coefficient on BTit, as reported in column 1 of
7 data item numbers, BTit, for firm i in year t is
Using the Compustat
table 2: the book-tax gap tends to be larger, other things
defined as follows:
equal, in firm-years in which tax sheltering is alleged.10
(#63), Probit and linear probability models also yield positive
(#272), t coefficients. Any conclusions from this validation check are
BTit= '
(#6h,
9 Some of TAit may be positive even in the absence of
where t is the U.S. federal corporate tax rate. As defined here, BTit component
of the gover Several alternative measures of "abnormal" or
includes elements that are innocuous from the perspective earnings management.
nance issues analyzed in this paper, such as book-tax differences arising "discretionary" accruals (e.g., Jones, 1991; Dechow, Sloan, and Sweeney,
Richardson, & Tuna, 2003) have been developed to better
from the treatment of depreciation or from the investment tax credit. The 1995; Dechow,
isolate the components of accruals that are truly under managerial control.
extent to which BTit can be corrected for these factors is limited by the
of information (e.g., on tax depreciation) in Compustat. these alternative proxies for earnings management instead of TAit
unavailability Using
and investment tax credits leads to very similar results. An alternative approach to calculating
However, controlling for depreciation expense
in robustness checks leads to consistent results (see section V). Note also accruals (Hribar & Collins, 2002) uses information on cash flows; this
or pension expense) also gives consistent results. The results are also robust to
that tax deductions (such as those for interest expense approach
measure of earnings management,
that are treated symmetrically for book and tax purposes (i.e., also including a control for another pro
deducted from financial income) do not mechanically affect BTit. posed by Phillips, Pincus, and Rego (2003), namely deferred tax expense
8
Lev and Nissim (2004) and Hanlon (2005) investigate how book-tax (see section V).
10Note that firm fixed effects cannot be used in table 2, because the
gaps predict the quality of future earnings, essentially interpreting the
entire book-tax gap as being due to earnings management activity. They fixed-effects logit model does not use for identification any observations
the consequences of these managed earnings for subsequent for firms for which Lit is always 0 or always 1. Thus, given that the vast
analyze
to use shelters, the remaining sample
accounting and market outcomes, but do
not focus on the contemporane majority of firms are never alleged
in a fixed-effects model would be extremely small.
ous valuation of the tax avoidance component of the book-tax gap.

This content downloaded from 132.211.1.50 on Mon, 21 Sep 2015 13:02:52 UTC
All use subject to JSTOR Terms and Conditions
 CORPORATE TAX AVOIDANCE AND FIRM VALUE 541

and Tax
(2006). To control for changes over time in the risk associ
Table 2.?Book-Tax Gaps Shelter Litigation

Indicator (=1) for Alleged ated with a firm's stock price, a measure of volatility is also
Variable Tax Shelter Activity
Dependent included.13 As NOL carryforwards are not taken into ac
Book-tax gap (scaled) 3.7185* count in themeasure of tax avoidance (and because NOLs
(1.9377) can affect the incentives to engage in tax avoidance), NOL
Accruals measure (scaled) ?5.6702***
(0.9005) carryforwards scaled by assets (with missing values treated
Controls for changes in sales, assets, market Y as zeroes) are also included.
value, and the ratio of the value of stock The tax avoidance measure is restricted to domestic U.S.
option grants to total compensation for tax expense and U.S. federal taxes, but tax liabilities and the
top 5 executives?
Number of firms 1,018 incentives for tax avoidance may be influenced by foreign
Number of observations
0.0671
4,985
activity under theU.S. system of worldwide taxation. Thus,
Pseudo-tf2
a proxy for foreign activity?the absolute value of foreign
Note: The dependentvariableis an indicatorvariable= 1 inanyfirm-year inwhich thefirm was
subsequentlyalleged tobe engaginginaggressivetax sheltering.The sampleincludesall firmsin the income or loss?is included inXff. As tax shields can affect
mergedCompustatandExecucompdatabasesforwhichtherequired use
dataareavailable.The estimates
a logitmodel.The sampleperiod is 1993-2001.The specifications includeyeareffectsand thesetof
the value of engaging in tax avoidance, changes in firms'
leverage are controlled for by including measures of long
controlsspecified.
Robuststandard errorsthatareclusteredat thefirmlevelarepresentedinparentheses;
* and *** denote at the 10% and 1% levels,respectively.
significance
term debt and debt in current liabilities. Changes in intan
gibles that affect q but are imperfectlymeasured in the book
necessarily tentative, given the small number of firms that value of assets are proxied for by research and development
have been involved in tax shelter litigation. Nonetheless, it
(R&D) expenditures. A number of additional control vari
appears that themeasure of tax avoidance employed below ables are used in robustness checks, as described below.
captures a critical element of tax sheltering activity, as it Note also thatbecause firm fixed effects are employed in the
takes on higher values for those firm-years for which there
specification described below, many of the sources of cross
is some independent evidence for alleged tax shelter activ sectional variation in q across firms that have been dis
ity. cussed in the literature are effectively controlled for here.
The specification used to test whether the valuation of
IV. OLS Approach and Results
corporate tax avoidance is dependent on firm governance
extends equation (2) as follows:
While the central hypothesis of the paper concerns the
interaction of governance institutions and tax avoidance
=
qit $xBTit + ?2TAit + fo/,, + Ml%BTit) + Xity
activity, the question of whether tax avoidance tends to be
associated with increases or decreases in firm value is also
+
(3)
of considerable interest. This is addressed using the follow p.,- + e, + vit,

