Tax Reporting Behavior under Audit Certainty

BENJAMIN C. AYERS, University of Georgia

JERI K. SEIDMAN,† McIntire School of Commerce, University of Virginia

ERIN M. TOWERY, University of Georgia

† Corresponding author: McIntire School of CommerceUniversity of VirginiaP.O. Box 400173Charlottesville, VA, 22904434-294-8976 (office)jseidman@virginia.edu.

The authors appreciate helpful comments from Christine Cheng (discussant), Lisa De Simone, Danielle Higgins (discussant), Lillian Mills, Thomas Omer, Jeffrey Pittman (editor), George Plesko (discussant), two anonymous referees, and workshop participants at the College of William & Mary, University of Houston, University of Kansas, University of Virginia McIntire School of Commerce, the 2014 American Accounting Association Annual Meeting, the 2015 IRS-TPC Research Conference, and the 2015 University of Illinois Symposium on Tax Research. We also thank John Miller and Barbara Hecimovich for providing information about the IRS Coordinated Industry Case (CIC) program.

The Internal Revenue Service (IRS) provided confidential tax information to Towery pursuant to provisions of the Internal Revenue Code that allow disclosure of information to a contractor to the extent necessary to perform a research contract for the IRS. None of the confidential tax information received from the IRS will be disclosed in this treatise. Statistical aggregates will be used so that a specific taxpayer cannot be identified from information supplied by the IRS. All opinions are those of the authors and do not reflect the views of the IRS.

ABSTRACTThis study uses a confidential dataset of firms assigned to the Internal Revenue Service (IRS)’s Coordinated Industry Case (CIC) program to examine the effect of audit certainty on firms’ tax reporting behavior. We first model the determinants of assignment to the program. Though the ability and incentive to avoid taxes are related to CIC assignment, we find that the IRS assigns firms primarily based on size and complexity. We then test whether audit certainty has a detectable effect on tax payments. Our results show that tax payments do not change when firms enter the CIC program, suggesting the CIC program does not have higher deterrence or enforcement effects relative to the IRS’s standard selection and audit process for large corporations not included in the CIC program. However, supplemental analysis suggests that audit certainty does alter managers’ expectations regarding future tax payments. Our paper provides new empirical evidence on the strategic game between the taxpayer and the tax authority and has important implications for tax authorities as they consider the costs and benefits of certain audit programs.

Keywords: Tax examination; Internal Revenue Service; Coordinated Industry Case program; Tax avoidance

JEL Classification: H25; M20; M41

1. Introduction

This study investigates the effect of tax audit certainty on tax reporting behavior.1

Understanding how audit certainty affects taxpayers is important because many of the largest

firms in the U.S. are assigned the Internal Revenue Service (IRS)’s Coordinated Industry Case

(CIC) program, where the likelihood of audit examination is 100 percent. We estimate that firms

assigned to the CIC program account for between 65 and 70 percent of the U.S. market

capitalization in 2011. Further, the IRS invests a sizeable portion of its resources in these efforts.

The mean (median) CIC audit used approximately 2,500 (1,700) IRS-personnel-hours during our

sample period, or the equivalent of more than one full-time employee per CIC audit each year.

We also observe that only about one-third of tax returns filed in the year preceding CIC

assignment are audited.2 Thus, the IRS commitment to a program of audit certainty is both an

economically substantial resource allocation decision for the U.S. government and a nontrivial

increase in audit probability for assigned firms. Studying the effect of audit certainty on tax

payments has been difficult because corporate taxpayers are not required to publicly disclose

whether they face certain audit. We overcome this data limitation using a confidential dataset of

corporate taxpayers in the CIC program.

The effect of audit certainty on tax payments is not clear ex ante. Over the past several

decades, both theoretical and empirical studies have documented that the risk of tax audit

examination affects tax reporting behavior. These studies generally predict that taxpayers enter

into fewer uncertain tax positions to reduce their probability of audit when tax uncertainty

increases the probability of audit selection.

1 We define tax audit certainty as a 100 percent likelihood that a tax return is subject to audit, and we define tax reporting behavior as the amount of tax reported to the revenue authority conducting the audit. 2 We use IRS audit examination data to observe hours per CIC audit and the audit rate before CIC assignment.

2

However, because tax uncertainty does not affect the probability of tax audit examination

for taxpayers facing audit certainty, these firms likely have different incentives to engage in

uncertain tax avoidance.3 On one hand, taxpayers facing audit certainty could have less incentive

to claim uncertain tax positions if the increased detection risk lowers the expected benefit of tax

uncertainty such that some positions are no longer value‐creating. This would be consistent with

the negative relation between audit selection risk and tax avoidance documented at lower points

on the audit probability spectrum (Hoopes, Mescall, and Pittman 2012).

On the other hand, because taxpayers facing certain audit have no incentive to reduce

their tax planning to avoid IRS audit selection, audit certainty could result in increased tax

uncertainty. Consistent with this intuition, Slemrod, Blumenthal, and Christian (2001) find that

high-income individual taxpayers reported lower taxable income amounts when told before filing

their tax return that they will be audited with certainty. The authors conjecture that taxpayers

claim additional tax benefits to create a more aggressive starting point for negotiations with the

tax authority, with the goal of minimizing their tax liability and under the assumption that the

audit will not detect and punish all tax avoidance. In sum, how tax audit certainty affects tax

reporting behavior is an empirical question. To our knowledge, the only existing empirical

evidence on the effect of corporate tax audit certainty is reported in Hanlon, Mills, and Slemrod

(2007). Although the study primarily focuses on determinants of IRS proposed audit

adjustments, the authors in a supplemental test find that CIC and non-CIC firms have similar

GAAP effective tax rates (computed using current tax expense from the financial statements)

using a levels analysis. In contrast, our study uses both levels and changes analyses to investigate

how CIC assignment impacts initial tax return filing positions, initial filing positions adjusted for

3 While the probability of audit is 100 percent for firms in the CIC program, the probability of audit for any particular transaction remains less than 100 percent. Thus, tax reporting behavior could continue to affect the audit risk of any particular transaction.

3

future settlements, and taxes paid reported in the financial statements.

