Chyz (2013)

Journal of Accounting and Economics 56 (2013) 311–328
Contents lists available at ScienceDirect
Journal of Accounting and Economics

journal homepage: www.elsevier.com/locate/jae
Personally tax aggressive executives and

corporate tax sheltering$
James A. Chyz n
College of Business Administration, University of Tennessee, Knoxville, TN 37996, USA
a r t i c l e i n f o abstract
Article history: This paper investigates whether executives who evidence a propensity for personal tax
Received 12 June 2012 evasion (suspect executives) are associated with tax sheltering at the firm level. I adapt
Received in revised form recent research to identify the presence of these executives and examine associations
17 September 2013
between suspect executive presence and firm-level measures of tax sheltering. The results
Accepted 20 September 2013
indicate that the presence of suspect executives is positively associated with proxies
Available online 29 September 2013
for corporate tax sheltering. In addition, firm-years with suspect executive presence
JEL classification: have significantly higher cash tax savings relative to firm-years without suspect executive
K3 presence. I also investigate the firm value implications of suspect executive presence
L5
and find that increases in tax sheltering are incrementally more valuable for firms that
H30
have suspect executives than similar increments made by firms that do not have suspect
H3
executives.
Keywords: & 2013 Elsevier B.V. All rights reserved.
Tax aggressiveness
Tax sheltering
Executive personal traits
Stock option exercise backdating
1. Introduction
In this study I examine whether executives with a propensity to engage in questionable transactions for personal tax gain
are more likely to do the same for corporate tax purposes. Adapting the techniques in Dhaliwal et al. (2009) and Cicero
(2009), I identify executives (suspect) with evidence of individual tax evasion through manipulative stock option exercise
backdating then test for associations between the presence of these executives and corporate tax sheltering. My tests cover
firm-years with and without suspect executive presence allowing me to control for stationary firm characteristics and better
capture the impact suspect executives have on corporate tax sheltering.1 I find that executives who appear willing to push
the envelope for personal tax savings appear to do the same at the firms they manage. My study advances recent research
$
I appreciate helpful comments from Jerry Zimmerman (editor) and an anonymous reviewer. This study stems from my dissertation completed at the
University of Arizona. I thank my dissertation committee: Dan Dhaliwal (chair), Dan Bens, Kirsten Cook, and Oliver Li for their guidance on this project. I
also thank Ken Anderson, Don Bruce, John Campbell, Joe Carcello, Bryan Cloyd, Scott Dyreng, Michelle Hanlon, Pete Lisowsky, LeAnn Luna, Terry Neal, and
Logan Steele. I appreciate suggestions from workshop participants at the University of Arizona; McGill University; Notre Dame University; University of
Arkansas; University of Houston; Southern Methodist University; University of Tennessee; and Virginia Tech University. Finally, I gratefully acknowledge
financial support from the Deloitte Foundation's Doctoral Fellowship Program.
n
Tel.: þ 1 865 974 1701.
E-mail address: jchyz@utk.edu
1
I focus on CEOs and other top executives including executives designated as Board Chairman, CFO, President, or Chief Operating Officer, consistent
with Dhaliwal et al. (2009) and similar to Dyreng et al. (2010) who focus on the 5 most highly compensated executives.
0165-4101/$ - see front matter & 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.jacceco.2013.09.003
312 J.A. Chyz / Journal of Accounting and Economics 56 (2013) 311–328
including Dyreng et al. (2010) by isolating a personal attribute that helps researchers add to our understanding of the cross-
sectional variation in corporate tax sheltering (Shevlin, 2002) and tax aggressiveness (Shevlin, 2007).
A number of recent academic studies have explored the determinants of corporate tax avoidance, but only Dyreng et al.
(2010) examine whether individual top executives have incremental effects on their firms' tax avoidance that cannot be
explained by characteristics of the firm.2 Unlike Dyreng et al. (2010) I focus on the most severe or aggressive form of tax
avoidance, namely, corporate tax shelters. Tax sheltering is differentiated from the more benign “avoidance” documented by
Dyreng et al. (2010) in that the main goal of sheltering is to lower tax liabilities by exploiting discontinuities in the tax law
(Department of the Treasury Report (7-1999), 2009). In some cases, sheltering could enter gray areas of the law when, for
example, it does not exhibit “economic substance” or a “business purpose”, that is, when a shelter is created solely for
evading tax rather than for fulfilling a non (or pre-) tax economic need (Lisowsky, 2010). Focusing on sheltering provides
an empirical setting to align evidence of manipulative executive behavior for personal tax gain with constructs of tax
aggressiveness at the firm level. My results extend the work by Dyreng et al. (2010) who do not link tax avoidance to specific
executive biographical characteristics or managerial styles. Thus, their study provides evidence of an executive impact on
tax avoidance, but is unable to tell us how or why.
Research in the finance and management literatures links personal traits such as CEO narcissism (Chatterjee and
Hambrick, 2007) and underlying psychological “attitudes” (Graham et al., 2012) to corporate financial policy choices
including debt ratios, debt maturity, and acquisition activity. Cronqvist et al. (2012) go beyond an observation of personal
traits and examine whether specific actions a CEO takes in his/her personal life affects the actions they take in their
professional life. Specifically, they show that when CEOs take on larger amounts of leverage in their personal finances, they
also take on larger amounts of leverage at the firms they manage. My study contributes to this literature by investigating an
additional action the CEO takes in their personal lives (i.e., tax evasion). The tax setting of my study is novel because extreme
risk-taking with tax aggressiveness may result in unique negative externalities to the firm such as fines, penalties, losses in
reputation, and tighter monitoring by external parties such as the Internal Revenue Service.
For my empirical tests I combine large sample estimates of corporate tax sheltering probability with data on suspect
executive presence from the Thomson Financial Insiders Database of equity and equity derivative transactions. Because
firms generally do not disclose their tax shelter involvement I use the TSSCORE inferred shelter probability score from
Lisowsky (2010). Lisowsky (2010) notes that his results support the use of TSSCORE as a reliable way to identify the most
severe forms of tax aggressiveness.3 I document a significant positive relation between suspect executive presence and
TSSCORE, suggesting that executives who appear to evade personal taxes are more likely to shelter at the firm-level.
Next, I examine the timing of differences in tax sheltering before, during, and after suspect executive tenure. Relative to
periods with suspect executive presence, I find that tax sheltering probability is lower in both time periods before suspect
executive arrival and time periods after suspect executive departure. Furthermore, reductions in sheltering before suspect
arrival are statistically indistinguishable from reductions in tax sheltering after suspect executive departure. This result
further supports the role of suspect executives in corporate tax sheltering.
In supplemental analyses I document evidence of a negative association between suspect executive presence and cash
effective tax rates (CETR), but no such relationship with GAAP effective tax rates (ETR). Although both are likely noisy
measures of corporate tax sheltering, including these tests helps align my results with prior research (Chen et al., 2010;
Dyreng et al., 2010; Higgins et al., 2013; Chyz et al., 2013; and others). These tests also provide some evidence of whether tax
sheltering translates into cash or GAAP tax savings and highlights the relative importance to suspect executives of financial
reporting tax expense versus cash taxes paid.
An additional question given my findings is whether the presence of suspect executives benefits shareholders. On one
hand, suspect executives impose costs on the firm. Cicero (2009) estimates that stock option exercise backdating results in
an average tax cost to the firm of $85,326 from reduced corporate tax deductions.4 In addition, the use of corporate tax
shelters increases the potential for penalties and back taxes assessed against firms, as well as the potential for higher agency
costs (Desai and Dharmapala, 2009; Chen et al., 2010). On the other hand, backdating stock options could signal executive
expertise in general and corporate tax savvy in particular, suggesting suspect executives could improve firm value through
tax sheltering. In empirical analyses, I find no evidence that tax sheltering or suspect executive presence is individually
related to firm value. Consequently the total effect of suspect executive presence and tax sheltering on firm value appears
to be zero. However, I find that increases in tax sheltering are incrementally more valuable for firms that have suspect
executives than similar increments made by firms that do not have suspect executives.
In addition to the studies mentioned above, my paper also contributes to the accounting literature by providing an
additional explanation for the cross-sectional variation in the levels of corporate tax avoidance. Though empirical studies
have identified factors that impact tax avoidance (Gupta and Newberry, 1997; Phillips, 2003) significant unexplained cross-
sectional variation remains (Desai and Dharmapala, 2006; Dyreng et al., 2008; Shackelford and Shevlin, 2001). My results
2
See Hanlon and Heitzman (2010) for a review of the recent literature documenting determinants of tax avoidance. I define tax aggressiveness as
activities along the higher/more uncertain end of the tax avoidance continuum framework described in Hanlon and Heitzman (2010).
3
I thank Petro Lisowsky for providing me with his TSSCORE measure. The TSSCORE measure is constructed using a more comprehensive set of
determinants but is similar in design to the SHELTER measure from Wilson (2009) and Rego and Wilson (2012).
4
Reduced corporate tax deductions are the result of a lower compensation expense that results from a lower spread between strike price and
opportunistically back-dated stock price. See Section 2.1 and footnote 10 for more detail.
J.A. Chyz / Journal of Accounting and Economics 56 (2013) 311–328 313
also support the hypothesis that individual executive characteristics are an important determinant of tax shelter partici-
pation and generally factor into the pursuit of tax benefits, thereby contributing to the upper echelons theory stream of
literature in management and organizational theory (Hambrick and Mason, 1984; Hambrick, 2007).
The remainder of this paper is structured as follows: In Section 2, I place my study in the context of the existing
literature; in Section 3, I describe the data, the tax sheltering proxy, and empirical tests; in Section 4, I discuss the main
results; in Section 5, I provide additional tests and analyses; and in Section 6, I summarize the findings and conclude.
2. Related literature and hypothesis development
2.1. Tax motivation for stock option exercise backdating
Several studies document a prevalence of options granted on days with relatively low share prices (Yermack, 1997;
Aboody and Kaznick, 2000; Lie, 2005; Bebchuk et al., 2010). These studies assumed executives timed option grants to
precede the release of good news and thereby maximize option value. Lie (2005) was the first to suggest that the lenient SEC
reporting requirements surrounding stock option grants could potentially lead to systematic stock option grant backdating
to coincide with the lowest stock price of the period. Such opportunistic backdating was possible because, prior to SOX,
firms were not required to report stock option grants to the public until after the fiscal year end (Dhaliwal et al., 2009).
Effective August 29, 2002, revised SEC reporting rules requiring firms to report option terms within two days of granting the
option curtailed the ability of executives to retroactively time stock option grants (Heron and Lie, 2007).
Although not mentioned in these earlier studies (Yermack, 1997; Aboody and Kaznick, 2000; Lie, 2005), pre-SOX report-
ing requirements left open the possibility to backdate an exercise to the most advantageous stock price in the month
(Dhaliwal et al., 2009).5 Two recent studies (Cicero, 2009; Dhaliwal et al., 2009) have examined this behavior and suggest
that some executives opportunistically backdate stock option exercises for personal tax reasons.
Upon option exercise, the spread between the option strike price and the stock price at the exercise date is taxed at
ordinary income tax rates.6 If held until the stock qualifies for long-term capital gains tax treatment, any gain subsequent to
exercise is taxed at the lower long-term capital gains rate. Though backdating will not alter the executive's total gain, it does
defer gains and subject more of the gain to lower long-term rates. According to Dhaliwal et al. (2009) the only identifiable
incentive to backdate a stock option exercise to a date with a lower stock price is to minimize the option holders' income tax
liability. Their study estimates that suspect stock option exercises provide average personal tax savings of $55,000 over non-
suspect exercises. Thus, it is likely that backdated stock option exercises capture a measure of executives' personal tax
aggressiveness and propensity to avoid individual income taxes.
It is important to note that studies such as McDonald (2004), and Scholes et al. (2009) have pointed out that exercising
a stock option early to start the clock running on capital gains taxation is not optimal. However, it is unclear whether the
stock option exercises I capture are truly early exercises or are closely coinciding with option expiration dates. In addition,
although my study could also be documenting deviations from decisions that are economically optimal along the dimen-
sions envisioned by McDonald (2004) and Scholes et al. (2009), it does not necessarily disqualify stock option exercise
backdating from signaling personal tax aggressiveness. For executives that choose to exercise-and-hold, the most tax
aggressive implementation of this strategy is to backdate the exercise date to the lowest price available.
2.2. Executive characteristics and economic outcomes
Despite a growing literature on executive characteristics and firm outcomes, the tax literature has largely been silent on
whether individual executives impact firms' tax strategies.7 One exception is Dyreng et al. (2010). Using effective tax rates as
their tax avoidance proxy, Dyreng et al. (2010) apply the approach developed in Bertrand and Schoar (2003) and track the
most highly compensated managers listed in the Execucomp Database to see whether these managers have any impact
on tax avoidance. In initial tests, the authors look for statistically significant individual manager fixed effects in pooled
regressions. In additional tests, the authors form quintiles of managers based upon the level of tax avoidance at the firm
they left (both high and low tax avoidance) to see whether tax avoidance changes at the manager's new firm. They also
examine how tax avoidance changes at the firms that managers leave. The outcome of these tests provides evidence of
a significant manager fixed effect on both GAAP and cash effective tax rates. In their last set of tests, the authors attempt
to associate tax avoidance with managerial styles and individual manager characteristics including highest degree earned,
5
Before the SOX effective date, insiders could wait until the tenth calendar day of the month following the stock option exercise before reporting the
transaction to the SEC. After SOX, insiders are now required to report stock option exercises within two business days of the exercise.
6
It should be noted that this tax treatment applies to non-qualified stock option (NQSO) grants, and not incentive stock option (ISO) grants. As
suggested by Dhaliwal et al. (2009), incentive stock options carry numerous restrictions that increase the likelihood stock options in my sample period
are non-qualified. Jin and Kothari (2008), and Cicero (2009) make similar assumptions regarding ISOs and NQSOs. Furthermore, Cicero (2009) notes that
executives subject to the Alternative Minimum Tax (AMT) would still have an incentive to backdate ISOs.
7
See Hambrick and Mason (1984), Hambrick (2007), Chatterjee and Hambrick (2007), Frank and Goyal (2007), Bamber et al. (2010), Cronqvist et al.
(2012), Graham et al. (2012); and others.
age, confidence, and others. However, they are unable to document consistently significant associations. Consequently their
study, though important, does not provide insights into the mechanism that drives their documented association.
Unlike Dyreng et al. (2010), I consider the most aggressive of corporate tax strategies – tax shelters, and identify a
characteristic not considered in their study-personal tax aggressiveness. Though the literature has not considered this link
specifically, most empirical evidence to date supports a link between executive characteristics and economic outcomes at
the firm level.8 By observing CEOs' personal residence financing levels, Cronqvist et al. (2012) find that CEOs' personal
residence debt financing is a revealed preference of the CEO's personal attitude towards debt that is reflected in higher
corporate leverage. I propose a similar line of reasoning. Specifically, opportunistic stock option exercises are a revealed
preference of an executive's attitude toward the aggressiveness of firm level tax policy. This linkage leads to my first
hypothesis:
H1. There is a positive association between corporate tax sheltering and the presence of suspect executives.
The neoclassical view of the firm suggests managers are perfect substitutes and as such, personal traits and tendencies
will be uncorrelated with firm outcomes (Bertrand and Schoar, 2003). If this view is descriptive with respect to suspect
executives, I would not expect to find support for H1.9
2.3. Firm value and tax sheltering
According to Scholes and Wolfson (1992), tax avoidance should increase a firm's value if the after-tax benefits of tax
avoidance exceed the non-tax costs. Non-tax costs potentially associated with particularly aggressive tax avoidance such as
tax sheltering include tax strategy implementation costs (e.g., promoter and attorney fees), costs associated with IRS audits
and subsequent litigation (e.g., accounting and legal fees), and reputational penalties (Rego and Wilson, 2012). Wilson
(2009) notes that in 14 cases of tax sheltering, the interest charges paid by firms to tax authorities amounted to 40 percent
of the tax savings originally generated by the tax shelter transaction. Hanlon and Slemrod (2009) find evidence that the
stock market penalizes all firms confirmed to have engaged in tax sheltering and Desai and Dharmapala (2006, 2009) assert
that tax sheltering increases the salience of agency costs. Suspect executives also divert the benefit of tax deductions away
from shareholders when they opportunistically exercise stock options to minimize their personal tax liability.10
Despite these non-tax costs, estimates suggest that, on average, benefits outweigh costs. For example, Mills et al. (1998)
find a $4 return for every $1 invested in tax planning, while Rego and Wilson (2012) note that in a recent study of tax
shelters, firms were able to generate annual tax deductions large enough to shield income equal to almost 10 percent of
assets. In addition to this, Weisbach (2002) asserts that observed levels of IRS detection and the resulting penalties, even for