ing specification:
where Iit is themeasure of institutional ownership defined
= + $2TAit + Xity + ^ + e, + vit, above. The hypothesis in section II implies that p$4> 0: the
qit ^BTit (2)
effect of tax avoidance on q is greater in firm-years inwhich
where the variables BTit and TAit are as defined above, |X/ institutional
ownership is higher (and governance is stron
and are firm and year fixed effects, respectively, and vit is
ger).
the error term (note that all regressions reported in this The results using OLS estimation on equations (2) and (3)
paper use both firm and year fixed effects). are reported in table 3; note that all results reported in this
Xit is a vector consisting of the following control vari paper use robust (White, 1980) standard errors clustered at
ables. Changes in firm size over time are controlled for the firm level. Column 1 presents the results from the
using sales.11 The value of stock option grants to executives estimation of equation (2).14 The overall effect on firm value
as a fraction of total compensation12 is included because a
of the proxy for tax avoidance is positive but insignificant.
substantial literature (e.g., Morck, Shleifer, & Vishny, 1988; The test of the hypothesis using equation (3) is reported in
Mehran, 1995) finds stock-based compensation to be a column 2. Here, the coefficient on the interaction term
determinant of firm value, presumably through incentive
(I*BTit)?p4 in equation (3)?is positive, consistent with
alignment effects. In addition, the structure of executive the paper's hypothesis, and is of borderline statistical sig
compensation plays a central role inDesai and Dharmapala nificance. The intuition can be reinforced by running equa
tion (2) separately for firm-years with high and low levels of
11
Assets and market value enter into the definition of q and so would be
institutional ownership (columns 3 and 4, respectively),
mechanically correlated with the variable.
dependent
12
This is calculated from data at the manager-year level in the Execu
13
comp database and is defined as the ratio of the Black-Scholes value of Using a firm's beta (calculated
using CRSP monthly data for the
stock option grants to total compensation (i.e., the sum of the value of preceding five years) instead of the volatility measure has no effect on the
stock options, salary, and bonus). This is similar to the stock-based results.
14The
compensation measures used inMehran (1995) and in a large subsequent sample is restricted to firm-years for which the CDA/Spectrum
literature. Adding a measure based on stock option exercises data on institutional ownership are available
(defined (although that variable is not
analogously) does not affect the results. used in equation (2)), for comparability with the other columns of table 3.

This content downloaded from 132.211.1.50 on Mon, 21 Sep 2015 13:02:52 UTC
All use subject to JSTOR Terms and Conditions
 542 THE REVIEW OF ECONOMICS AND STATISTICS

Table 3.?Tax Avoidance, Firm Value, and Governance Institutions: OLS Results

Tobin's q

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

Firm-Years with High Firm-Years with Low
Dependent Variable All Firms All Firms Institutional Ownership Institutional Ownership

Book-taxgap (scaled) 0.5776 -2.1655 2.7624* 0.2718

(0.5590) (1.4957) (1.5394) (0.3678)
Book-tax gap interacted with institutional 5.6687*
ownership (3.3307)
Institutional ownership 0.7706**
(0.3287)
Total accruals(scaled) 1.3267** 1.2689** 0.5313 1.1159*
(0.3811) (0.3613) (0.5676) (0.5892)
Ratio of value of stock option grants to total 0.4391** 0.4371** 0.5627** 0.2253
fortop5 executives
compensation (0.1208) (0.1202) (0.2004) (0.1745)
Sales 0.0442* 0.0471* 0.0195 0.0592
(0.0249) (0.0250) (0.0158) (0.0404)
Volatility -2.114006** -1.9465** -3.8771** -1.0303*
(0.6682) (0.6466) (1.2553) (0.5697)
Controls for tax shields (NOLs, long-term
debt,and current
debt) Y Y YY
Controls for foreign income/loss and R&D Y Y Y Y
Year andfirmeffects? Y Y YY
Numberof firms 862 862 583 614
Numberof observations 4,492 4,492 2,324 2,168
R2 0.6483 0.6500 0.7765 0.6213
Note: The dependent variableisTobin'sq, as definedinsectionIII.The sample(overtheperiod 1993-2001) isdrawnfromthemergedCompustatandExecucompdatabasesand is restricted tofirm-yearsfor
whichCDA/Spectrum dataon institutional
ownership areavailable.All specifications
includeyeareffects, and thecontrolslisted.Incolumn3, thesampleis restricted
fixedeffects,
firm tofirm-years
with institutional
> 0.6. In column4, thesampleis restrictedtofirm-years with institutional < * **
errorsthatare clusteredat thefirmlevelare presentedinparentheses; and denote
ownership ownership 0.6. Robust standard
at the 10% and 5% levels,respectively.
significance