Before testing our research question, we first analyze the determinants of assignment to

the CIC program using selection factors outlined in the Internal Revenue Manual (IRM), which

capture various size and complexity determinants. Though other research cites size and

complexity as determinants of CIC assignment (Mills 1998; Hanlon et al. 2007), these statements

are based on the IRM listed factors rather than on empirical tests. Our analysis tests these

statements and serves two purposes: (i) to determine whether CIC program assignment is

primarily mechanical in nature based upon the IRM listed factors or whether factors associated

with tax avoidance (beyond size and complexity) also contribute to assignment, and (ii) to

provide researchers without access to CIC assignment data a model of audit certainty.

We find that many of the IRM listed factors are positively associated with assignment

into the CIC program, with gross receipts being the most economically significant size

determinant across various specifications. When we include factors known to affect firms’

incentives or ability to avoid taxes (such as research and development expenses, excess stock

option deductions, and net operating loss carryforwards) as potential determinants, we estimate

that the associations between a number of these factors and CIC assignment are statistically

significant. However, their inclusion does not dramatically improve the fit of the model. These

results collectively suggest that although inclusion in the CIC program is associated with firms’

incentives or ability to avoid taxes, the CIC assignment decision is primarily based on firm size

and complexity.4

Next, we study the deterrence and enforcement effects of audit certainty. The taxpayer’s

initial federal tax rate is the total federal tax liability that taxpayers report on their initial tax

4 Importantly, these results do not imply that the CIC assignment decision is purely mechanical. Even though a firm will generally not be assigned to the CIC program if it has less than 12 CIC points, we confirm that not all firms with at least 12 points are assigned to the program, consistent with limited IRS resources.

4

return for the year, deflated by the taxpayer’s worldwide pre-tax income for the year; to the

extent audit certainty has a deterrence effect on tax avoidance, the initial federal tax rate should

increase. The advantage of using the taxpayer’s initial federal tax rate to evaluate deterrence

effects is that it captures initial tax payments to the tax authority and is not confounded by tax

reserves associated with current tax return positions, tax settlements paid in the current year for

prior tax years, or future tax settlements associated with the current tax year. We measure the

combined effect of deterrence and enforcement using: (i) the taxpayer’s initial federal tax rate

adjusted for settlements, and (ii) the cash effective tax rate (ETR).

We use both a levels approach and a changes approach to investigate the effects of audit

certainty. We implement our levels analysis using a pooled sample of firms assigned to the CIC

program and firms not assigned the CIC program over the period 2000 to 2011. Using a variety

of different scalars and fixed effects, we are unable to detect a statistically significant difference

in initial federal tax rates, adjusted federal tax rates (including settlements), or cash effective tax

rates between firms assigned to the CIC program and firms not assigned to the CIC program.

To implement our changes analysis, we identify 405 corporate taxpayers that are first

assigned to the CIC program between 2000 and 2011 (“newly-assigned firms”). We then

construct two samples of propensity-matched control firms: (i) firms never assigned to the

program between 2000 and 2011 (“non-assigned firms”); and (ii) firms in the program both

before and after newly-assigned firms enter the CIC program (“long-assigned firms”). The

matched sample design allows us to not only compare the tax payments of a firm to itself before

and after the change in its CIC program status, but to also compare its tax payments with the tax

payments of a firm that does not experience a change in CIC program status.

5

We find that, post-assignment, the initial federal tax rates, the adjusted federal tax rates,

and the cash effective tax rates of newly-assigned firms are not statistically different than those

of the matched sample of non-assigned firms. Further, post-assignment, none of the tax payment

rates of newly-assigned firms are statistically different than the tax payment rates of the matched

sample of long-assigned firms. Thus, our results suggest that the CIC program does not have

higher deterrence and enforcement effects relative to the IRS’s standard selection and audit

process for large corporations not included in the CIC program.

Even if audit certainty does not change firms’ tax reporting behavior, audit certainty

could revise taxpayer’s expectations about future tax settlements. Therefore, in supplemental

analysis, we consider whether audit certainty affects managers’ expectations regarding future tax

payments. Using both a changes and a levels analysis, we find that newly-assigned firms report

higher financial statement reserves for current year tax positions relative to both non-assigned

and long-assigned firms, suggesting that audit certainty does impact financial reporting for

income taxes. Our primary results suggest that the increased reserves do not represent an

increase in uncertain tax avoidance. More plausible explanations include: (i) firms systematically

underestimated their likelihood of sustaining a position prior to CIC assignment and

subsequently updated their expectations based on “learning” in the audit process, and/or (ii)

firms incorporated audit likelihood into their determination of reserves prior to CIC assignment,

which would be inconsistent with the U.S. GAAP requirement that firms assume audit certainty

with respect to each uncertain tax position.

Our study expands the literature in multiple ways. First, our model of CIC determinants

provides researchers without access to tax return data with a proxy for the audit risk of large,

6

publicly-traded companies.5 Second, we provide evidence that although inclusion in the CIC

program is associated with firms’ incentives or ability to avoid taxes, this substantial resource

allocation decision is primarily based on firm size and complexity. Third, to our knowledge, our

study is the first to analyze the deterrence and enforcement effects of audit certainty. In doing so,

we further our understanding of the strategic game between the taxpayer and the tax authority.

We do not view our results as diverging from prior research that documents a negative relation,

on average, between audit probability and tax avoidance (Mills and Sansing 2000; Hoopes et al.

2012). Our study focuses on the top end of the audit probability continuum, where tax reporting

behavior does not affect audit selection.

Our findings are also important to tax authorities. Understanding how audit risk affects

tax reporting behavior informs the IRS as it designs and implements new audit approaches. CIC

audits consume a substantial portion of IRS Large Business and International (LB&I) audit

resources. Whether and how firms alter tax reporting behavior within the CIC program informs

the cost-benefit assessment of the program.6 Our results suggest that the CIC program does not

have higher deterrence and enforcement effects than the IRS’s non-CIC audit process for large

corporations. Although not anticipated ex ante, these results are consistent with the IRS’s recent

announcement to potentially modify the CIC program to incorporate an as-yet-unspecified risk-

based approach that will likely leave many large corporate taxpayers uncertain of whether to

expect an audit.