the most aggressive tax positions, are quite low.
The relation between suspect executives, tax sheltering, and firm value is an empirical question. On one hand, stock
option backdating could signal a willingness to engage in transactions for personal gains that are costly to the firm and
shareholders. On the other hand, opportunistic stock option exercises could signal an awareness and comfort with tax
complexities, suggesting suspect executives could be better equipped to improve firm value through tax sheltering. Because
theory does not provide a clear prediction, I test the following two related null hypotheses:
H2a. Suspect executives are not associated with firm value.
H2b. The association between tax sheltering and firm value is not affected by the presence of suspect executives.
Unless suspect executives are also aggressive with respect to other regulators and suppliers, etc., it is possible that there
would be a negative association between firm value and suspect executive presence when tax sheltering activity is very low.
This would be possible since the cost of stock option backdating is unlikely to be offset by tax savings from tax sheltering.
3. Research design
3.1. Sample construction and executive identification
I begin sample construction by collecting all stock acquired by insiders through stock-option exercises between January 1,
1996 and August 29, 2002. These transactions are reported on SEC Form 4 and are machine readable in the Thomson
Financial Insider Filing Database. I retain transaction code “M” exercises (“Exercise of in the-money derivative security
8
Examining executives and not line-level managers is consistent with Rego and Wilson (2012) and Dyreng et al. (2010) that suggest CEOs have a
significant impact on corporate policies and decision-making, including tax planning (even if they are not directly involved in the tax-planning process). In
additional untabulated analyses I retain a sample of CFOs only and find results consistent with my tabulated analyses.
9
In addition, Dyreng et al. (2010) suggest some executives, such as CEOs, may not possess the expertise necessary to influence the implementation of
tax avoidance strategies.
10
When exercising non-qualified stock options, managers are taxed at ordinary income tax rates on the difference between the strike price and stock
price at the exercise date. At the same time, the stock granting firm receives a deduction equal to the amount managers report as ordinary income. Thus the
lower the stock price at exercise, the lower the deduction the firm receives.
acquired pursuant to Rule 16b-3 plan”). I use the role variables (rolecode1, rolecode2) to identify exercises by insiders whose
primary role is CEO, COO, CFO, President, or Chairman of the Board. I then look for dispositions subsequent to stock option
exercises coded “S”, “F”, and “D.” To create a preliminary sample of “exercise-and-hold” transactions, I retain only those
stock option exercises that are not followed by a disposition within 30 days. This approach is consistent with Dhaliwal et al.
(2009) and Aboody et al. (2008). After applying these data screens I am left with 4,154 exercise-and-hold transactions that
originate from executives working at 1,494 unique firms. Of these 1,494 firms, 1,055 meet the sample criteria (described in
more detail below) yielding a final sample of 7,821 firm-year observations.
It is important for my test design (described in more detail in Sections 3.2 and 3.3) to make four distinctions within this
sample of 7,821 firm-years. First, I need to identify firms that have ever been associated with a suspect executive, i.e., firms
that, at some point, employed an executive that engaged in a backdated or “suspect” stock option exercise (suspect firms).
Once I account for suspect firms, I am left with firms that have never employed an executive that engaged in a suspect
exercise (non-suspect firms). Thus, suspect firms and non-suspect firms are mutually exclusive by design. Even though non-
suspect firms never employed a suspect executive, they are included in my sample because they employed, at some point, an
executive that engaged in an exercise-and-hold transaction that was not backdated (non-suspect executives). I retain non-
suspect executives to provide a comparison group that exercised and held without evidence of personal tax aggressiveness.
Next, within each of these two partitions (i.e. suspect firms and non-suspect firms), I need to estimate firm-years when
the executives who engaged in the exercise-and-hold transactions were working at the firms. Doing so provides periods
with and without executive presence. For suspect firms, I use the dichotomous variable SUSPECT_EXEC to capture firm-
years when a suspect executive is present. For non-suspect firms I use the dichotomous variable NON_SUSPECT_EXEC to
capture firm-years when a non-suspect executive is present. Since suspect and non-suspect firms are mutually exclusive,
NON_SUSPECT_EXEC is always equal to zero for suspect firms. Similarly, SUSPECT_EXEC is always equal to zero for non-
suspect firms.11 Table 1 diagrams my sample partitioning and executive presence indicator variable construction.
I identify executives with suspect exercises consistent with Dhaliwal et al. (2009) and Cicero (2009) (i.e., where the
exercise date coincides with the lowest stock price day of the month). To ensure that my data collection results in similar
patterns documented in Dhaliwal et al. (2009), in Fig. 1 I plot the percentage of executive stock option exercises occurring on
the 10 lowest stock price days of the month. If the exercise date during the month were chosen at random, I would expect to
see approximately 4.6 percent of exercises on any given day in the month (Dhaliwal et al., 2009). Instead I find that over 13
percent of executive exercise-and-hold transactions occur on the day with the month's lowest closing stock price. Consistent
with Dhaliwal et al. (2009), I also find that this phenomenon diminishes after SOX as less than 8 percent of these
transactions occur on the day with the month's lowest closing stock price. Both pre-SOX and post-SOX percentages are
comparable to Dhaliwal et al. (2009).
To estimate periods where an executive was present at a given firm, I collect all stock transaction data available in the
Thomson Financial Insiders Database for the executives in my sample. This dataset includes stock option grants, exercises
and stock sales. By keeping the first and last date of any stock transaction for an executive at a given firm, I can estimate a
window where each executive was present at that firm. Since I am uncertain how closely the first available stock transaction
aligns with a given executive's initial presence, I make the simplifying assumption that executive presence begins in the year
following the year of the first stock transaction. In the case of suspect firms, if a firm has more than one executive that had
a suspect exercise, the variable SUSPECT_EXEC ¼1 if any of the suspect executives are present in a firm-year. Similarly, in the
case of non-suspect firms, NON_SUSPECT_EXEC¼1 if any of the non-suspect executives are present. As noted in Table 1,
for suspect firms I identify 1,402 firm-years with suspect executive presence and 680 firm-years without executive presence.
For non-suspect firms I identify 3,184 firm-years with non-suspect executive presence and 2,555 firm-years without
executive presence.12
3.2. Tax sheltering tests of Hypothesis 1
To test for evidence supporting H1, I pool my sample firms that have available TSSCORE and control variable data.
TSSCORE measures used in my study are provided by Petro Lisowsky and are based on the model in Lisowsky (2010).
Appendix A provides additional detail on the TSSCORE measure. All regression specifications use robust standard errors
clustered by firm to address independence concerns (Wooldridge, 2002; Greene, 2003) and include firm fixed effects that
allow each firm to act as its own control over time. Tests of H1 use the following regression:
TSSCOREit ¼ β0 þβ1 SUSPECT_EXEC it þβ2 NON_SUSPECT_EXEC it þ β3 NOLit þ β4 RDit
þβ5 INTANit þ β6 CAPINT it þβ7 LEV it þ β8 FORINC it þ β9 EQ INC it
þβ10 SIZEit þ β11 PT_ROAit þ ∑Firm þ ∑Industries þ ∑Years þ eit ð1Þ
where:
11
It is possible that a suspect firm also had an executive working for them at some point that engaged in an exercise-and-hold transaction but never
engaged in a suspect exercise. Because the firm had a suspect executive present at some point they would be included in the suspect firm group and
NON_SUSPECT_EXEC would never equal 1 for any of this firm's observations.
12
These totals include a small number of observations attributable to executives that have changed firms. After applying data and sample
requirements, tracking suspect executives to other firms adds an additional 80 firm-years, and an additional 11 firm-years tracking non-suspect executives.
Table 1
Sample partitioning and independent variable specification detail.
Notes: This table summarizes the partitioning of sample firms that I label suspect and non-suspect. Suspect firms are those firms that at some point
employed a suspect executive. Non-suspect firms are those firms that ever employed a non-suspect executive, but never employed a suspect executive.
Suspect executives are defined as those executives that have evidence of manipulative stock option exercise backdating [i.e., exercises coinciding with the
lowest stock price day of the month followed by a holding of the stock (as opposed to an immediate sale)]. Non-suspect executives are defined as those
executives that have exercised stock options and held the stock but do not have evidence of manipulative exercise backdating. The diagram also breaks
down the number of firm-years within each of these two partitions where executive presence has been detected. Finally, the diagram provides the reali-
zations of my primary indicator variables of interest (SUSPECT_EXEC) and primary comparison group indicator variable (NON_SUSPECT_EXEC) associated
with each sample partition. SUSPECT_EXEC equals one for firm-years where a suspect executive is deemed present and zero otherwise. NON_SUSPECT_EXEC
equals one for firm-years where a non-suspect executive is present and zero otherwise.
0.14
0.12
0.1
0.08
Pre-Sox Frequency
0.06 Post-Sox Frequency
0.04
0.02
0
1 2 3 4 5 6 7 8 9 10
Fig. 1. Replicating Dhaliwal et al. (2009): distribution of exercise-and-hold stock option exercises on the ten lowest stock price days of the month. This
figure plots the percentage of all exercise-and-hold transactions over 10 days ranked by lowest closing price of the month, with a value of 1 for the
lowest price.
TSSCOREit ¼The inferred estimated probability that a firm is engaged in a tax shelter from Lisowsky (2010). See Appendix A;
SUSPECT_EXECit ¼Indicator variable equal to one for firm-years when a suspect executive is present. This variable always
equals zero for non-suspect firms;
NON_SUSPECT_EXECit ¼ Indicator variable equal to one for firm-years when a non-suspect executive is present. This
variable always equals zero for suspect firms;
NOLit ¼Indicator variable equal to 1 if net operating loss (data52) at time t is greater than zero, and equal to 0 otherwise;
RDit ¼Research and Development expense (data46) at time t, divided by total assets (data6) at time t 1. Set to 0 if
missing;
INTANit ¼Intangible assets (data55) at time t divided by total assets (data6) at time t 1. Set to zero if missing;
CAPINTit ¼Gross property, plant, and equipment (data8) at time t, divided by total assets (data6) at time t 1;
LEVit ¼Total long-term debt (data9) at time t divided by total assets (data6) at time t 1;
FORINCit ¼Pre-tax foreign income (data273) at time t, divided by total assets (data6) at t 1. Set to zero if missing;
EQINCit ¼ Equity in earnings (data33), at time t divided by total assets (data6) at time t 1. Set to zero if missing;
SIZEit ¼Natural log of total sales (data12) at time t;
PT_ROAit ¼Pretax income (data170) less extraordinary items (data192) less special items (data17) divided by total assets
(data6) at time t 1.
As I discuss in more detail below I expect a positive and significant β1, which would be consistent with H1. I also expect
that the magnitude of β1 will exceed the magnitude of β2, but offer no prediction for the sign of β2.
Multivariate regression analysis with a treatment variable is common in tax avoidance studies. There is less precedence
however, when using measures similar to TSSCORE. TSSCORE is an inferred probability estimate constructed with the fitted
values from a first stage logistic regression that includes a number of the control variables commonly used in tax avoidance
research. Including some of these variables in a multivariate regression test could make it difficult to detect the incremental
variation in TSSCORE driven by suspect executive presence relative to some control variables that are, as demonstrated by
Lisowsky (2010), related to TSSCORE. Alternatively, a simple test of differences in means may not properly isolate the
association between suspect executives and sheltering or appropriately control for fundamental differences between firms
in my sample. For sensitivity, I tabulate reduced model results that exclude control variables.
Per Lisowsky (2010), the tax shelter score measure (TSSCORE) has strong discriminatory power in inferring corporate tax
sheltering. Lisowsky (2010) suggests that the TSSCORE model yields an 88 percent chance that a firm with a tax shelter has a
higher predicted TSSCORE than a firm without a tax shelter. The TSSCORE allows researchers to use “clues” in the financial
statements (such as subsidiaries located in tax havens, inconsistent book-tax treatment, litigation losses, and the use of
tax shelter promoters) to infer corporate tax shelter usage in broad samples (Lisowsky, 2010). Because the TSSCORE is an
inferred probability, it suffers from some limitations. Most importantly, the TSSCORE can capture probabilities of tax
sheltering but not direct evidence of tax sheltering. Consequently, my tests are designed to assess whether suspect
executives are more or less likely to be engaging in tax sheltering, but I cannot say for certain whether or not they are.
Despite these limitations, the TSSCORE performs well in identifying large sample evidence of tax sheltering relative to other
tax avoidance measures and is strongly linked to actual tax shelter use (Lisowsky, 2010).13
Firm fixed effects included in Model (1) capture firms' average TSSCORE. Because SUSPECT_EXEC is a time period indicator
equal to one only when suspect executives are present, β1 in Model (1) captures the average incremental effect of suspect
executive presence on TSSCORE. A positive and significant loading on β1 supports my first hypothesis that suspect executives
are positively associated with corporate tax sheltering.14 Aboody et al. (2008) present evidence suggesting that executives
following an exercise-and-hold strategy could be transacting on private information. Since it is unclear what that private
information might be, and more importantly, if it is a source of variation in my dependent variable that is unrelated to
personal tax aggressiveness, my regressions account for the presence of non-suspect executives. Non-suspect executives are
a natural comparison group because, like suspect executives, they have engaged in an exercise-and-hold transaction. Unlike
suspect executives, they have no evidence of tax motivated backdating. Inclusion of this comparison group helps isolate the
association between personal tax aggressiveness and my proxy for corporate tax sheltering. The impact that non-suspect
executive presence has on tax sheltering at non-suspect firms is captured with β2. In the event that the exercise-and-hold
strategy itself (present in both suspect and non-suspect executive periods) drives some of the variation in my left hand side
variable that is unrelated to personal tax aggressiveness, both β1 and β2 will be positively related to tax sheltering. It will be
important to show that the magnitude of any suspect executive effect exceeds that for non-suspect executives. Accordingly,
I include F-tests of the null hypothesis that β1 and β2 are equal.
To supplement results using the dichotomous variable SUSPECT_EXEC, I also construct a continuous firm level variable
(SUSPECT_RATIO) equal to the ratio of suspect exercise-and-hold transactions for all executives at a suspect firm to total
exercise-and-hold transactions at that firm. Accordingly, SUSPECT_RATIO increases in proportion to exercise-and-hold
13
Because of the way TSSCORE is derived (for detail refer to Appendix A), inferences could be problematic if there is reason to believe that inputs into
the shelter prediction model are correlated with the likelihood of being a suspect executive. In other words, there is some risk that existing firm policies, as
reflected in the TSSCORE inputs, encourage suspect stock option backdating. Though it is difficult to completely rule out this risk, in Section 5.2 I present
results that suggest sheltering probability responds to suspect executive arrivals and departures. This helps support the assertion that suspect executives
drive firm policy towards more tax sheltering that is reflected in the Lisowsky (2010) TSSCORE inputs, minimizing the risk of a problematic mechanical
relationship in my results.
14
As is evident in Fig. 1, some exercise-and-hold transactions are likely to randomly fall on the lowest stock price day of the month. Because I identify
all executives with exercise-and-hold transactions occurring on the lowest price day of the month as suspect, my measure potentially includes some
executives that were “lucky” and not necessarily personally tax aggressive. Consequently, coefficients capturing suspect executive presence potentially
suffer from measurement error in the form of attenuation bias. If it were possible to remove suspect executive identification measurement error, it is
reasonable to assume that coefficients capturing suspect executive presence and associated economic magnitudes would be larger than what I document.
Thus, my results likely document a lower bound of suspect executives' impact on tax sheltering.
transactions that are suspect.15 Because non-suspect firms by definition have no suspect exercise-and-hold transactions,
I continue to use the variable NON_SUSPECT_EXEC when testing with SUSPECT_RATIO. Lastly, I use the variable NUM_SUSPECT
to capture the number of suspect executives at a given firm (i.e., the number of executives at a firm who have engaged
in a suspect exercise-and-hold transaction). When using this specification, I also include NUM_NON_SUSPECT, which
equals the number of executives that engaged in a non-suspect exercise-and-hold transaction at a given non-
suspect firm.
Model (1) also includes year and industry fixed effects along with control variables that are common in the tax
accounting literature (Gupta and Newberry, 1997; Phillips 2003; Dyreng and Lindsey, 2009; Higgins et al., 2013; McGuire
et al., 2011; Brown and Drake, 2011; and others).16
3.3. Firm value tests of Hypotheses 2a and 2b
To test H2a and H2b and evidence of relationships among firm value, tax sheltering, and suspect executive presence, I use
the following regression:
TOBINSQ it ¼ β0 þβ1 SUSPECT_EXEC it þβ2 NON_SUSPECT_EXEC it þ β3 TSSCOREit
þ β4 SUSPECT_EXEC it TSSCOREit þ β5 NON_SUSPECT_EXEC it TSSCOREit
þ β6 NOLit þ β7 RDit þ β8 AVG_SGROW it þ β9 LOSSit
þ β10 LEV it þ β11 ABS_FORINC it þβ12 LN_AGEit
þ β13 SIZEit þβ14 PT_ROAit þ ∑Firm þ∑Industries þ ∑Yearsþ eit ð2Þ
where:
TOBINSQit ¼Market value of assets (data6þ (data24 data25) – data60) divided by book value of assets (data6) as in
Desai and Dharmapala (2009);
AVG_SGROWit ¼Average of last 3 years percentage growth in sales (data12) ending in year t.
LOSSit ¼Indicator variable equal to 1 if income before extraordinary items (data18) in year t is less than zero and zero
otherwise;
ABS_FORINCit ¼Absolute value of pre-tax foreign income (data273) at time t (set to zero if missing), divided by total assets
at t 1.
LN_AGEit ¼Log of the number of years the firm appears on Compustat;
All other variables as defined in Model (1).