where "high" institutional ownership is defined as being a in the proxy for tax avoidance, particularly if the extent of
fraction that exceeds 0.6, which is approximately themean measurement error differs by governance institutions. For
of the sample. For well-governed firm-years, the effect of example, if the proxies used for earnings management are
tax avoidance on q is positive and of borderline signifi
incomplete, the remaining component of the book-tax gap
cance. For less well-governed firm-years (with institutional as tax avoidance when it actually
may be mischaracterized
ownership below 0.6), the effect is also positive but statis represents earnings management. Accordingly, it is possible
tically insignificant and considerably smaller inmagnitude. that the results are driven by differential market reactions to
Thus, while the estimated overall effect of tax avoidance on earnings management by well-governed and poorly gov
firm value is indistinguishable from 0, the effect appears to erned firms. The second is the potential endogeneity of tax
be more positive for well-governed firm-years than for avoidance activity. For example, firms that are performing
poorly governed firm-years. This finding is consistent with worse for other reasons may be more likely to engage in tax
the hypothesis that agency problems mitigate the benefits to
avoidance.
shareholders of corporate tax avoidance.
In order to address these concerns, an exogenous source
of variation in firms' opportunities for tax avoidance is
V. Instrumental Variables Approach and Results
required. Fortunately, a 1997 regulatory change with unre
A. Instrumental Variables lated objectives lowered the costs of tax avoidance for a
subset of firms. In late 1996, the Treasury issued what are
OLS estimation of equations (2) and (3) gives rise to two
known as the check-the-box (CTB) regulations. These reg
types of potential problems.15 The first ismeasurement error ulations enable firms to choose form for
their organizational

15Another alternative
tax purposes?for example, whether to be taxed as a C
explanation for the paper's findings (that is not
entirely addressed by the IV approach) is that the differences in the corporation or as a pass-through entity such as a partnership
valuation of tax avoidance between the well-governed and poorly gov or sole proprietorship?by filing a one-page form on which
erned subsamples relate to differences in the types of tax shelters used by
these firms. It is possible that the smaller effect for poorly governed firms
they simply check the appropriate box. In replacing a
is due to these firms' investing in riskier shelters that are discounted at complex set of rules by which the IRS determined firms'
higher rates or in shelters with shorter expected lives. If this alternative tax status, the CTB regulations were intended to reduce
is true, the tax avoidance measure would be expected to
explanation the administrative burdens for small firms. Researchers
exhibit lower autocorrelation for poorly governed firms relative to well
governed firms. However, computing simple autocorrelation coefficients
for BTi t in the two subsamples of firm-years shows that the degree of shelters with shorter expected lives. While this is by no means a definitive
autocorrelation is actually larger (more positive) for less well-governed test, the available evidence does not appear to support the alternative
firm-years. This is not consistent with poorly governed firms' using tax explanation.

This content downloaded from 132.211.1.50 on Mon, 21 Sep 2015 13:02:52 UTC
All use subject to JSTOR Terms and Conditions
 CORPORATE TAX AVOIDANCE AND FIRM VALUE 543

Table 4.?Book-Tax Tax and the Check-the-Box

studying international taxation argue that the CTB regula Gaps, Shields,
First-Stage IV
tions also had the unintended consequence of facilitating tax _Regulations: Results_
Book-Tax
avoidance by large U.S.-based multinational firms through Gap