5 However, given that we find no significant difference in the initial federal tax rates, the adjusted federal tax rates, or the cash effective tax rates for CIC firms relative to non-CIC firms, we caution researchers to consider whether IRS audit risk is expected to have the same behavioral effect at the top end of the continuum as the effect documented at lower points of the continuum. 6 Scott (1991) outlines five factors that influenced the creation of the CIC program, which we detail in Section 2. Our research tests only one of these factors—whether audit certainty via the CIC program raises government revenue. Thus, the cost-benefit analysis we discuss here cannot speak to the overall success of the CIC program.

7

2. Background and hypothesis development

Coordinated Industry Case (CIC) program

Many of the largest corporate taxpayers in the United States are subject to certain audit as

part of the Internal Revenue Service’s CIC program. Between 500 and 1,500 taxpayers are

assigned to the CIC program in a given year, though the number varies over time. The IRS

implemented the CIC program, formerly the Coordinated Examination Program, in the 1960s.

Scott (1991) provides a detailed history of IRS large case audits. He describes the primary

factors that led to the change in large case enforcement in the 1960s as: (i) increasing complexity

of tax law; (ii) increased demand for government revenue; (iii) increasing size and complexity of

U.S. businesses; (iv) limited IRS resources; and (v) inefficient large case audits. Thus, the

program was intended to address difficulties encountered when auditing large U.S. firms, but

factors (ii) and (iv) indicate that increasing tax revenues in a setting of limited IRS resources was

also a major motivation for the program.

For CIC firms, a team from the IRS LB&I division spends a substantial amount of time in

the taxpayer’s primary place of business throughout the year. The IRS team consists of the

examination team manager, field agents, industry specialists, and subject matter experts in areas

such as engineering, excise taxes, and employment. Although the number of team members

varies depending on the size and complexity of the taxpayer, CIC audit teams generally provide

more in-depth audits than traditional IRS audits. In our sample, the mean (median) number of

hours for CIC audits is 2,500 (1,700) audit hours, while the mean (median) number of hours for

non-CIC audits is 270 (150) audit hours, suggesting non-CIC audits are not as resource-

intensive. Among audited firms with at least $250 million in assets, the average proposed

adjustment for CIC firms is $12.4 million, while the average proposed adjustment for non-CIC

8

firms is only $781,000. Together, these figures suggest the IRS generates a higher return for

hours spent on CIC audits ($4,960 per CIC audit hour and $2,893 per non-CIC audit hour, on

average).

Firms are assigned to, rather than invited to join, the CIC program.7 Per the Internal

Revenue Manual, this assignment is currently made based on a point system involving seven

main criteria: (i) gross assets; (ii) gross receipts; (iii) operating entities; (iv) number of industries;

(v) total foreign assets; (vi) related transactions; and (vii) foreign taxes paid (Internal Revenue

Manual, 4.46.2.5). Each criterion has a point value, and a firm may be assigned to the CIC

program if its total point value is greater than or equal to 12. However, because of limited IRS

resources, not all firms with at least 12 points are assigned to the program.8 Further, firms with a

point value of less than 12 can also be assigned to the CIC program if they are sufficiently

complex to warrant certain audit.

Appendix 1 provides a timeline of CIC program assignment. The IRS makes the CIC

program assignment decision using tax return data: the decision for Year t is made in Year t+1

after the taxpayer has filed its tax return for Year t. The IRS then informs the taxpayer of the

assignment. Thus, the annual report and tax return for Year t are filed before the taxpayer

becomes aware of its CIC assignment, and the annual report and tax return for Year t+1 are filed

after the firm becomes aware of its CIC assignment.9 The CIC team audits the Year t tax return

and often also audits tax returns for multiple years filed prior to CIC assignment.10 Once assigned 7 This contrasts with the IRS Compliance Assurance Program (CAP), where a corporate taxpayer voluntarily agrees to discuss uncertain tax issues with an IRS team prior to filing its annual tax return. See De Simone, Sansing, and Seidman (2013) and Beck and Lisowsky (2013) for evidence on the effect of voluntary audits on tax reporting behavior. 8 In untabulated analysis, we use tax return assets and tax return gross receipts data to confirm that not all firms with 12 or more tax return points are assigned to the program. 9 Untabulated analysis shows that CIC program assignment is gradually increasing in total points, which suggests assignment is not easily predicted. Thus, the firm will file its annual report and tax return for Year t without knowing it will be assigned to the CIC program. 10 For example, a taxpayer will be notified in late 2006 after filing its 2005 tax return that it has been assigned to the CIC program. The CIC team may also audit years prior to 2005 in which the statute of limitations remains open.

9

to the CIC program, a taxpayer typically remains in the CIC program until the audit no longer

requires a team audit approach (their point value falls), or until the taxpayer ceases to exist

because of either bankruptcy or combination with another operating entity.

Related literature

The strategic game between the taxpayer and the tax authority has been studied

extensively in the accounting, economics, and finance literatures. Early models characterize

taxpayer compliance as a function of tax rates, probability of detection and punishment,

penalties, and taxpayer risk-aversion (e.g., Allingham and Sandmo 1972).11 Graetz, Reinganum,

and Wilde (1986) extends these basic models to allow the IRS to condition its audit rules on the

reports it receives from taxpayers. This ‘strategic tax compliance model’ provides different

predictions than earlier models because a taxpayer must now consider how its tax return affects

the probability of audit when it chooses the amount to report. Thus, a high-income taxpayer that

considers reporting low income not only takes into account the tax savings and potential

penalties if caught, but also that reporting low income increases the probability of audit. The

most basic strategic tax compliance model predicts a deterrence effect that leads to a negative

association between audit risk and tax uncertainty, consistent with firms engaging in less tax

uncertainty to reduce their probability of audit because uncertain tax positions increase the

probability of audit (Graetz et al. 1986).

The strategic tax compliance model has been extended in many ways. Our paper is most

closely related to a set of models of enforcement, specifically the probability of detection. For

example, Sansing (1993) introduces certain and verifiable information, which allows the IRS to

11 Many models assume perfect detection, meaning that they assume all uncertain tax positions are detected and only those with strong facts are upheld if the firm is audited. This simplifying assumption does not generalize well to the reality of audits of large corporations with several uncertain tax positions. However, detection risk is likely to increase under certain audit both because the probability of the firm being audited increases to one and because the coordinated nature of these programs allows the tax authority to gain greater knowledge of the taxpayer’s business, thus potentially enabling the tax authority to better identify uncertain tax positions.