Model (2) uses robust standard errors clustered by firm to address independence concerns (Wooldridge, 2002; Greene,
2003) and includes firm fixed effects that allow each firm to act as its own control over time. As in Model (1), I include three
different specifications to capture suspect executive presence. TOBINSQ, my proxy for firm value, is based on Desai and
Dharmapala (2009).17 Coefficient β1 captures the average effect of suspect executive presence on firm value and β3 captures
the average effect of tax sheltering on firm value. Hypothesis 2a, stated in the null form, predicts no relationship between
firm value and suspect executive presence. This result would be supported by a statistically insignificant β1 coefficient
estimate. The coefficient β4 helps to illustrate whether firm value is associated with tax sheltering in the presence of suspect
executives. Firm fixed effects capture the firm average TOBINSQ allowing β4 to capture the average incremental effect
of suspect executive presence in the relationship between TSSCORE and TOBINSQ. Hypothesis 2b, stated in the null form,
predicts that the association between tax sheltering and firm value is not affected by the presence of suspect executives. This
result would be supported by a statistically insignificant β4 coefficient estimate. I also include industry, and year fixed effects
along with control variables that are common in the recent literature (Adams and Santos, 2006; Brown and Caylor, 2006;
Desai and Dharmapala, 2009; and others).18
3.4. Data, sample and descriptive statistics
I collect firm-level variables from the Compustat Database for the period spanning 1995–2006 to coincide with my data
availability for TSSCORE. This period approximates the start of the FAS 109 reporting environment, consistent with prior
15
I re-run my analyses after dropping firms that had only one exercise and hold transaction to address concerns that a suspect stock option exercise
may occur by chance, or that an exercise and hold transaction is improperly identified. I find similar results.
16
Because industry affiliation varies over time for at least some of the firms in my sample, the inclusion of industry and firm fixed effects is
appropriate. As a robustness check I run tests of Model (1) excluding industry fixed effects. My inferences are unchanged.
17
Tobin's Q accounts for the book as well as the market value of equity and the value of debt. A simpler alternative as suggested by Desai and
Dharmapala (2009) is the market value of common equity scaled by the book value of assets. I measure market value of common equity as the calendar-
year closing stock price multiplied by the number of common shares, and the book value of assets is measured as total assets. Using this variable instead of
TOBINSQ does not alter my inferences.
18
As a robustness check I run tests of Model (2) excluding industry fixed effects. My inferences are mostly unchanged. The one exception is the
statistical significance of the coefficient loading on the interaction between SUSPECT_RATIO and TSSCORE. Without controlling for industry variation over
time, this coefficient loading drops below statistical significance at conventional levels (1.315, t-stat ¼ 1.64).
Table 2
Sample composition and descriptive statistics.
Panel A: Industry distribution
Suspect firms Non-suspect firms