the use of what are known as hybrid entities (see in partic Dependent Variable (2) (1)
ular Altshuler & Grubert, 2005). Hybrid entities are classi PostCTB X -0.0349** -0.0162
(NOL carryforwards)
fied in two distinct ways: as separately incorporated subsid (0.0139) (0.0356)
PostCTB X (long-term debt) -0.0127 -0.0474
iaries under the tax rules of one country and as
(0.0160) (0.0384)
unincorporated branches under the tax rules of another PostCTB X (current debt) -0.0879* -0.3103**
country.16 (0.0507) (0.1197)
PostCTB X (NOL carryforwards) X -0.0579
The instruments for tax avoidance involve interacting a
(institutional ownership) (0.1341)
dummy variable for the post-CTB time period (the years PostCTB X (long-term debt) X 0.0548
since 1997) with firm-year-level variables that capture the (institutional ownership) (0.0570)
PostCTB X (current debt) X 0.4149**
incentive to engage in tax avoidance. The central idea
(institutional ownership) (0.1644)
underlying the identification strategy is that for a given F-statistic for joint significance of 3.33** 3.00***
incentive to engage in tax avoidance, a firmwill engage in the instruments (p-value) (0.0189) (0.0063)
more actual tax avoidance after the CTB regulations were Controls for total accruals, executive
compensation, sales, volatility,
adopted than it would have before, other things equal. A foreign income/loss, and R&D Y Y
crucial determinant of the incentives to engage in tax Controls for tax shields (NOLs, long
term debt, and current debt) Y Y
avoidance is the availability of tax shields (tax deductions
Control for institutional ownership N Y
from other sources, such as interest deductions or NOL Year and firm effects? Y Y

carryforwards resulting from losses in previous years). For Number of firms 862 862
Number of observations
instance, Graham and Tucker (2006) emphasize the substi 4,492
R20.5993 0.6009
tutability of tax shelters and other kinds of tax shields. Note: The dependentvariableis thebook-taxgap, as definedin thetext."PostCTB" isan indicator
Instruments for tax avoidance can thus be constructed by variableforyearsafterthecheck-the-box were introduced
regulations (1997-2001).The sample(overthe
period 1993-2001)isdrawnfromthemergedCompustatandExecucompdatabasesand is restricted to
interacting a dummy variable for the period after the CTB firm-yearsforwhichCD A/Spectrum data on institutional
ownershipare available.All specifications
includeyeareffects,firmfixedeffects,
and thecontrolslisted.
Robust standarderrors thatare clustered
regulations with each of the following variables: NOL *, **, and ***denotesignificance
at thefirmlevelarepresentedinparentheses. at the10%,5%, and 1%
carryforwards and two different measures of debt (long levels,respectively.
term debt and debt in current liabilities).
The IV approach involves instrumenting for the endoge
nous variable BTit in equation (2) using as the set of In equation (3), there are effectively two endogenous
instruments the variables listed above, each interacted with variables?BTit and (I*BTit)?and the set of instruments
a dummy variable for the post-CTB period. Let Pt be the thus includes interactions with Iit. In particular, the instru
dummy for the post-CTB period, NOLit be theNOL carry ments for BTit and (I*BTit) are the following: (P$V0Li7),
forwards for firm / in year t,DLit be long-term debt for firm (P*DLit), (P*DC/r), (I%P*NOLit), (/f,P*DLl7), and
/ in year t, and DCit be debt in current liabilities for firm / (I*P*DCit). The first-stage results (presented in column 2
in year t. Then the instruments for BTit are {P^NOLit), of table 4) show the expected negative relationship between
(P*DLit), and (P?DC/f). The first-stage regression (reported each of the firstthree of these instruments and BTit\ the full
in column 1 of table 4) shows that these instruments are set of instruments is also strongly jointly significant.17
indeed strongly related toBTit. Specifically, the coefficients The basic rationale for this IV approach is that a given
are negative, as expected: lower values of tax shields (which incentive to engage in tax avoidance should lead to more
imply a greater incentive to engage in tax avoidance) are actual tax avoidance after the CTB regulations than before.
associated with larger values of BTit after the CTB regula The crucial exclusion restriction underlying the use of these
tions than before, controlling for other factors. The instru instruments is the following. The underlying tax shield
ments are jointly significant at the 5% level. variables (NOLs and the debt measures) used in construct
ing the instruments may directly affect firm value; this
16
The following is a simple example of how these entities can be used direct effect is controlled for by including the tax shield
to reduce tax liabilities. Suppose that a U.S.-based multinational (A) sets variables in the specification. However, the tax shield vari
up a tax haven subsidiary (B) that provides loans to another subsidiary (C)
in a high-tax foreign country. The interest on these loans is tax deductible ables should not affect firmvalue differently after the CTB
regulations other than through their influence on incentives
in the high-tax foreign country, to the government of which B is
reported
to be a separately incorporated
entity from C. Prior to 1997, the interest for tax avoidance. This restriction is conditional on the
received by B (while untaxed by its tax haven location) would have been
subject to immediate U.S. taxation under the Subpart F rules relating to controls included in themodel. For example, even if firm
interest payments from one Controlled Foreign Corporation (CFC) to valuations were in general higher in the late 1990s, the year
another. However, the CTB regulations made it possible forA to elect (for
dummies included in the specification would account for
U.S. tax purposes) to have B treated as an unincorporated branch of C.
This makes the interest payments received by B "invisible" to theU.S. tax
17
system and so facilitates the avoidance (or at least deferral until repatri The instruments are also of borderline significance in the first-stage
ation) of U.S. tax on the interest income paid by C to B. for {I*BTit).
regression