10

better focus audits on taxpayers with unverifiable information. Rhoades (1999) allows the audit

decision to be made at the component level and determines that the IRS conditions its second

audit decision on the results of the first component’s audit. Mills and Sansing (2000) allows the

difference between book and taxable income, both of which the IRS observes, to be informative

to the IRS. Their empirical results show that the magnitude of book-tax differences is positively

associated with IRS proposed audit adjustments but not associated with IRS settlements. This

result suggests corporate taxpayers require strong facts to enter into uncertain tax positions that

will generate a book-tax difference because they appreciate that these differences convey

information to the tax authority.

However, for firms facing certain audit, engaging in less tax uncertainty does not reduce

the probability of audit. Though tax reporting behavior may reduce the probability of a particular

line item or transaction being subject to audit, Rhoades (1999) suggests that the probability of a

line item claimed by a certain audit firm being audited is greater than or equal to the probability

of a line item claimed by a non-certain audit firm being audited. In other words, firms facing

certain audit generally have less ability to influence the probability that a particular line item is

audited than firms not facing certain audit. Thus, it is not clear that the familiar deterrence effect

will hold for firms facing audit certainty.

The strategic tax compliance model discussed above provides some insight into tax

reporting behavior in our setting. On one hand, taxpayers could have less incentive to engage in

uncertain tax avoidance if the increased audit probability decreases the expected benefit of tax

uncertainty such that a subset of tax positions are no longer value-creating. On the other hand,

Mills and Sansing (2000) suggests certain audit could increase the incentive to engage in

uncertain tax avoidance. Specifically, because the IRS will audit the firm regardless of the level

11

of tax planning, certain audit firms no longer have incentive to reduce their tax planning to avoid

IRS audit selection. This suggests audit certainty could result in increased tax uncertainty.

This intuition is consistent with results presented in Slemrod et al. (2001), which reports

results of a 1995 experiment by the Minnesota Department of Revenue under which a random

sample of individual taxpayers were told that the returns they were about to file would be

“closely examined.” Relative to the sample of individual taxpayers not told that their returns

would be closely examined, the high-income taxpayers of the “audit certain” sample

significantly decreased their reported tax liability. The authors conjecture that these individual

taxpayers claimed more tax benefits to create a more aggressive starting point for negotiations

with the tax authority with the goal of minimizing their tax liability. This suggests an

enforcement effect but not a deterrence effect of audit certainty. Further, the authors postulate

that this effect is observed in high-income taxpayers, but not low- or middle-income taxpayers,

because high-income taxpayers believe that the final outcome of the certain audit is more

manipulable due to their ability to hire tax professionals.

This logic likely applies to the corporate taxpayers that we study, as firms assigned to the

CIC program tend to be large and/or have complex operations, both of which are expected to be

correlated with the use of professional tax assistance. However, the corporate taxpayers in our

sample have financial reporting obligations that individuals do not, which could cause these two

types of taxpayers to have different tax avoidance preferences. In sum, how tax certainty affects

corporate tax reporting behavior is an empirical question.

Our paper is also related to three recent applied papers regarding the strategic game

between the IRS and corporations. Hoopes et al. (2012), DeBacker, Heim, Tran, and Yuskavage

(2015), and Lennox, Li, Pittman, and Wang (2015) suggest that increased enforcement leads to

12

greater tax payments. Using data on IRS audit probability by firm size over time, Hoopes et al.

(2012) provides evidence that firms facing greater risk of IRS audit report higher cash effective

tax rates relative to firms facing lower risk of IRS audit. However, because CIC program

assignment is not publicly available, the study is unable to examine the effect of certain audit on

taxpayer behavior.

DeBacker et al. (2015) finds that tax payments are lower immediately following an audit,

when the taxpayer is assumed to have lower expectations of another audit. As time passes and

the probability of another audit increases, tax payments increase. Thus, similar to Hoopes et al.

(2012), Debacker et al. (2015) concludes that firms engage in less tax avoidance when audit

selection probability is higher. Finally, Lennox et al. (2015) examines the effect of tax audit

examination on Chinese firms. Their results suggest firms report higher GAAP effective tax rates

and lower book-tax differences after being audited. Our paper differs from these three studies

because we test how taxpayers report when facing audit certainty, an economically important

setting between taxpayers and the tax authority that is fundamentally different from those

previously examined.

Hypothesis development

Although the aforementioned literature provides rich insight into tax reporting behavior,

no studies have modeled the case where audit probability equals one. While we can input

parameters to force the probability of audit to one, prior theoretical models are designed for the

tax authority to have an audit decision (to audit or not to audit). Thus, while our research

question is informed by prior models, we develop our hypothesis primarily based on the

following intuition.

A taxpayer’s expected total tax payment equals the tax liability on their originally filed

13

return plus the expectation of any future settlements:

E[total tax payment] = tax liability as originally filed + prob(audit)*E(settlement)12 (1)

Prior literature primarily considers situations where the probability of the firm being audited is a

function of the tax liability as originally filed:

Prob(audit) = f (tax liability as originally filed, political environment, IRS budget, complexity and size of taxpayer, etc.) (2)

In Equation (2), when the probability of audit selection increases, the taxpayer can increase the

tax liability on the originally filed return (for example, by not entering into the most aggressive

tax positions) to lower the probability of audit selection.

In our setting, however, the probability of audit selection is not a function of the tax

liability as originally reported because Prob(audit) equals one upon assignment to the program. If

a taxpayer believes that an audit reveals complete truth (i.e., perfect detection), the taxpayer may

choose to accept the expected higher tax payment (increase initial filing liability), which likely

decreases the time associated with the audit and eliminates interest and penalties associated with

expected disallowed tax positions. This would be consistent with a deterrence effect of certain

audit.

Alternatively, the taxpayer may not change the level of tax avoidance or even become

more aggressive to begin negotiations with the tax authority from a more favorable starting point

if the taxpayer does not believe in perfect detection risk or that a higher total tax payment is

certain despite a certain audit.13 In this case, audit certainty would not have a deterrence effect

but may still have an enforcement effect if we observe that the tax liability as originally reported

stays constant or decreases but the total tax payment increases. Thus, taxpayers may increase,

12 For simplicity, this model ignores the time value of money.13 Imperfect detection could arise for a number of reasons, including noise in the audit process, not all transactions being audited, or if a nonstrategic model better describes tax reporting behavior.