Full sample Firm-years with executive presence Firm-years with executive presence
Number Percent Number Percent Number Percent
Consumer Nondurables 479 6.12 60 4.28 211 6.63

Consumer Durables 246 3.15 57 4.07 88 2.76
Manufacturing 1140 14.58 194 13.84 490 15.39
Oil, Gas and Coal Extraction and Products 293 3.75 31 2.21 143 4.49
Chemicals and Allied Products 250 3.20 45 3.21 111 3.49
Business Equipment 2545 32.54 562 40.09 959 30.12
Telephone and Television Transmission 165 2.11 8 0.57 77 2.42
Wholesale, Retail and Some Services 829 10.60 148 10.56 333 10.46
Healthcare, Medical Equipment and Drugs 1072 13.71 170 12.13 471 14.79
Insurance/Real Estate 73 0.93 24 1.71 19 0.60
Other 729 9.32 103 7.35 282 8.86
7,821 1,402 3,184
Panel B: Full sample descriptive statistics
Suspect firms – all years Non-suspect firms – all years

n¼2,082 n ¼5,739
Mean Median Std dev. Mean Median Std dev.
TSSCORE 0.749 0.894nnn 0.306 0.757 0.929 0.314

SUSPECT_EXEC 0.673nnn 1.000nnn 0.469 0.000 0.000 0.000
SUSPECT_RATIO 0.330nnn 0.250nnn 0.337 0.000 0.000 0.000
NUM_SUSPECT_EXEC 0.913nnn 1.000nnn 0.855 0.000 0.000 0.000
NON_SUSPECT_EXEC 0.000nnn 0.000nnn 0.000 0.555 1.000 0.497
NUM_NON_SUSPECT_EXEC 0.000nnn 0.000nnn 0.000 0.956 1.000 1.095
NOL 0.333nnn 0.000nnn 0.471 0.345 0.000 0.475
RD 0.098nnn 0.046nnn 0.152 0.071 0.019 0.131
INTAN 0.114nnn 0.023nnn 0.243 0.156 0.051 0.378
CAPINT 0.485nnn 0.386nnn 0.386 0.572 0.452 0.456
LEV 0.157nnn 0.063nnn 0.244 0.189 0.120 0.292
FORINC 0.013 0.000nnn 0.049 0.013 0.000 0.042
EQINC 0.000 0.000nnn 0.025 0.000 0.000 0.018
SIZE 5.599nnn 5.560 1.969 5.956 5.957 2.126
PT_ROA 0.070 0.084 0.220 0.071 0.086 0.199
Panel C: Suspect firms descriptive statistics – all years, partitioned on suspect executive presence
With executive presence Without executive presence

Total n¼2,082 n¼1,402 n¼ 680
Mean Median Std dev. Mean Median Std dev.
TSSCORE 0.781nnn 0.919nnn 0.289 0.683 0.799 0.329

SUSPECT_EXEC 1.000nnn 1.000nnn 0.000 0.000 0.000 0.000
SUSPECT_RATIO 0.489nnn 0.444nnn 0.300 0.000 0.000 0.000
NUM_SUSPECT_EXEC 1.355nnn 1.000nnn 0.696 0.000 0.000 0.000
NON_SUSPECT_EXEC 0.000 0.000 0.000 0.000 0.000 0.000
NUM_NON_SUSPECT_EXEC 0.000 0.000 0.000 0.000 0.000 0.000
NOL 0.323 0.000 0.468 0.353 0.000 0.478
RD 0.098 0.056nnn 0.135 0.099 0.034 0.180
INTAN 0.111 0.030n 0.232 0.121 0.015 0.265
CAPINT 0.472nn 0.376nnn 0.373 0.512 0.421 0.412
LEV 0.157 0.057 0.248 0.157 0.077 0.235
FORINC 0.015 0.000nnn 0.047 0.008 0.000 0.051
EQINC 0.001nnn 0.000 0.029 0.000 0.000 0.012
SIZE 5.709nn 5.704nnn 1.954 5.373 5.241 1.982
PT_ROA 0.078nnn 0.091nnn 0.215 0.054 0.069 0.229
Panel D: Non-suspect firms descriptive statistics – all years, partitioned on non-suspect executive presence

Total n¼5,739 n¼3,184 n¼2,555
Mean Median Std dev. Mean Median Std Dev
nnn nnn
TSSCORE 0.779 0.941 0.299 0.730 0.906 0.329
Table 2 (continued )
Panel D: Non-suspect firms descriptive statistics – all years, partitioned on non-suspect executive presence

Total n¼ 5,739 n¼ 3,184 n¼ 2,555
Mean Median Std dev. Mean Median Std Dev
SUSPECT_EXEC 0.000 0.000 0.000 0.000 0.000 0.000