This content downloaded from 132.211.1.50 on Mon, 21 Sep 2015 13:02:52 UTC
All use subject to JSTOR Terms and Conditions
 544 THE REVIEW OF ECONOMICS AND STATISTICS

Table 5.?Tax Avoidance, Firm Value, and Governance Institutions: Second-Stage IV Results

Market Value
Tobin's q Tobin's q (scaled) Tobin's q

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

Book-taxgap (scaled) 14.523 -5.8710 -6.9313 -4.2465
(12.288) (5.1474) (4.9001) (4.0640)
Book-tax gap interacted with institutional ownership 32.8204** 31.4461** 21.4885**
(13.0267) (12.8540) (10.8603)
Institutional ownership 1.0331** 1.0593** 0.9614**
(0.4376) (0.4250) (0.3976)
Total accruals (scaled) -2.8305 -1.3588 -0.4447 -0.2585
(3.6467) (2.3935) (2.3651) (1.8085)
Ratio of value of stock option grants to total 0.3495 0.4839** 0.5532** 0.4732***
fortop5 executives
compensation (0.3092) (0.2142) (0.1919) (0.1609)
Sales 0.0434 0.0473* 0.0581 0.0418
(0.0276) (0.0264) (0.0245) (0.0255)
Volatility -1.0221 -1.2096 -1.2202 -1.5364**
(0.9795) (0.7703) (0.7869) (0.7227)
Controls for tax shields (NOLs, long-term debt, and
currentdebt) Y Y Y Y
Controlsforforeignincome/loss
andR&D Y Y YY
Interactions between tax shield variables and time trends N N NY
Year and firmeffects? Y Y Y Y
Numberof firms 862 862 862 862
Numberof observations 4,492 4,492 4,470 4,492
Note: The dependentvariableincolumns1,2, and4 isTobin'sq, as definedinsectioniii.The dependent variableincolumn3 ismarketvalue (scaledby thebook valueof assets),as definedin section
v. The
sample(overtheperiod 1993-2001) is drawnfromthemergedCompustatandExecucompdatabasesand is restricted tofirm-yearsforwhichCDA/Spectrum ownershipare available.All
data on institutional
includeyeareffects,
specifications firmfixedeffects, and thecontrolslisted.
The book-taxgap variable and the between
interaction thebook-taxgap and institutional are
ownership instrumented, as describedin
thetext.Robust standard *, **, and *** denotesignificance
errorsthatare clusteredat thefirmlevelare presentedinparentheses. at the 10%,5%, and 1% levels,respectively.

this. Of course, the validity of the exclusion restriction tivitywill continue in the future.With reasonable discount
depends on there being no other changes over time in the rates used, the estimated coefficient (2.76) for the well
effect of the tax shield variables on firm value. To test for governed sample in the OLS specification (column 3 of
possible violations of this exclusion restriction, interactions table 3) would correspond to an expected life of tax savings
between the tax shield variables and time trends are in of seven to nine years for well-governed firms. However,
cluded in themodel as a robustness check. interpreting this OLS coefficient in thismanner is compli
cated by the identification concerns discussed above and the
B. Instrument Variables Results
marginal significance of the coefficient in table 3.
The IV estimate in column 3 of table 5 can be used to
The second-stage results from the IV analysis are pre
overcome these difficulties. For a firmwith a mean value of
sented in table 5. Column 1 reports the results from esti
institutional ownership, these coefficients imply that the
mating equation (2), using the instruments for BTit de
scribed above. The overall estimated effect of tax avoidance
market interprets an increase in the book-tax gap as a
quasi-permanent change in tax obligations. As such,
on firm value is substantially larger than in theOLS results
changes in the book-tax gap are not interpreted as transitory
in table 3. Column 2 reports the results from estimating
the instruments for BTit and (I*tBTit) items associated with particular shelters but as signals of
equation (3), using
described above. The coefficient (34of the interaction be general tax planning ability.18The larger effects using the IV
tween governance and tax avoidance is positive and highly approach suggest thatmeasurement error in the tax avoid
ance proxy may lead to attenuation bias in the OLS esti
significant, consistent with the paper's main hypothesis. The
IV results thus support the notion that the benefits to mates. Alternatively, the OLS estimate of the effect of tax
shareholders from corporate tax avoidance depend on firms' avoidance on firm value may be biased toward zero by the
form of endogeneity noted above, where firms that are
governance institutions.
more likely to
The results in table 5 are in the same direction as theOLS performing poorly for other reasons are
results in table 3 but are considerably stronger. The coeffi engage in tax avoidance.19
cients from tables 3 and 5 can be interpreted as reflecting an
18In the
expected duration of a particular tax shelter or the ability to
data set on corporate tax shelter cases constructed by Graham
and Tucker (2006, table 1), the active life of alleged tax shelters ranges up
engage in tax planning for a given period. Suppose a firm to 10 years. The active life in theGraham-Tucker sample refers only to the
unexpectedly increases its book-tax gap by $1.00. The longevity of specific tax sheltering strategies. As such, the
IV estimates