14

decrease, or not change their level of tax avoidance when the probability of audit increases to

one.

HYPOTHESIS. Audit certainty does not affect tax reporting behavior.

3. Research design

Determinants of CIC assignment

We first study the determinants of CIC program assignment following the IRS’s stated

assignment criteria. This determinants model is intended to examine whether the CIC assignment

process is based primarily on the point system outlined in the Internal Revenue Manual and to

provide a model that other researchers without access to tax return data can use. The Internal

Revenue Manual lists seven criteria to be used in identifying cases for the CIC program. 14 These

criteria encompass various measures of size and complexity. Following IRM Exhibit 4.46.2-2,

we specify the following logistic regression:

CICFirm = α + β*Size + γ*Complexity + ε (3)

where CICFirm equals one if the firm is assigned to the CIC program and zero otherwise.15 We

define the various Size and Complexity measures with a discrete point system similar to that

specified in the Internal Revenue Manual.

Although the IRS uses tax return disclosures to assign points, we use only publicly-

available data to estimate the determinants of CIC prediction in our model specifications so that

future researchers can use the model to estimate CIC assignment. Both total assets (AT) and net

14 Parts of the Internal Revenue Manual were revised in March of 2006. However, the specific changes to the CIC point system (if any) are not disclosed on the IRS website. In addition, FASB Interpretation No. 48, Accounting for Uncertainty in Income Taxes (FIN 48/ASC 740-10) was introduced in 2006, which changed how firms accounted for income tax uncertainty in their public financial statements for years starting after 2006. To address concerns that a change in the point system and/or a change in accounting standards could affect our research design and results, we re-estimate our prediction model and our tests of the effect of audit certainty on taxpayer reporting behavior for years 2007 through 2011. Our inferences are unchanged, suggesting any changes to the Internal Revenue Manual or accounting for income tax uncertainty do not materially affect our pattern of results. 15 We estimate all equations on firm-year observations. Firm and year subscripts are suppressed in all equations.

15

sales (SALE) proxy for Size. We employ four proxies for Complexity: the number of geographic

segments in the Compustat Segments dataset, the number of business segments in the Compustat

Segments dataset, foreign sales (SALEFO) and foreign tax (TXFO+TXDFO). We are unable to

create a publicly available proxy for the related transactions criterion. Thus, our main prediction

model is as follows:

CICFirm = α + β1*AssetPoints + β2*GrossReceiptsPoints+ γ1*GeoSegPoints + γ2*BusSegPoints + γ3*ForeignSalesPoints + γ4*ForeignTaxPoints + ε (4)

See Appendix 2 for detailed information about the point assignment for each variable.

We also test whether firm attributes that affect the incentive or ability to avoid taxes

influence the CIC assignment decision. We incorporate five firm attributes into Equation (4) as

follows:

CICFirm = α + β1*AssetPoints + β2*GrossReceiptsPoints+ γ1*GeoSegPoints + γ2*BusSegPoints + γ3*ForeignSalesPoints + γ4*ForeignTaxPoints + δ*FirmAttributes + ε (5) Leverage (DLTT/AT) captures a conforming book-tax strategy; consistent with the tax

exhaustion theory outlined in DeAngelo and Masulis (1980), firms with high levels of debt tax

shields should engage in fewer additional tax avoidance strategies. R&D (XRD/SALE) may

positively affect CIC assignment as the research and experimentation credit is an area of

considerable complexity, aggressiveness, and conflict, and the IRS notes that CIC assignment is

based in part on complexity. However, following the tax exhaustion logic outlined for Leverage,

R&D may negatively affect CIC assignment because firms with substantial R&D credits should

engage in fewer alternative tax avoidance strategies. CapInt (PPENT/AT) proxies for the capital

intensity of a firm; similar to R&D, this variable may positively affect CIC assignment if the IRS

is concerned about asset category assignment or negatively affect CIC assignment under the tax

exhaustion hypothesis. ExcessStockBen (one if TXBCOF > 0, zero otherwise) captures tax-

saving stock option deductions in excess of stock option expenses and should be negatively

16

related to CIC assignment under the tax exhaustion theory (Graham, Lang, and Shackelford

2004). Finally, NOL (one if TLCF > 0, zero otherwise) captures whether the firm has a net

operating loss (NOL) carryforward; firms with NOL carryforwards have lower incentives to

engage in additional tax avoidance strategies. The full specification of this regression is

presented in equation (6):

CICFirm = α + β1*AssetPoints + β2*GrossReceiptsPoints+ γ1*GeoSegPoints + γ2*BusSegPoints + γ3*ForeignSalesPoints + γ4*ForeignTaxPoints + δ1*Leverage + δ2*R&D + δ3*CapInt + δ4*ExcessStockBen + δ5*NOL + ε (6)

Multivariate regression specification

We first estimate a pooled analysis to test whether firms assigned to the CIC program

report greater tax liabilities.

Tax = β0 + β1*CICParticipationInd + Controls + ε (7)

We measure Tax in three ways. Fed_Cash_ETR represents the taxpayer’s initial federal

tax rate and equals the total tax reported on the U.S. tax return (Page 1 Line 31 of Form 1120)

divided by worldwide pretax income (PI). Adj_Fed_Cash_ETR represents total tax payments

made to the IRS and equals total tax reported on the tax return (Page 1 Line 31 of Form 1120)

plus settlements paid to the IRS divided by worldwide pretax income (PI).16,17 Cash_ETR

represents payments made to tax authorities worldwide and equals cash taxes paid (TXPD)

divided by worldwide pretax income (PI). Our Fed_Cash_ETR measure represents the amount

taxpayers report on their original return and therefore captures deterrence effects. If the CIC

16 We scale the domestic income tax measures by worldwide pretax income because U.S. firms are taxed on their worldwide income. However, firms only pay U.S. tax on foreign income that is repatriated or deemed repatriated. Thus, in untabulated results, we alternatively scale both Fed_Cash_ETR and Adj_Fed_Cash_ETR by domestic pretax income (PIDOM) in place of worldwide pretax income, and our core results hold.17 We measure settlements paid to the IRS using a confidential IRS database. Our use of these data demands an important caveat. Even in the CIC program, proposed audit adjustments often take a number of years to become settlements as the firm applies its right to various methods of appeal and litigation. Thus, firms with missing settlement data may still be appealing the proposed audit adjustment. Incorrectly assuming that the final settlement equals zero for ∆Firms or for non-∆Firms would understate the settlement amount.