SUSPECT_RATIO 0.000 0.000 0.000 0.000 0.000 0.000
NUM_SUSPECT_EXEC 0.000 0.000 0.000 0.000 0.000 0.000
NON_SUSPECT_EXEC 1.000nnn 1.000nnn 0.000 0.000 0.000 0.000
NUM_NON_SUSPECT_EXEC 1.723nnn 1.000nnn 0.915 0.000 0.000 0.000
NOL 0.342 0.000 0.474 0.348 0.000 0.477
RD 0.070 0.018 0.138 0.071 0.020 0.122
INTAN 0.162 0.054 0.356 0.149 0.048 0.402
CAPINT 0.561nn 0.442nn 0.447 0.586 0.466 0.466
LEV 0.195 0.127 0.276 0.182 0.115 0.312
FORINC 0.012n 0.000 0.043 0.014 0.000 0.039
EQINC 0.000 0.000 0.020 0.000 0.000 0.014
SIZE 6.065nnn 6.087nnn 2.064 5.822 5.815 2.194
PT_ROA 0.077nn 0.089nnn 0.186 0.064 0.083 0.214
Notes: This table presents information on the primary sample composition including descriptive statistics for the pooled sample and partitions of suspect
and non-suspect firms detailed in Table 1. Panel A presents industry composition with classifications based on the Fama and French 12 industries. Panel A
compares the firm-year industry distribution of the non-suspect executive group of firm-years to the suspect executive group of firm-years when these
types of executives are present at the firm. Panels B through D provide descriptive statistics for the primary sample. All continuous variables are truncated
at 1% and 99% to avoid the influence of outliers on statistical inference. Panel B provides descriptive statistics for all sample firms partitioned by suspect and
non-suspect firm. Panel C provides descriptive statistics for suspect firms, partitioned on firm-years with and without the presence of suspect executives.
Panel D provides descriptive statistics for non-suspect firm-years, partitioned on firm-years with and without non-suspect executive presence. Data
definitions are as follows: TSSCORE is the tax shelter probability score from Lisowsky (2010). SUSPECT_RATIO is the firm level ratio of total suspect exercise-
and-hold transactions to all exercise-and-hold transactions when suspect executives are present and zero otherwise. NUM_SUSPECT_EXEC is the number of
executives associated with a firm that engaged in a suspect exercise and hold transaction when suspect executives are present and zero otherwise.
NUM_NON_SUSPECT_EXEC is the number of executives associated with a firm that engaged in an exercise-and-hold transaction (that was not suspect) when
non-suspect executives are present and zero otherwise. NOL is a dummy variable equal to 1 if the net operating loss in the observation year is greater than
zero, and zero otherwise. RD is research and development expense scaled by lagged total assets. INTAN is intangible assets scaled by lagged total assets.
CAPINT is a measure of capital intensity, defined as gross PP&E scaled by lagged total assets. LEV is total long-term debt scaled by lagged total assets. FORINC
equals pre-tax foreign income scaled by lagged total assets. EQINC is the equity in earnings scaled by lagged total assets. SIZE is defined as the natural log of
total sales. PT_ROA equals pretax income less extraordinary items and special items, all scaled by lagged total assets. All other items as defined in Table 1. n,
nn
And nnn next to the mean (median) indicate a 10%, 5% and 1%, respectively, significant difference between partitions using a two-tailed t-test (two-tailed
difference in medians test).
research, and ends with my access to the TSSCORE measure. Consistent with extant prior research I exclude financial firms
(SIC code 6000–6411) and utilities (SIC code 4900–4999). After truncating at the 1st and 99th percentiles of all primary
regressor variables, I merge data on suspect and non-suspect executives described in Section 3.1 above. This results in a final
primary sample of 7,821 firm-years.19 My sample, as detailed in Table 1, consists of 1,402 (3,184) suspect (non-suspect) firm-
years where executives are present representing 284 (771) distinct firms. My sample also includes 680 (2,555) firm-years for
suspect (non-suspect) firms with no executive presence. This sample attribute appears to provide a reasonable number of
observations to test for differences in tax sheltering over periods with and without executive presence.
Panel A of Table 2 presents an industry breakdown of my sample firms into the Fama-French 12 groupings.20 I partition
the sample by suspect and non-suspect firms and find that, for the most part, the industry distribution for suspect firms is
similar to that for non-suspect firms. The one exception is the Business Equipment category, which has a 10 percent higher
proportion in the suspect firm partition.21 To help control for the influence of industry affiliation, regression tests include
industry fixed-effects.
Panel B of Table 2 summarizes statistics for variables used in primary regressions. I present the data for suspect and non-
suspect firms separately. Average realizations of TSSCORE are higher in the non-suspect partition, but this difference is
statistically significant in medians only and not in means. Panel C presents summary statistics for suspect firms, partitioned
by firm-years with and without suspect executive presence. Both the mean and median differences in TSSCORE are greater
when suspect executives are present. Across the partitions, the mean value of TSSCORE is higher by approximately 10
percentage points and the median value is higher by approximately 12 percentage points. Both increases are statistically
significant. Panel D presents summary statistics for non-suspect firms partitioned on firm years with and without non-
19
Increasing the sample size to include all Compustat firms that meet my data requirements (not only firms with executives that followed an exercise-
and-hold strategy) does not alter my inferences.
20
http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/Data_Library/det_12_ind_port.html.
21
For both suspect and non-suspect firms, the Business Equipment category represents a substantial portion of my sample. This could reflect
compensation practices in this industry that leads to more stock option granting. Because I include firm and industry fixed effects in my regression tests, I
mitigate concerns that my results are industry driven.
suspect executive presence. As in Panel C, TSSCORE increases when non-suspect executives are present, but the magnitude of
this increase is much smaller relative to suspect firms (5 percentage points at the mean and 3.5 percentage points at the
median).
4. Empirical results
4.1. Tax sheltering regression tests
Table 3 reports regression results of Model (1). Columns (1), (3), and (5) present regression results without control
variables. In all columns, the three specifications that capture suspect executive presence are positive and statistically
significant, supporting H1. The variables that capture non-suspect executive presence also are positive and significant in
these columns, but the magnitude of suspect executive coefficients are larger. More importantly, reported F-tests confirm
that I can reject the null of equality between suspect and non-suspect coefficients (all p-values are less than 0.10). After
adding control variables in columns (2), (4), and (6), I find that all specifications capturing suspect executive presence
remain positive and significant, while none of the variables capturing non-suspect executive presence are statistically
significant. This lack of statistical significance in fully specified multivariate regressions could be consistent with control
variables accounting for variation in the TSSCORE measure that is related to the exercise-and-hold strategy but unrelated to
personal tax aggressiveness.
Table 3
Firm fixed-effect regressions of corporate tax sheltering on suspect executive presence.
Predicted sign (1) (2) (3) (4) (5) (6)
SUSPECT_EXEC þ 0.057 0.034 – – – –

(4.16)nnn (3.15)nnn – – – –
SUSPECT _RATIO þ – – 0.075 0.043 – –
– – (3.37)nnn (2.41)nn – –
NUM_SUSPECT_EXEC þ – – – – 0.039 0.024
– – – – (4.31)nnn (3.14)nnn
NON_SUSPECT_EXEC ? 0.022 0.008 0.021 0.007 – –
(2.82)nnn (1.28) (2.61)nnn (1.09) – –
NUM_NON_SUSPECT_EXEC ? – – – – 0.012 0.004
– – – – (2.78)nnn (1.06)
NOL 0.004 0.004 0.004
(0.49) (0.50) (0.50)
RD ? 0.236 0.236 0.236
( 5.82)nnn ( 5.84)nnn ( 5.78)nnn
INTAN þ 0.023 0.023 0.023
(2.37)nn (2.37)nn (2.36)nn
CAPINT ? 0.024 0.024 0.025
( 1.46) ( 1.49) ( 1.50)
LEV 0.031 0.031 0.031
(2.53)nn (2.56)nnn (2.54)nn
FORINC þ 0.681 0.682 0.680
(5.97)nnn (5.97)nnn (5.96)nnn
EQINC þ 0.030 0.028 0.036
(0.15) (0.14) (0.18)
SIZE þ 0.086 0.086 0.085
(10.53)nnn (10.50)nnn (10.56)nnn
PT_ROA þ 0.167 0.167 0.167
(8.27)nnn (8.28)nnn (8.28)nnn
F-test: β1 ¼β2 (p-value) 0.0211 0.0271 0.0180 0.0496 0.0054 0.0130
Firm, Industry & Year Fixed Effects Yes Yes Yes Yes Yes Yes
Adjusted-R2 0.8237 0.8671 0.8233 0.8669 0.8236 0.8670
N obs 7821 7821 7821 7821 7821 7821
Notes: This table presents the results for tests of Model (1):
TSSCOREit ¼ β0 þ β1 SUSPECT_EXEC it þ β2 NON_SUSPECT_EXEC it þ β3 NOLit þ β4 RDit

þ β5 INTAN it þ β6 CAPINT it þ β7 LEV it þ β8 FORINC it þ β9 EQ INC it
þ β10 SIZEit þ β11 PT_ROAit þ ∑Firm þ ∑Industries þ ∑Years þ eit
all variables are as described in Tables 1 and 2. All regressions include firm, year, and industry fixed-effects. Amounts in parentheses are t-statistics based
on White (1980) standard errors clustered by firm. n, nn And nnn next to the coefficient estimates indicate a 10%, 5% and 1%, respectively, significance level
using a two-tailed test. F-test p-values denote the statistical significance of tests for the difference in coefficient loadings between the variables that capture
suspect and non-suspect executive presence.
SUSPECT_EXEC in column (2) provides some sense of the economic significance suspect executive presence has on tax
sheltering. The coefficient on SUSPECT_EXEC is positive and significant (0.034, t-stat ¼3.15), which suggests that after
controlling for other sources of variation, suspect executive presence increases by 3.4 percentage points the inferred
probability a tax shelter is being used.22 Without the addition of control variables [column (1)], the increase is 5.7
percentage points, an increase of approximately 8 percent relative to the TSSCORE sample mean. The results of these tests
support H1 that suspect executive presence is positively associated with tax sheltering.
Coefficients on INTAN, LEV, FORINC, SIZE, and, PT_ROA are positive and statistically significant across the three specifi-
cations. The coefficient on RD is negative and consistently statistically significant. Coefficients on NOL, CAPINT, and EQINC are
never statistically significant. The direction and statistical significance of these coefficients have precedence in tax avoidance
research. This also is true of the large t-statistics on SIZE. In terms of t-statistic magnitude, firm size is frequently the
strongest predictor of tax avoidance. Because research and development activities (R&D) are often tax favored, the negative
coefficient on RND could be the result of a substitution effect between tax sheltering and R&D for the firms in my sample. In
addition, firms engaged in R&D often are immature and less profitable having have fewer resources or need to invest in tax
sheltering. The positive association between LEV and the probability of tax sheltering is somewhat surprising given the
results in the extant tax avoidance literature. This positive association could be a result of firms in my sample seeking tax
savings from tax sheltering due to constraints from adding to debt as a more conventional tax shield. Many commonly used
tax shelters, including the Delaware Trademark Holding Company and Double Irish, involve the licensing of intangible
assets, often to or from foreign subsidiaries. Such strategies could explain and predict the positive relationship between tax
sheltering probability and both INTAN and FORINC. Large firms (SIZE) and better performing firms (PT_ROA) potentially have
more future taxable income. A dollar invested in tax sheltering for larger and better performing firms is likely to have a
higher expected future payoff. Economies of scale in tax sheltering could also explain the positive association with firm size
as sheltering costs are unlikely to increase proportionately with size.
4.2. Firm value regression tests
Table 4 presents results of tests for H2a and H2b. I test for both the main effect of suspect executive presence on firm
value (H2a) and the interactive effect of tax sheltering and suspect executive presence on firm value (H2b). I cannot reject
H2a that suspect executives alone are not related to firm value, as none of the specifications capturing suspect executive
presence are statistically significant. This also is true of the specifications capturing non-suspect executives.
I can reject H2b that the association between tax sheltering and firm value is not affected by the presence of suspect
executives. I find that all specifications of suspect executive presence interacted with tax sheltering are positive and
statistically significant. Interactions with non-suspect executive presence and tax sheltering are not statistically significant.
Furthermore, I can reject the null that coefficients on interactions with suspect executive presence and coefficients on
interactions with non-suspect executive presence are equal. P-values for F-tests of this difference in all columns are 0.10
or less.
The results in Table 4 allow for some interesting interpretations of the roles that sheltering and suspect executives play in
firm value. The main effect of sheltering on firm value in the absence of suspect executives (i.e. the coefficient on TSSCORE) is
negative but statistically indistinguishable from zero. In addition, an untabulated F-test confirms that the effect of sheltering
on firm value in the presence of suspect executives, (i.e. the sum of the negative coefficient on TSSCORE and positive
coefficient on SUSPECT_EXEC TSSCORE) is positive but is also statistically indistinguishable from zero. The results appear to
support a slight (but statistically insignificant) decrease in firm value for increases in sheltering among firms that do not
have a suspect executive, and a slight (but statistically insignificant) positive increase in firm value for increases in sheltering
among firms with a suspect executive. The positive and significant coefficient on SUSPECT_EXEC TSSCORE suggests that the
difference between this slight decrease and slight increase is statistically significant. In other words, increases in tax
sheltering are incrementally more valuable for firms that have suspect executives than similar increments made by firms
that do not have suspect executives. Finally, it appears that the total effect of sheltering and suspect executive presence (i.e.
the combination of SUSPECT, TSSCORE, and SUSPECT TSSCORE) is essentially zero. This suggests that for suspect firms, firm-
years with very low sheltering and no suspect executive presence are no different in terms of TOBINSQ than firm-years with
very high levels of sheltering and suspect executive presence.
As noted above, Table 4 provides some support for the notion that increases in tax sheltering are incrementally more
valuable for firms that have suspect executives than similar increments in sheltering made by firms with no suspect
executive presence. This relationship does not appear to hold in the presence of non-suspect executives, suggesting that
suspect executives play a unique role in the association between firm value and tax sheltering. One interpretation of this role
is that tax sheltering by suspect executives is valued more than tax sheltering by non-suspect executives. This is an
22
Because my tests cannot observe actual sheltering activity it is difficult to gauge the economic magnitude of the suspect executive effect derived
from multivariate regressions. However it is possible to make some comparisons of the marginal effect of suspect executive presence relative to other firm-
level variables in my model. I express marginal effects in terms of the inter-quartile change (i.e. 75th percentile less 25th percentile) of the independent
variables. For both SUSPECT and NON_SUSPECT I assume a movement from no executive presence to executive presence (1–0). Results from this analysis
suggest that the marginal effect of suspect executive presence (SUSPECT_EXEC) is high relative to other regressor variables, ranking second only to the effect
of size (SIZE).
Table 4
Firm fixed-effect regressions of firm value on suspect executive presence and corporate tax sheltering.
Dependent variable TOBINSQ
(1) (2) (3)
SUSPECT_EXEC 0.566 – –
( 1.16) – –
SUSPECT_RATIO – 0.838 –
– ( 1.15) –
NUM_SUSPECT_EXEC – – 0.371
– – ( 1.08)
NON_SUSPECT_EXEC 0.105 0.094 –
( 0.35) ( 0.32) –
NUM_NON_SUSPECT_EXEC – – 0.068
– – ( 0.46)
TSSCORE 0.553 0.443 0.632
( 1.19) ( 0.95) ( 1.36)
SUSPECT_EXEC TSSCORE 1.112 – –
(2.06)nn – –
SUSPECT_RATIO TSSCORE – 1.321 –
– (1.66)n –
NUM_SUSPECT_EXEC TSSCORE – – 0.954
– – (2.44)nn
NON_SUSPECT_EXEC TSSCORE 0.002 0.045 –
(0.00) ( 0.13) –
NUM_NON_SUSPECT_EXEC TSSCORE – – 0.101
– – (0.64)
NOL 0.037 0.035 0.034
( 0.30) ( 0.28) ( 0.27)
RD 6.649 6.655 6.648
(3.44)nnn (3.43)nnn (3.43)nnn
AVG_SGROW 0.171 0.170 0.170
(1.84)nn (1.83)n (1.83)n
LOSS 0.068 0.074 0.061
( 0.40) ( 0.44) ( 0.35)
LEV 0.582 0.580 0.579
( 2.73)nnn ( 2.73)nnn ( 2.73)nnn
ABS_FORINC 0.329 0.381 0.306
(0.12) (0.14) (0.11)
LN_AGE 0.144 0.146 0.149
( 2.36)nn ( 2.38)nn ( 2.42)nn
SIZE 0.0767 0.766 0.778
( 2.93)nnn ( 2.93)nnn ( 2.96)nnn
PT_ROA 4.062 4.053 4.069
(3.14)nnn (3.13)nnn (3.14)nnn
F-test: β4 ¼β5 (p-value) 0.0563 0.0944 0.0282
Firm, Industry & Year Fixed Effects Yes Yes Yes