current-year tax benefit (including federal and state tax imply that current tax avoidance activity signals general tax planning

ability, which may be expected to persist even beyond the life of any
benefits) would be approximately $0.40. Market reactions
particular strategy.
are given by the coefficient on the book-tax gap and should 19 is
It that the apparent effects of tax avoidance are in fact
possible
incorporate an expectation of how long tax sheltering ac attributable to managerial incentives (noting that Desai & Dharmapala,

This content downloaded from 132.211.1.50 on Mon, 21 Sep 2015 13:02:52 UTC
All use subject to JSTOR Terms and Conditions
 CORPORATE TAX AVOIDANCE AND FIRM VALUE 545

C. Robustness of the IV Results While the baseline specification in table 5 includes an

extensive set of controls, it is possible that unobserved
TheseIV results appear to be robust to concerns regard
changes in firms' investment opportunities or expected fu
ing themeasurement of the book-tax gap, governance, and tureperformance may affect q. To address these concerns, it
firmvalue. For instance, the basic result in column 2 of table is possible to include capital expenditures and future reve
5 is robust to the (unreported) inclusion of additional vari nue growth as additional controls; including these controls
ables?specifically, deferred tax expense,20 depreciation ex leads to consistent results. Furthermore, although Tobin's q
pense, investment tax credits, interest expense, pension is a standard measure of firm value in the literature, it is
expense, and a proxy for employees' stock option exercis nonetheless important to consider alternative proxies. As q
es21?that control for the potential mismeasurement of the takes account of the book as well as market value of equity
book-tax gap. It is also robust to adding lagged tax avoid and the value of debt, a simpler measure is themarket value
ance activity to themodel; this does not change the effect of of common stock (Execucomp variable MKTVAL, the clos
contemporaneous tax avoidance (interacted with Iit), and ing share price for the fiscal year multiplied by the number
the effect of the lagged variable is small and insignificant. of common shares outstanding). This is scaled by the book
Thus, there is no evidence to suggest a substantial delayed value of assets in order to conform to the scaling of the
market reaction to firms' tax avoidance activity. independent variables. As shown in column 3 of table 5,
The results are also robust to using alternative measures using this variable instead of q leads to essentially identical
of governance. Specifically, the findings are unaffected results.

when Iit is replaced by the index of anti-takeover provisions Finally, the identification strategy used depends on the
constructed by Gompers et al. (2003). This index represents validity of the exclusion restrictions. In particular, it re
a count of anti-takeover provisions that apply to a firm quires that there are no changes over time in the effect on
firm value of the tax shield variables used in constructing
(through either its corporate charter or state law).22 It takes
the instruments (other than the change due to the impact of
on values up to 18, with lower values indicating better
the CTB regulations on tax avoidance activity). This as
governance. As the cardinal properties of this index are
sumption may be violated if there are trends unrelated to the
unclear, the robustness check involves constructing an in
CTB regulations in the effect of the tax shield variables on
dicator variable for better-governed firms by dividing the
q. This possibility can be tested for by adding to themodel
sample at themean (with values of 9 or lower corresponding interactions between a time trend and each of the tax shield
to "well-governed" firms). The interaction between this variables. Specifically, these additional control variables are
indicator variable forwell-governed firms and BTit is very - - -
(NOL%(t 1997)), (DL%(t 1997)), and (DCt(t
similar inmagnitude and significance to that in column 2 of where t is the year (1997, the midpoint of the
1997)),
table 5. This suggests that the results are robust to alterna is used as the base year). The second-stage
sample period,
tive notions of governance, as the Gompers et al. (2003) IV results with the addition of these controls are presented
index measures managerial entrenchment rather than the in column 4 of table 5.While the coefficient of the interac
quality of monitoring. Moreover, this also indicates that the tion term of interest is somewhat smaller, it remains signif
results are unaffected by the potential endogeneity of Iit, icant at the 5% level. Thus, itdoes not appear that the effect
where institutional investors may choose to buy firms that of the instrumental variables is driven by time trends in the
are expected to increase in value. This is less applicable to impact of the tax shield variables on firmvalue. Rather, the
theGompers et al. index, as its values were predominantly results seem to depend on only the discontinuous change in
determined in the 1980s and generally do not change during the effect of tax shields on firm value associated with the
the sample period. CTB regulations.