17

program has a deterrence effect, Fed_Cash_ETR should be higher for firms assigned to the CIC

program. Our Adj_Fed_Cash_ETR and Cash_ETR measures represent both initial tax payments

and payments made after the audit and therefore capture both deterrence and enforcement

effects. If the CIC program has a deterrence effect or an enforcement effect, Adj_Fed_Cash_ETR

and Cash_ETR should be higher for firms assigned to the CIC program. We include the proxies

for Size, Complexity, and FirmAttributes from Equation (6) to control for underlying differences

between firm-years assigned to the CIC program and those not assigned to the CIC program.

CICParticipationInd equals one if a firm is aware of CIC program assignment during the current

year and zero otherwise.18 We interpret an insignificant coefficient on CICParticipationInd as

consistent with the null hypothesis that audit certainty does not affect tax payments.

We next use changes analyses to examine whether firms that are assigned to the CIC

program during our sample period (“newly-assigned firms”) alter their tax reporting behavior.

An important issue in examining whether tax reporting behavior changes when a firm enters the

CIC program is that we do not observe the counterfactual—what would have happened had the

firm not been assigned to the CIC program. Specifically, the trend in tax payments could be

explained by the nature of the firms that are experiencing a program status change, by industry

trends, or by mean reversion.

To address the counterfactual issue, we construct two control samples. The first matched

sample for newly-assigned firms begins with all firms that are not assigned to the CIC program

at any time during our sample period (“non-assigned firms”). A non-assigned firm is matched

with a newly-assigned firm in the year the newly-assigned firm is first assigned to the CIC

18 Specifically, CICParticipationInd equals one for Year t+1 and future years for newly-assigned firms, one for all years for long-assigned firms, and zero for all years for non-assigned firms. Though the tax return for Year t for newly-assigned firms is subject to CIC audit, it was not prepared with that knowledge so CICParticipationInd equals zero for newly-assigned firms in Year t. Appendix 1 presents a timeline of CIC program assignment.

18

program on both year and the propensity score generated in Equation (4).19 The second matched

sample for the newly-assigned firms begins with all firms that have been assigned to the CIC

program for at least four years prior to the newly-assigned firms’ initial assignment and remain

assigned to the CIC program (“long-assigned firms”). As with non-assigned firms, long-assigned

control firms are matched with newly-assigned treatment firms on year and propensity score.

This matched sample design allows us to compare the tax payments of a newly-assigned firm to

itself before and after the change in its CIC program status and to compare its tax payments with

the tax payments of a firm that does not experience a change in CIC program status.

We estimate Equation (8) for each set of matches separately:

Tax = β0 + β1*Post + β2*∆Firm + β3* Post*∆Firm + Controls + ε (8)

As above, we measure Tax as Fed_Cash_ETR, Adj_Fed_Cash_ETR, or Cash_ETR. Post

equals one for both the ∆Firm and its matched non-∆Firm for all years after the change in CIC

program status and zero otherwise. Thus, Post represents years under which the ∆Firm is aware

that it faces certain audit (i.e., Post equals one for years on and after Year t+1). ∆Firm equals

one for firms that experience a change in their CIC program assignment status during our sample

period, and zero otherwise. In other words, ∆Firm equals zero for non-assigned firms and long-

assigned firms. When we estimate Equation (8), we continue to include the proxies for Size,

Complexity, and FirmAttributes detailed in Equation (6) to control for differences between the

firms whose CIC assignment status changes and the matched sample that does not experience a

status change.20 We interpret an insignificant coefficient on β3 as consistent with our hypothesis.

4. Samples and results

19 A taxpayer first coded as a CIC firm in Year t is notified of the assignment in mid- to late Year t+1. Because Year t financial statement data is available at the time of assignment, we use Year t data to calculate the propensity score for matching purposes.20 Though the average covariate balance for both sets of matched samples is below the generally accepted cut-off of 0.25 (Ho, Imai, King, and Stewart 2007), we include the covariates as control variables because our matched sample firms are significantly different from our newly-assigned firms on some dimensions.

19

CIC prediction model

The sample used to estimate Equations (4) and (6) for the determinants of CIC

assignment begins with all publicly-traded corporate firm-year observations with at least $250

million in total assets.21,22 We impose this size restriction to at least partially control for

differences in size between CIC and non-CIC firms.23 We then exclude: (i) observations without

Compustat data; (ii) observations with missing values for the required variables; and (iii)

observations missing one year of lag data and one year of lead data. These sample criteria result

in a sample of 23,094 firm-year observations used to estimate Equations (4) and (6). Panel A of

Table 1 details the selection of this sample, and Panel B details the number of observations by

year.

[Insert Table 1 around here]

Table 2 provides descriptive statistics on our CIC prediction model sample. Panel A

presents descriptive statistics for the 4,493 firm-years assigned to the CIC program, and Panel B

presents descriptive statistics for 18,601 firm-years not assigned to the CIC program. Firm-years

assigned to the CIC program report greater assets and higher gross receipts than firm-years not

assigned to the CIC program. Additionally, firm-years assigned to the program report more

geographic and business segments, higher foreign sales, and higher foreign taxes than firm-years

not assigned to the CIC program. These differences are consistent with the CIC identification

21 We chose to limit our sample to firms with at least $250 million in total assets because the CIC program was intended to audit large corporate taxpayers. Although a specific size threshold for CIC taxpayers is not defined in the Internal Revenue Manual, multiple sources have unofficially defined large case enforcement as taxpayers with at least $250 million in assets. For example, in remarks to Tax Executives Institute in 2012, former IRS Deputy Commissioner Steven Miller described the large corporate sector as “corporations with assets of $250 million or over” (https://www.irs.gov/uac/remarks-of-steven-t-miller-irs-deputy-commissioner-service-and-enforcement-before-the-tax-executives-institute-mid-year-conference).22 We focus on public firms because doing so enables us to construct our determinants model using Compustat data, which are accessible to researchers without access to tax return data. However, our inferences are unchanged when we include both public and private firm-year observations in our sample.23 Non-CIC firms can be under audit even though they are not participating in the CIC program. However, unlike CIC firms that know they will be audited with certainty before filing their tax return, non-CIC firms are not informed of the audit until after filing their tax return.