Adjusted-R2 0.3981 0.3975 0.3991
N obs 7815 7815 7815
Notes: This table presents the results for tests of Model (2):
TOBINSQ it ¼ β0 þ b1 SUSPECT_EXEC it þ β2 NON_SUSPECT_EXEC it þ β3 TSSCOREit

þ β4 SUSPECT_EXEC it nTSSCOREit
þ β5 NON_SUSPECT_EXEC it nTSSCOREit þ β6 NOLit þ β7 RDit
þ β8 AVG_SGROW it þ β9 LOSSit þ β10 LEV it þ β11 ABS_FORINC it þ β12 LN_AGEit
þ β13 SIZEit þ β14 PT_ROAit þ ∑Firm þ ∑Industries þ ∑Years þ eit
TOBINSQ is the market value of assets divided by the book value of assets as defined in Desai and Dharmapala (2009). AVG_SGROW is the average of the last
3 years worth of sales growth. ABS_FORINC is the absolute value of pre-tax foreign income scaled by lagged total assets. LN_AGE is the log of the number of
years the firm appears on Compustat. All other variables are as described in Tables 1 and 2. All regressions include firm, year, and industry fixed-effects.
Amounts in parentheses are t-statistics based on White (1980) standard errors clustered by firm. n, nn And nnn next to the coefficient estimates indicate a
10%, 5% and 1%, respectively, significance level using a two-tailed test. F-test p-values denote the statistical significance of tests for the difference in
coefficient loadings between the variables that capture interactions between tax sheltering and suspect and non-suspect executive presence.
intriguing result within the context of the agency view of tax avoidance (Desai and Dharmapala, 2006, 2009) because
of the suspect nature of backdated stock option exercises. My results could be partially explained if any value derived
from sheltering activities by suspect executives were not offset by other forms of rent extraction or market participants'
price protection.23 Another related alternative is that, relative to non-suspect executives, suspect executives are more
23
Blaylock (2011) asserts that there is little empirical evidence to support a link between tax sheltering and rent extraction in U.S. settings.
opportunistic when engaging in tax sheltering activities and are more likely to do so when they expect higher future taxable
income. Higher future taxable income likely increases the expected payoff from tax sheltering activities, and is likely
associated with higher levels of TOBINSQ.
Coefficient loadings on control variables in Table 4 appear reasonable and generally consistent with extant research.
Firms in my sample that are more profitable (PT_ROA), that invest more in research and development (RND), and firms with
greater sales growth (AVG_SGROW) are positively related to TOBINSQ. Older (LN_AGE) and larger (SIZE) firms, and firms with
higher leverage (LEV) are negatively related to TOBINSQ.
5. Additional analysis and robustness testing
5.1. Option grant backdating
I argue that suspect stock option exercise backdating reflects a propensity for executives to engage in questionable
transactions for tax purposes. To support this claim, I find a different independent variable that also captures questionable
transactions which are not driven by personal tax motivations, specifically, stock option grant backdating. I follow the
method used in Bebchuk et al. (2010) to collect a sample of unscheduled grants to executives in the pre-Sox period. Bebchuk
et al. (2010) capture potentially backdated stock option grants by identifying those executives (BACKDATE_EXEC) who
received stock option grants at the lowest price of the month. In the same manner that SUSPECT_EXEC is constructed in my
primary tests, the variable BACKDATE_EXEC is an indicator that equals one for firm years when stock option grant backdating
executives are deemed present and zero for years when these executives are not present. In Table 5, I present results
that adapt Model (1) to include BACKDATE_EXEC. I run regressions with and without the inclusion of SUSPECT_EXEC and
NON_SUSPECT_EXEC across three samples. The first sample [columns (1) and (2)], follows an approach similar to my primary
Table 5
Firm fixed-effect regressions of corporate tax sheltering on stock option grant backdating executive presence.
Dependent variable TSSCORE
Sample Option Grant Firms Option Grant Firms and Exercise and Hold Firms All Firms
(1) (2) (3) (4) (5) (6)
BACKDATE_EXEC 0.0102 0.005 0.010 0.006 0.015 0.010