2006, find a relationship between managerial incentives and tax avoidance VI. Conclusion
activity). However, adding an interaction term between the executive
compensation measure and tax avoidance to the model leaves the results
essentially unchanged.
Although there is an extensive literature on how firms
respond to taxes, little analysis of activities has been de
20
Recall that deferred tax expense was omitted from the computation of
q, while taxable income was inferred using only current tax expense. This signed solely or primarily to reduce tax liabilities. This
omission could be important because current tax sheltering activity may
take the form of deferring tax liabilities to the future. Also, a focus on paper contributes to the emerging literature on this topic by
current tax avoidance ignores current actions by the firm that reduce its investigating whether such activities advance shareholder
future tax liabilities, and hence increase the present value of the firm. Note
interests, using evidence on how markets capitalize these
that adding deferred tax expense to the definition of q, as in and
Kaplan activities. The simple presumption that corporate tax avoid
Zingales (1997), leads to results that are highly consistent with those in
table 5. ance represents a transfer of value from the state to share
21
The value of stock option exercises by the top five executives (scaled holders does not appear to be validated in the data. Rather,
by the book value of assets) is used as a proxy for these deductions. On the patterns in the data are more consistent with the agency
the importance of stock option deductions for certain firms, see Graham,
Lang, and Shackelford (2004). perspective on corporate tax avoidance, which emphasizes
22 See
Gompers et al. (2003, appendix 1) for more details. the mediating role of governance. The basic result that

This content downloaded from 132.211.1.50 on Mon, 21 Sep 2015 13:02:52 UTC
All use subject to JSTOR Terms and Conditions
 546 THE REVIEW OF ECONOMICS AND STATISTICS