20

criteria outlined in the Internal Revenue Manual.

Untabulated statistics suggest that CIC program assignment is quite sticky. Of the 3,611

unique firms in our sample, 683 (18.9%) are assigned to the CIC program at least once during

our sample period and 2,882 (79.8%) are never assigned to the CIC program during our sample

period. Of the 683 unique sample firms assigned to the CIC program at least once during our

sample period, 68 (10%) are assigned for one year, 264 (39%) are assigned for between two and

five years, 281 (41%) are assigned for between seven and eleven years, and 70 (10%) are

assigned for all years during our sample period. In addition, of the firms assigned to the CIC

program at least once, the median firm is assigned to the program for 83 percent of the years it

exists in our sample and 35 percent of firms are assigned to the CIC program for every year they

are in the sample.24

[Insert Table 2 around here]

Table 3 presents our CIC prediction model. Panel A presents results of estimating

Equation (4), which assumes that Size and Complexity are the primary assignment factors and

thus includes only proxies outlined in IRM Exhibit 4.46.2-2. Column [1] presents the

determinants of CIC participation for our main sample, which includes all firm-year observations

outlined in Table 1. Column [2] presents results estimated using a sample of firm-years with a

nontrivial likelihood of CIC assignment by including only firm-years where we estimate a score

of 12 or higher using the point system outlined in Appendix 2.25 In Columns [1] and [2],

CICFirm equals one for firm-years in which the firm is assigned to the CIC program, and zero

24 Some firms exist in our sample for fewer than 12 years. Thus, they may be assigned to the program for less than 12 years but still be assigned to the program for all years they exist in our sample.25 Though we do find CIC assignment for firm-years with an estimated score of as low as seven, we select 12 as the cut-off for the sample for two reasons. First, 12 coincides with the IRS’s stated CIC-inclusion point. However, because we use Compustat variables to generate the score rather than confidential tax return data, our pool of observations with a score of 12 does not perfectly align with the pool the IRS estimates. Second, 12 is the lowest score where at least 5 percent of observations with that score are assigned to the CIC program.

21

otherwise. Tests in Columns [1] and [2] essentially assume that the participation decision is

annual and independent. Column [3] presents determinants of the initial assignment and includes

only observations from our main sample that were not assigned to the CIC program in Year t-1.26

In Column [3], CICFirm equals one for observations that have initially been assigned to the CIC

program in Year t, and zero otherwise.

[Insert Table 3 around here]

Results presented in Panel A are consistent with IRM Exhibit 4.46.2-2 and suggest that

both Size and Complexity factors contribute to CIC assignment. We assess the classification

accuracy of our model using the area under the Receiver Operating Characteristic (ROC) curve,

which Hosmer, Lemeshow and Sturdivant (2013) suggest is the appropriate statistic for

evaluating model fit. In Columns [1], [2], and [3], we estimate areas of 94.02, 86.24, and 86.51,

respectively. Guidelines offered in Hosmer et al. (2013) classify the discriminatory power of the

three models in Table 3 as outstanding, excellent, and excellent, respectively.

Across all three specifications, we estimate that sales are the most economically

significant Size determinant of CIC assignment. Coefficient estimates suggest that a one point

increase in GrossReceiptPoints (corresponding to additional sales of between $1 and $3 billion)

increases the probability of CIC participation by between 3.8 and 5.5 percent and increases the

probability of initial CIC assignment by 0.6 percent. Regarding important Complexity

determinants of CIC assignment, both the number of geographic segments and the number of

business segments are statistically significant in all specifications. The relative importance of

these two segment variables varies, though, from one of the most economically significant

determinants in Column [1] to nearly the least economically important in Columns [2] and [3]. In 26 In Column [3], a firm assigned to the CIC program during our sample period is removed from the determinant sample beginning the year after its assignment. Thus, a firm never assigned to the CIC program remains in the sample for all years, while a firm assigned to the CIC program for every year between 2000 and 2011 is never included.

22

Columns [1] and [2], foreign complexity is also statistically related to CIC assignment. However,

when CIC firm-years are excluded after the initial year of CIC assignment (Column [3]), foreign

complexity is no longer directly related to CIC assignment but likely remains indirectly related to

CIC assignment because of its effect on the number of segments. Overall, Panel A of Table 3

confirms that both Size and Complexity factors contribute to CIC assignment, consistent with the

Internal Revenue Manual.

In untabulated tests, we use results in Panel A, estimated using data from 2000 to 2011,

to test the ability of our model to predict CIC participation in a “hold out” sample. We generate

measures of CIC_Likelihood_Score using 2012 Compustat data and coefficients from each of the

three columns in Panel A. We then test whether the CIC_Likelihood_Score is associated with

CIC participation or CIC initial assignment in 2012. We estimate that the CIC_Likelihood_Score

explains CIC participation in specifications derived from Columns [1] and [2] well, with an area

under the ROC curve of 92.19 and 82.75 percent, respectively. Though only six firms are

initially assigned to the program in 2012, the CIC_Likelihood_Score derived from Column [3]

explains their assignment well, with an area under the ROC curve of 90.43 percent. These results

further confirm that our model explains economically significant determinants of CIC program

participation and assignment.27

Table 3 Panel B presents results of estimating Equation (6), which models CIC

participation as a function of Size factors, Complexity factors, and firm attributes. The results

suggest that a number of firm attributes are associated with CIC assignment. Consistent with

R&D expenses contributing to tax complexity, we find that firms with higher R&D expenses are

more likely to be assigned to the CIC program. Consistent with tax exhaustion, we generally find 27 We also estimate our determinants models for years 2000 through 2009 and define the holdout period as 2010 through 2012, which increases our newly-assigned observations to 24. We estimate that the CIC_Likelihood_Score explains CIC participation in specifications derived from Table 3 Panel A Columns [1], [2], and [3] well, with an area under the ROC curve of 93.98 percent, 82.13 percent, and 85.02 percent, respectively.