(1.03) (0.54) (0.99) (0.63) (1.47) (0.90)
SUSPECT_EXEC 0.052 0.036 0.038
(3.84)nnn (3.22)nnn (3.36)nnn
NON_SUSPECT_EXEC 0.001 0.008 0.010
( 0.12) (1.31) (1.67)n
NOL 0.008 0.008 0.002 0.002 0.002 0.002
(1.05) (1.05) (0.24) (0.28) ( 0.47) ( 0.46)
RD 0.164 0.163 0.160 0.159 0.107 0.106
( 3.14)nnn ( 3.13)nnn ( 3.81)nnn ( 3.82)nnn ( 14.17) ( 4.18)nnn
INTAN 0.062 0.061 0.034 0.034 0.035 0.034
(5.30)nnn (5.24)nnn (3.24)nnn (3.25)nnn (4.94)nnn (4.95)nnn
CAPINT 0.017 0.016 0.021 0.021 0.016 0.016
( 1.03) ( 1.00) ( 1.60) ( 1.57) ( 2.14)nn ( 2.11)nn
LEV 0.027 0.026 0.032 0.032 0.035 0.034
(1.90)n (1.87)n (2.84)nnn (2.83)nnn (4.51)nnn (4.51)nnn
FORINC 0.782 0.782 0.716 0.712 0.787 0.788
EQINC 0.653 0.658 0.034 0.047 0.052 0.047
( 3.04)nnn ( 3.05)nnn ( 0.18) ( 0.24) (0.51) (0.46)
SIZE 0.099 0.099 0.089 0.089 0.078 0.079
PT_ROA 0.143 0.143 0.161 0.160 0.121 0.121
Firm, Industry & Year Fixed Effects Yes Yes Yes Yes Yes Yes
Adjusted-R2 0.8711 0.8715 0.8716 0.8719 0.8921 0.8923
N obs 8,777 8,777 12,555 12,555 37,352 37,352
Notes: This table presents the results for tests that adapt Model (1) to include firm-years with and without the presence of executives that have evidence of
stock option grant backdating (BACKDATE_EXEC). All other variables are as described in Tables 1 and 2. Results for three samples are presented. Columns (1)
and (2) contain all firms with available data that have ever granted a non-scheduled stock option in the sample period based on the method used in
Bebchuk et al. (2010). Columns (3) and (4) add firms from my primary exercise–and–hold sample, and columns (5) and (6) contain the full universe of firms
with available data in the sample period. All regressions include firm, year, and industry fixed-effects. Amounts in parentheses are t-statistics based on
White (1980) standard errors clustered by firm. n, nn And nnn next to the coefficient estimates indicate a 10%, 5% and 1%, respectively, significance level using
a two-tailed test.
80
78
76
74
72
70 TSSCORE
68
66
64
62
Before During After
Fig. 2. Average tax sheltering probability (TSSCORE) relative to periods with suspect executive presence. This figure plots the average realizations of
TSSCORE over three periods for firms that ever employed a suspect executive (suspect firms). These periods are: firm-years before suspect executive arrival
(before), firm-years during suspect executive tenure (during), and periods subsequent to suspect executive departure (after).
sample. Specifically, I include only those firms, based on the method in Bebchuk et al. (2010), that have at some point
granted a non-scheduled stock option (backdated or not backdated) in the pre-SOX sample period (Option Grant Firms).
In columns (3) and (4), I add to this initial sample any other firms from my primary exercise-and-hold sample (Option Grant
Firms and Exercise and Hold Firms) to allow for better comparisons with my primary sample. In columns (5) and (6), I include
all Compustat firms meeting my data requirements (All Firms) to ensure my findings are robust to a broader sample.
Across all six columns, BACKDATE_EXEC is never statistically significant. In addition, SUSPECT_EXEC is consistently positive
and significant when included in the regression. NON_SUSPECT_EXEC is negative in column (2) and positive in columns (4)
and (6) but is statistically significant only in column (6). The results of this counter factual test provide evidence that my
main results are driven by questionable tax transactions and not just questionable transactions in general.
5.2. Assessing the timing of differences in tax sheltering
In tests of H1, the coefficient on SUSPECT_EXEC captures the average difference in the level of tax sheltering when suspect
executives are present relative to periods when they are not. It could be useful to understand whether the timing of diffe-
rences in TSSCORE plays a role in my results. I begin with a simple plot presented in Fig. 2 of average TSSCORE realizations
for suspect firms across periods before, during, and after suspect executive tenure. This univariate analysis suggests that
the inferred probability of tax sheltering increases after suspect executives arrive and also decreases after suspect execu-
tives depart firms. Furthermore, these changes relative to firm-years with suspect executive presence appear similar in
magnitude.24 To test whether these magnitudes are still comparable and statistically significant in regression analysis,
I replace SUSPECT_EXEC in Model (1) with two time-period indicator variables that capture periods before suspect executive
arrival (SUSPECT_BEFORE) and periods after suspect executive departure (SUSPECT_AFTER). As with Model (1), I continue to
include firm fixed-effects, thus allowing SUSPECT_BEFORE and SUSPECT_AFTER to capture differences in tax sheltering before
and after suspect executive arrival respectively, relative to periods with suspect executive presence. Because prior research
documents time trends of increasing tax aggressiveness (Desai and Dharmapala, 2009; Plesko, 2004; Yin, 2003), these tests
provide some additional comfort that my results are not solely attributable to overall time trends. In unreported results,
I find that the inferences from Fig. 2 are robust to multivariate regression analysis. Specifically coefficients on SUSPECT_
BEFORE and SUSPECT_AFTER are both negative and significant. An F-test provides no support for rejection of the null that
these two coefficients are equal. This result suggests that the increase in tax sheltering following a suspect executive arrival
is comparable to the decrease in tax sheltering following a suspect executive departure and that overall time trends are not
driving my results.
5.3. Effective tax rates
To determine whether suspect executives are also associated with effective tax rates, I run Model (1) with cash effective
tax rates (CETR) and GAAP effective tax rates (ETR) as dependent variables. Though CETR and ETR are likely noisy measures of
24
In untabulated tests, I examine whether firm performance decreases subsequent to suspect executive departure. I find no evidence of a statistically
significant relationship between periods after suspect executive departure and firm performance.
tax sheltering, testing for the influence of suspect executives on effective tax rates helps compare with prior research
and understand whether suspect executives are equally concerned about cash and GAAP effective tax rates. Because I find
a positive association between suspect executives and tax sheltering, tests with effective tax rates could also help illustrate
whether tax sheltering by suspect executives results in cash tax savings, GAAP tax savings, or both. CETR is constructed as
cash taxes paid divided by pretax income less special items. Similarly ETR is GAAP tax expense divided by pretax income less
special items.
I follow two approaches when selecting my sample for testing. In the first approach I maintain the same sample used
in my primary tax sheltering tests by winsorizing both CETR and ETR at 0 and 1 consistent with Chyz et al. (2013). This
approach allows for comparability between tests but does not adjust for negative denominators as is common in related
research (Dyreng et al., 2010). In the second approach, I set CETR and ETR to missing if pretax income less special items is less
than zero. I find similar results using either approach. Specifically, I find negative and significant associations between CETR
and two of the three suspect executive specifications and none of the non-suspect executive specifications.25 These results
provide some evidence that suspect executive presence is negatively associated with cash effective tax rates. Given my
findings in Table 3, CETR results could be the outcome of more sheltering by suspect executives, or it could be evidence of a
tendency towards a broader set of tax avoidance activities by suspect executives.
The magnitude of the increase in suspect executive option portfolio wealth derived from my estimated firm-level cash
tax savings appears economically meaningful. Using the method in Core and Guay (2002) I collect the option portfolio deltas
for the executives in my sample with available data. Option deltas measure the change in option portfolio value for a
1percent change in stock price. I assume that the estimated annual cash tax savings at the firm level, net of the estimated tax
loss from backdated stock option exercises, is capitalized into firm value using a dividend discount model.26 Using estimates
of cash tax savings derived from my second sample approach that drops firms with negative pretax income, the average
suspect executive realizes an $83,310 annual increase in stock option portfolio value. Under estimates derived from my first
sample approach, which preserves all firm years, the average annual increase in suspect executive stock option portfolio
value is $34,812.
I find positive and significant associations between ETR and the number of suspect executives present at a given firm
(NUM_SUSPECT_EXEC) and the presence of non-suspect executives (NON_SUSPECT) but no statistically significant associa-
tions with other suspect or non-suspect executive specifications. Given the results in Table 3, this result could mean that tax
sheltering by suspect executives is not necessarily focused on GAAP effective tax rate savings. One interpretation of this
assertion is that suspect executives are not aggressive for financial reporting purposes. Taken with the negative association
between suspect executives and cash effective tax rates, the lack of a reliable association with GAAP effective tax rates also
could be the result of suspect executives minimizing scrutiny and political costs stemming from low GAAP effective tax
rates. This notion is not without precedence. In his article discussing the disconnect between Apple's relatively low cash
effective tax rate (approximately 13 percent) and their relatively high GAAP effective tax rate (24 percent), Martin Sullivan
implies that Apple is potentially upwardly managing their effective tax rate to minimize negative publicity with groups like
Citizens for Tax Justice and US Uncut.27 Hanlon and Heitzman (2010) also note that contrary to the GAAP effective tax rate,
the cash effective tax rate measure uses a numerator that is affected by tax deferral strategies. Finding results in tests of CETR
and not in tests of ETR could therefore be evidence of suspect executive participation in tax shelter strategies that defer taxes
that would not impact levels of ETR.28
6. Conclusion
Using a proxy for executives' personal tax aggressiveness adapted from Dhaliwal et al. (2009), I study whether executives
who are aggressive in personal tax matters (suspect executives) influence their firms' tax policy. I use the Thomson Financial
Insiders Database to detect the presence of these executives and construct a sample of firm-years that allows me to evaluate
the impact of personally tax aggressive executive presence on corporate tax policy relative to periods when they are absent.
I focus on the most aggressive of tax policy choices, tax sheltering, and document that the presence of personally tax
aggressive executives is associated with the aggressiveness of some firms' tax policy choices. My proxy for personally tax
aggressive executives is robust to counter-factual testing further suggesting that this executive trait is an important factor
for the aggressiveness of firms' tax policy choices. My findings are consistent with the extant research in managerial
organization and evolving research in accounting and finance that suggests executives' individual characteristics lead to
firm-level economic outcomes. I also find that sheltering tends to increase firm value somewhat in the presence of suspect
25
Under the first approach, coefficient estimates and t-statistics are as follows: SUSPECT_EXEC ( 0.023, t-stat ¼ 1.76), SUSPECT_RATIO ( 0.040, t-stat¼
2.06), NUM_SUSPECT ( 0.011, t-stat¼ 1.19). Under the second approach, coefficient estimates and t-statistics are as follows: SUSPECT_EXEC ( 0.029, t-stat¼
1.79), SUSPECT_RATIO ( 0.048, t-stat¼ 1.75), NUM_SUSPECT ( 0.014, t-stat¼ 1.20).
26
To estimate the pricing of the net cash tax savings using a dividend discount model I need to include estimates of discount and growth rates. I use
the average cost of equity capital estimates from Dhaliwal et al. (2006) plus the 10-year Treasury bond yield on June 30th of each year to proxy for the
discount rate. I include the historical S&P earnings growth rate to proxy for the expected growth rate. I estimate annual cash tax savings as pretax income
multiplied by the coefficient loadings on SUSPECT_EXEC from untabulated tests described in footnote 25.
27
News and Analysis – Economic Analysis. February 13, 2012.
28
In untabulated results of firm-value Model (2) tests, I replace TSSCORE with CETR and generate inferences similar to those reported in Table 4.
executives but decreases firm value somewhat when suspect executives are not present. This result could reflect suspect
executive expertise and ability with respect to tax complexities. The presence of suspect executives or tax sheltering alone,
however, is not associated with firm value.
Since tax sheltering activity is rarely disclosed, my study relies on the ability of my chosen proxy to quantify the
probability that an executive is making aggressive tax policy choices. In addition, regulatory intervention after the Sarbanes-
Oxley Act of 2002 led to changes in the way executives reported stock option exercises to the SEC, thereby precluding
identification of suspect executives using stock option exercise data after August 29, 2002. As executives identified as
suspect retire, move to privately-held firms not captured in Compustat, or move to non-executive roles, the power of my
suspect executive identification technique is likely to wane in later sample years. Despite these limitations, my study
remains the first to directly identify a personal executive trait that is associated with aggressive corporate tax policy choices.
Appendix A. Tax sheltering score (TSSCORE) – construction and validity testing
The variable TSSCORE used in this paper was supplied by Petro Lisowsky. It is a measure of the inferred probability that a
firm is involved in tax sheltering activities. The construction of the TSSCORE is detailed in Lisowsky (2010). Below I provide
more background on how the measure is collected and tests completed by Lisowsky (2010) to support the measure's
validity. Readers wanting additional detail should consult Lisowsky (2010).
Lisowsky (2010) merges proprietary data on publicly traded Subchapter C Corporations obtained from the Internal
Revenue Service (IRS) and IRS Office of Tax Shelter Analysis (OTSA) with the Compustat Database. OTSA provides
information on whether and when a firm participated in a tax shelter. Specifically, Lisowsky (2010) accesses Form 8886
“Reportable Transactions Disclosure Statement” that is required to be filed with a firm's tax return. This form includes
information on the use of listed transactions, which are specifically identified by the IRS as abusive tax shelters, as well as
non-listed transactions, which exhibit features of tax shelter transactions that the IRS is interested in obtaining more details.
With this data, Lisowsky (2010) generates a model using publicly available (but not necessarily machine readable) financial
statement information to estimate the likelihood of tax sheltering for a given firm-year. Specifically, he runs a pooled, cross-
sectional logistic regression model by firm f in year t:
ln ½P TaxShelter =1 P TaxShelter ¼ αþ βX þ ε ðA1Þ
where PTaxShelter ¼[1/1þe (α þ βX þ ε)]¼probability that the firm engages in a tax shelter, and X is a vector of control variables
including year and industry fixed effects. Lisowsky (2010) finds that tax shelter likelihood is positively related to subsidiaries
located in tax havens, foreign-source income, inconsistent book-tax treatment, litigation losses, profitability, size, and the
use of tax shelter promoters; and negatively related to leverage. Fitted values using the coefficient estimates generated by
this model are the inferred estimated probability of tax sheltering realizations, or TSSCORE, used in my study.
To examine the validity of this model, Lisowsky (2010) performs several validity tests. In the first, Lisowsky (2010)
obtains a set of firm-years with known tax shelter participation that was not part of his original sample (Shelter_Valid1).
He then matches, one-to-one, a group of control firm-years from different firms (again, not part of his original sample)
that have not participated in a tax shelter. For both sets of firm-years Lisowsky (2010) generates TSSCORE measures and then
runs a logistic regression of Shelter_Valid1 on TSSCORE. Results of this logistic regression indicate that TSSCORE is useful in
identifying out-of-sample tax shelters in a one-to-one matched sample setting.
In the second validity test Lisowsky (2010) calculates the area under the Receiver Operator Characteristic (ROC) curve
(see Hosmer and Lemeshow, 2001) as a way to evaluate the both the validity and predictive ability of his logistic tax shelter
probability model on out-of-sample period observations. Using the one-to-one matched sample described above, Lisowsky
(2010) finds that his model has strong discriminatory power to identify tax shelter firms.
In the third set of tests, Lisowsky (2010) replicates the analyses described above on a larger sample that matches out-of-
sample tax shelter participants to 10 (instead of 1) non-shelter out-of-sample firms. He continues to find support for the
validity of his model.
In the last set of validity tests, Lisowsky (2010) re-estimates his logistic model (1) above, dropping one year of data to
preserve a “holdout” sample. Lisowsky (2010) calculates TSSCORE for the holdout sample of firms based on the coefficient
estimates generated by the re-estimated model using the reduced sample. He finds that TSSCORE is useful in identifying
hold out sample tax shelters. In ROC tests with this holdout sample, Lisowsky (2010) again finds that his model has strong
discriminatory power to indentify tax shelter firms.
References
Aboody, D., Hughes, J., Liu, J., Su, W., 2008. Are executive stock option exercises driven by private information? Review of Accounting Studies 13, 551–570.
Aboody, D., Kaznick, R., 2000. CEO stock option awards and corporate voluntary disclosures. Journal of Accounting and Economics 29, 73–100.
Adams, R.B., Santos, J.A.C., 2006. Identifying the effect of managerial control on firm performance. Journal of Accounting and Economics 41, 55–85.
Bamber, L., Jiang, J.(X.), Wang, I.Y., 2010. What's my style? The influence of top managers on voluntary corporate financial disclosure. Accounting Review 85,
1131–1162.
Bebchuk, L.A., Grinstein, Y., Peyer, U., 2010. Lucky CEOs and lucky directors. Journal of Finance 65, 2363–2400.
Bertrand, M., Schoar, A., 2003. Managing with style: the effect of managers on firm policies. Quarterly Journal of Economics 118, 1169–1208.
Blaylock, B., 2011. Do managers extract economically significant rents through tax aggressive transactions? Working paper. Oklahoma State University.
Brown, L.D., Caylor, M.L., 2006. Corporate governance and firm valuation. Journal of Accounting and Public Policy 25, 409–434.
Brown, J.L., Drake, K.D., 2011. Network ties among low tax firms. Working paper. Arizona State University.
Chatterjee, A., Hambrick, D.C., 2007. It's all about me: narcissistic chief executive officers and their effects on company strategy and performance.
Administrative Science Quarterly 52, 351–386.
Chen, S., Chen, X., Cheng, Q., Shevlin, T., 2010. Are family firms more tax aggressive than non-family firms? Journal of Financial Economics 95, 41–61.
Chyz, J., Leung, W., Li, O., Rui, O., 2013. Labor unions and tax aggressiveness. Journal of Financial Economics 108, 675–698.
Cicero, D.C., 2009. The manipulation of executive stock option exercise strategies: information timing and backdating. Journal of Finance 64, 2627–2663.
Core, J., Guay, W., 2002. Estimating the value of employee stock option portfolios and their sensitivities to price and volatility. Journal of Accounting
Research 40, 613–630.
Cronqvist, H., Makhija, A.K., Yonker, S.E., 2012. Behavioral consistency in corporate finance: CEO personal and corporate leverage. Journal of Financial
Economics 103, 20–40.
Department of the Treasury, 2009. The Problem of Corporate Tax Shelters: Discussion, Analysis and Legislative Proposals, 7-1999.
Desai, M., Dharmapala, D., 2006. Corporate tax avoidance and high-powered incentives. Journal of Financial Economics 79, 145–179.
Desai, M., Dharmapala, D., 2009. Corporate tax avoidance and firm value. Review of Economics and Statistics 91, 537–546.
Dhaliwal, D., Heitzman, S., Li, O., 2006. Taxes, leverage, and the cost of equity capital. Journal of Accounting Research 44, 691–723.
Dhaliwal, D., Erickson, M., Heitzman, S., 2009. Taxes and the backdating of stock option exercise dates. Journal of Accounting and Economics 47, 27–49.
Dyreng, S., Hanlon, M., Maydew, E., 2008. Long-run corporate tax avoidance. Accounting Review 83, 61–82.
Dyreng, S., Lindsey, B., 2009. Using financial accounting data to examine the effect of foreign operations located in tax havens and other countries on U.S.
multinational firms' tax rates. Journal of Accounting Research 47, 1283–1316.
Dyreng, S., Hanlon, M., Maydew, E., 2010. The effects of executives on corporate tax avoidance. Accounting Review 85, 1163–1189.
Frank, M.Z., Goyal, V.K., 2007. Corporate leverage: how much do managers really matter? Working paper. University of Minnesota.
Graham, J.R., Harvey, C.R., Puri, M., 2012. Managerial attitudes and corporate actions. Journal of Financial Economics 109, 103–121.
Greene, W.H., 2003. Econometric Analysis, 5th edition Prentice Hall, Englewood Cliffs, NJ.
Gupta, S., Newberry, K., 1997. Determinants of variability in corporate effective tax rates: evidence from longitudinal data. Journal of Accounting and Public
Policy 16, 1–34.
Hambrick, D., Mason, P., 1984. Upper echelons: the organization as a reflection of its top managers. Academy of Management Review 9, 193–206.
Hambrick, D., 2007. Upper echelons theory: an update. Academy of Management Review 32, 334–343.
Hanlon, M., Slemrod, J., 2009. What does tax aggressiveness signal? Evidence from stock price reactions to news about tax aggressiveness. Journal of Public
Economics 93, 126–141.
Hanlon, M., Heitzman, S., 2010. A review of tax research. Journal of Accounting and Economics 50, 127–178.
Heron, R.A., Lie, E., 2007. Does backdating explain the stock price pattern around executive stock option grants? Journal of Financial Economics 83, 271–298.
Higgins, D.M., Omer, T.C., Phillips, J.D., 2013. The influence of a firm's business strategy on its tax aggressiveness. Working paper. University of Connecticut.
Hosmer, D., Lemeshow, S., 2001. Applied Logistic Regression, 2nd edition John Wiley & Sons, Inc., New York, NY.
Jin, L., Kothari, S.P., 2008. Effect of personal taxes on managers' decision to sell their stock. Journal of Accounting and Economics 46, 23–46.
Lie, E., 2005. On the timing of CEO stock option awards. Management Science 51, 802–812.
Lisowsky, P., 2010. Seeking shelter: empirically modeling tax shelters using financial statement information. Accounting Review 85, 1693–1720.
McDonald, R., 2004. Is it optimal to accelerate the payment of income tax on shared-based compensation? Working paper. Northwestern University.
McGuire, S., Wang, D., Wilson, R., 2011. Dual class ownership and tax avoidance. Working paper. Texas A&M University.
Mills, L., Erickson, M., Maydew, E., 1998. Investments in tax planning. Journal of the American Taxation Association 20, 1–21.
Phillips, J., 2003. Corporate tax-planning effectiveness: the role of compensation-based incentives. Accounting Review 78, 47–874.
Plesko, G., 2004. Corporate tax avoidance and the properties of corporate earnings. National Tax Journal 57, 729–737.
Rego, S., Wilson, R., 2012. Equity risk incentives and corporate tax aggressiveness. Journal of Accounting Research 50, 775–810.
Scholes, M., Wolfson, M., 1992. Taxes and Business Strategy. Prentice Hall, Englewood Cliffs, NJ.
Scholes, M., Wolfson, M., Erickson, M., Maydew, E., Shevlin, T., 2009. Taxes & Business Strategy, 4th edition Prentice Hall Publishing, New York.
Shackelford, D., Shevlin, T., 2001. Empirical tax research in accounting. Journal of Accounting and Economics 31, 321–387.
Shevlin, T., 2002. Symposium on Corporate Tax Shelters, Part II: Commentary: Corporate Tax Shelters and Book-tax Differences. New York University Tax
Review 55 Tax L. Rev 427.
Shevlin, T., 2007. The future of tax research: from an accounting professor's perspective. Journal of the American Taxation Association 14, 58–79.
Weisbach, D.A., 2002. Ten truths about tax shelters. Tax Law Review 55, 325–384.
White, H., 1980. A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica 48, 817–838.
Wilson, R., 2009. An examination of corporate tax shelter participants. Accounting Review 84, 969–999.
Wooldridge, J.M., 2002. Econometric Analysis of Cross Section and Panel Data. The MIT Press, Cambridge, MA.
Yermack, D., 1997. Good timing: CEO stock option awards and company news announcements. Journal of Finance 52, 449–476.
Yin, G., 2003. How much tax do large corporations pay? Estimating the effective tax rates of the S&P 500. Virginia Law Review 89, 1793–1856.