-
higher-quality firm governance leads to a larger effect of tax
"The Degradation of Reported Corporate Profits," Journal of
Economic Perspectives 19 (2005), 171-192.
avoidance on firm value is reinforced by using an exoge Mihir A., Alexander
Desai, Dyck, and Luigi Zingales, "Theft and Taxa
nous source of variation due to changes in tax regulations to tion," Journal of Financial Economics 84 (2007), 591-623.
construct instrumental variables for tax avoidance activity. Desai, Mihir A., and Dhammika Dharmapala, "Corporate Tax Avoidance
and High Powered Incentives," Journal of Financial Economics 79
The results are robust to a wide variety of tests for alterna 145-179.
(2006),
tive explanations. -
"Earnings Management, Corporate Tax Shelters, and Book-Tax
The findings of this paper shed new light on what Weis Alignment," National Tax Journal 62 (2009), 169-186.
Desai, Mihir A., and James R. Hines Jr., "Expectations and Expatriations:
bach (2002) terms the "undersheltering puzzle": why firms the Causes and Consequences of Corporate
Tracing Inversions,"
do not engage in sheltering activity more extensively, given National Tax Journal 55 (2002), 409-441.
"Tax Rates and Tax Evasion:
availability of shelters and the low risk of
the widespread Fisman, Raymond, and Shang-Jin Wei,
Evidence from 'Missing Imports' in China," Journal of Political
penalties. Undersheltering may not be as puzzling as it first Economy 112 (2004), A1X-A96.
appears, given that investors doubt the value of such activ Gompers, Paul A., Joy Ishii, and Andrew Metrick, "Corporate Governance
ities in the absence of good governance. More generally, the and Equity Prices," Quarterly Journal of Economics 118 (2003),
107-155.
result that the valuation of tax avoidance is a function of John R., "Taxes and Corporate Finance: A Review," Review of
Graham,
firm governance suggests that tax avoidance and managerial Financial Studies 16 (2003), 1074-1128.
efforts to divert value from shareholders may be inter Graham, John R., Mark H. Lang, and Douglas A. Shackelford, "Employee
Stock Options, Corporate Taxes, and Debt Policy," Journal of
twined. This paper thus shows that incorporating agency Finance 59 (2004), 1585-1618.
issues into the analysis of corporate tax avoidance leads to Graham, John R., and Alan Tucker, "Tax Shelters and Corporate Debt
theoretical and empirical conclusions that are substantially Policy," Journal of Financial Economics 81 (2006), 563-594.
Hanlon, Michelle, "The Persistence and Pricing of Earnings, Accruals, and
different from those that would be predicted by a model Firms Have
Cash Flows When Large Book-Tax Differences,"
where managers are perfect agents. These findings open up Accounting Review 80 (2005), 137-166.
Healy, Paul, "The Effect of Bonus Schemes on Accounting
several lines of inquiry, including the implications of this Decisions,"
Journal of Accounting and Economics 1 (1985), 85-107.
agency perspective for the analysis of tax policy, which we Hribar, Paul, and Daniel W. Collins, "Errors in Estimating Accruals:
leave for future research. Implications for Empirical Research," Journal of Accounting Re
search 40 (2002), 105-134.
Jones, Jennifer, "Earnings Management During Import Relief Investiga
tions," Journal of Accounting Research 29 (1991), 193-228.
REFERENCES
Kaplan, Steven N., and Luigi Zingales, "Do Investment-Cash Flow
Altshuler, Rosanne, and Harry Grubert, "The Three Parties in the Race to Sensitivities Provide Useful Measures of Financing Constraints?"
the Bottom: Host Governments, Home Governments, and Multi Quarterly Journal of Economics 112 (1997), 169-216.
national Companies," Florida Tax Review 1 (2005), 137-209. Lev, Baruch, and Doron Nissim, "Taxable Income, Future Earnings, and
Auerbach, Alan J., "Taxation and Corporate Financial Policy" (pp. 1251? Equity Values," Accounting Review 19 (2004), 1039-1074.
and M. Feldstein (Eds.), Handbook Gil B., Jr., and George A. Plesko, "The Relation between
1292), in A. J. Auerbach of Manzon,
Public Economics, vol. 3 (Amsterdam: North-Holland, 2002). Financial and Tax Reporting Measures of Income," Tax Imw
Tax Journal 57 Review 55 (2002), 175-214.
Bankman, Joseph, "The Tax Shelter Problem," National
925-936. Mehran, Hamid, "Executive Compensation Structure, Ownership, and
(2004),
"Executive Finan Firm Performance," Journal of Financial Economics 38 (1995),
Brown, JeffreyR., Nellie Liang, and Scott Weisbenner,
cial Incentives and Payout Policy: Firm Responses to the 2003 163-184.
Randall, Andrei Shleifer, and Robert Vishny, "Management Own
Dividend Tax Cut," NBER working paper no. 11002 (2004). Morck,
Jour
Chen, Kong-Pin, and C. Y. Cyrus Chu, "Internal Control vs. External ership and Corporate Performance: An Empirical Analysis,"
A Model of Corporate Income Tax Evasion," RAND nal of Financial Economics 20 (1988), 293-315.
Manipulation:
Journal of Economics 36 (2005), 151-164. Phillips, John, Morton Pincus, and Sonja Olhoft Rego, "Earnings Man
agement: New Evidence Based on Deferred Tax Expense," Ac
Chetty, Raj, and Emmanuel Saez, "Dividend Taxes and Corporate Behav
ior: Evidence from the 2003 Dividend Tax Cut," Quarterly Journal counting Review 78 (2003), 491-521.
120 (2005), 791-833. Slemrod, Joel, "The Economics of Corporate Tax Selfishness," National
of Economics
Crocker, Keith J., and Joel Slemrod, "Corporate Tax Evasion with Agency Tax Journal 51 (2004), 877-899.
89 (2005), 1593-1610. Slemrod, Joel, and Shlomo Yitzhaki, "Tax Avoidance, Evasion and Ad
Costs," Journal of Public Economics
and Irem Tuna, "Why Are Earnings ministration" inA. J.Auerbach and M. Feldstein
Dechow, Patricia, Scott Richardson, (pp. 1423-1470),
of the Earnings Management (Eds.), Handbook of Public Economics, vol. 3 (Amsterdam: North
Kinky? An Examination Explana
tion," Review of Accounting Studies 8 (2003), 355-384. Holland, 2002).
Tax Shelters:
Dechow, Patricia, Richard and Amy Sweeney, "Detecting Earnings
Sloan, U.S. Department of the Treasury, "The Problem of Corporate
Management," Accounting Review 70 (1995), 193-225. Discussion, Analysis and Legislative Proposals" (Washington, DC:
Demsetz, Harold, and Kenneth Lehn, "The Structure of Corporate Own U.S. Government Press Office, 1999).
and Consequences," Journal of Political Economy Weisbach, David, "Ten Truths about Tax Shelters," Tax Law Review 55
ership: Causes
93 (1985), 1155-1177. (2002), 215-253.
Mihir "The Divergence Between Book and Tax Income" White, Halbert, "A Heteroskedasticity-Consistent Covariance Matrix Es
Desai, A., (pp.
the Economy, vol. timator and a Direct Test for Heteroskedasticity," Econometrica 48
169-206), in J.M. Poterba (Ed.), Tax Policy and
17 (Cambridge, MA: MIT Press, 2003). (1980), 817-830.

This content downloaded from 132.211.1.50 on Mon, 21 Sep 2015 13:02:52 UTC
All use subject to JSTOR Terms and Conditions