Chyz (2013)

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Chyz (2013)

Uploaded by

Copyright:

Available Formats

Journal of Accounting and Economics 56 (2013) 311–328

Contents lists available at ScienceDirect

Journal of Accounting and Economics

Personally tax aggressive executives and

2. Related literature and hypothesis development

2.1. Tax motivation for stock option exercise backdating

2.2. Executive characteristics and economic outcomes

2.3. Firm value and tax sheltering

H2a. Suspect executives are not associated with firm value.

3.1. Sample construction and executive identification

3.2. Tax sheltering tests of Hypothesis 1

3.3. Firm value tests of Hypotheses 2a and 2b

All other variables as defined in Model (1).

3.4. Data, sample and descriptive statistics

Panel A: Industry distribution

Suspect firms Non-suspect firms

Number Percent Number Percent Number Percent

Consumer Nondurables 479 6.12 60 4.28 211 6.63

Panel B: Full sample descriptive statistics

Suspect firms – all years Non-suspect firms – all years

Mean Median Std dev. Mean Median Std dev.

TSSCORE 0.749 0.894nnn 0.306 0.757 0.929 0.314

With executive presence Without executive presence

Mean Median Std dev. Mean Median Std dev.

TSSCORE 0.781nnn 0.919nnn 0.289 0.683 0.799 0.329

With executive presence Without executive presence

Mean Median Std dev. Mean Median Std Dev

With executive presence Without executive presence

Mean Median Std dev. Mean Median Std Dev

SUSPECT_EXEC 0.000 0.000 0.000 0.000 0.000 0.000

4.1. Tax sheltering regression tests

Predicted sign (1) (2) (3) (4) (5) (6)

SUSPECT_EXEC þ 0.057 0.034 – – – –

F-test: β1 ¼β2 (p-value) 0.0211 0.0271 0.0180 0.0496 0.0054 0.0130

TSSCOREit ¼ β0 þ β1 SUSPECT_EXEC it þ β2 NON_SUSPECT_EXEC it þ β3 NOLit þ β4 RDit

4.2. Firm value regression tests

Dependent variable TOBINSQ

(1) (2) (3)

F-test: β4 ¼β5 (p-value) 0.0563 0.0944 0.0282

Firm, Industry & Year Fixed Effects Yes Yes Yes

TOBINSQ it ¼ β0 þ b1 SUSPECT_EXEC it þ β2 NON_SUSPECT_EXEC it þ β3 TSSCOREit

5. Additional analysis and robustness testing

5.1. Option grant backdating

Dependent variable TSSCORE

(1) (2) (3) (4) (5) (6)

BACKDATE_EXEC 0.0102 0.005 0.010 0.006 0.015 0.010

5.2. Assessing the timing of differences in tax sheltering

5.3. Effective tax rates

Appendix A. Tax sheltering score (TSSCORE) – construction and validity testing

You might also like