23

that firms reporting excess tax benefits from stock options or an NOL carryforward are less

likely to be assigned to the CIC program. Despite the statistically significant coefficients,

however, the inclusion of these firm attributes does not change the area under the ROC curve in a

meaningful way. Specifically, we estimate the largest increase in the area under the ROC curve

in Column [2], which increases from 86.24% to 86.90%. Thus, while CIC assignment is

associated with common factors related to firms’ incentives or ability to avoid taxes, it appears

that Size and Complexity are the primary determinants of CIC assignment.28

Finally, in untabulated results, we include only the firm attributes that affect the incentive

or ability to avoid taxes as potential determinants of CIC assignment in the broad sample. We

find that the firm attributes alone give an area under the ROC curve of 56.70 percent. When

compared to the areas under the ROC curve of 94.02 percent estimated in Column [1] of Table 3

Panel A, these results continue to suggest that the Size and Complexity variables outlined by the

Internal Revenue Manual are the most economically significant determinants of CIC assignment.

Effect of audit certainty on tax reporting behavior

CIC assignment represents a positive shock to a firm’s perceived audit risk. Only 34

percent of newly-assigned CIC firms were audited in the year prior to assignment. Extending the

period, only 62 percent of newly-assigned firms were audited in at least one of the three years

prior to assignment. Thus, we expect CIC assignment to represent a substantial increase in audit

probability for newly-assigned firms and we test whether this audit probability shock affects tax

reporting behavior.

We use the CIC prediction model sample of 23,094 firm-year observations for our pooled

levels analysis. For our changes analyses, we begin the construction of the non-assigned control

28 In an alternative specification, we also included measures of tax avoidance in the likelihood model to test whether tax avoidance increases the likelihood of CIC assignment. None of the tax avoidance measures are statistically significant determinants of CIC assignment.

24

sample with the 405 firms first assigned to the IRS CIC program between 2000 and 2011. We

calculate the CIC prediction score for these firms in the year of their CIC assignment using

coefficients estimated in Column [1] of Panel A. Using the sample of firms that are never in the

CIC program between 2000 and 2011, we construct a sample of non-assigned control firms that

are matched on both year and CIC prediction score without replacement. For the 810 test and

control firms, we obtain Compustat data for Years t-2 to t+4, where Year t represents the year

the newly-assigned firm was assigned to the CIC program and Years t+1 through t+4 represent

the post-assignment years.29 This results in a final sample of 4,310 firm-year observations.

Panel A of Table 4 provides descriptive statistics for the newly-assigned and non-

assigned firms in Year t, the year of the match. Even though firms are matched on year and

propensity score, firms still differ on some dimensions. Specifically, newly-assigned firms are

larger in terms of both assets and gross receipts and have more foreign sales, pay more foreign

taxes, and have higher leverage and R&D than non-assigned firms. Accordingly, we include

these variables in our subsequent models to control for the effects that these differences may

have on tax reporting behavior.

[Insert Table 4 around here]

As above, we begin the construction of the long-assigned control sample with the 405

firms first assigned to the IRS CIC program between 2000 and 2011. Using the sample of firms

that are in the CIC program for the four years prior to the newly-assigned firm’s assignment to

the program, we construct a sample of long-assigned control firms that are matched on both year

and CIC prediction score without replacement. Panel B of Table 4 provides descriptive statistics

for the newly-assigned and long-assigned firms in the year of the match. For the 360 test and 29 We require both the newly-assigned firm and the non-assigned match firm to have sufficient data in Years t-1 to t+1 for the firms to remain in the sample. If either firm is not in the database in Years t-2, t+2, t+3, or t+4, we omit both firms in those years so that the match remains one-to-one based on firm-year. We follow the same methodology for the long-assigned matched sample.

25

control firms, we obtain Compustat data for Years t-2 to t+4, which results in a sample of 1,798

firm-year observations. Panel C of Table 4 presents descriptive statistics for each of our Tax

measures in: (i) the pooled sample, (ii) the newly-assigned/non-assigned sample, and (iii) the

newly-assigned/long-assigned sample.

Tables 5 through 7 present results of testing the null hypothesis that tax payments do not

change when a firm’s audit probability increases to one. In all three tables, Panel A presents the

pooled analysis while Panels B and C present the changes analysis using the non-assigned and

long-assigned matched samples, respectively. In Panels B and C, the first column presents results

of estimating Equation (7) on only the newly-assigned firms (“∆Firms”); the second column

mirrors the first for the matched control sample (“non-∆Firms”); and the third column presents

results of estimating Equation (8) with interactions to test the significance of any difference in

tax payments.

[Insert Table 5 around here]

We proxy for Tax using the initial federal tax rate (Fed_Cash_ETR) in Table 5. In Panel

A, we estimate no statistically significant difference in Fed_Cash_ETR between firms assigned

to the CIC program and firms not assigned to the CIC program. Results presented in Panels B

and C confirm this finding: firms newly-assigned to the CIC program do not report different

Fed_Cash_ETR relative to either a sample of firms never assigned to the CIC program or to a

sample of firms already assigned to the CIC program. In untabulated results, we include industry,

year or industry-year fixed effects in Panel A or define the initial federal tax rate as a function of

total assets reported in Compustat (AT). In all of these alternative specifications, we continue to

estimate no statistically significant change in Fed_Cash_ETR when firms are assigned to the CIC

program. Thus, results suggest that audit certainty does not have a larger deterrence effect

26

relative to the IRS’s standard selection and audit process for large corporations not included in

the CIC program.

We proxy for Tax using Adj_Fed_Cash_ETR in Table 6 and Cash_ETR in Table 7. We

discuss the results presented in these two tables together because both measures are intended to

capture total tax payments, which is a combination of deterrence and enforcement. In Panel A of

both tables, we estimate no statistically significant difference in Adj_Fed_Cash_ETR or

Cash_ETR between firms assigned to the CIC program and firms not assigned to the CIC

program. The changes analysis in Panels B and C of both tables also estimate that firms do not

report a statistically significant change in total federal cash payments or cash taxes paid, relative

to income, once they are assigned to the CIC program. As above, inferences are unchanged when

we include various fixed effects in Panel A or define the Tax variables as a function of total

assets in all three panels of both tables.

In additional untabulated analysis, we disaggregate the Post variable into four separate

years to examine whether payments change over time. We estimate no statistically significant

difference in Fed_Cash_ETRs, Adj_Fed_Cash_ETRs, or Cash_ETRs between ∆Firms and non-

∆Firms in any year after the change in program status. Overall, our evidence suggests that audit

certainty does not have greater deterrence or enforcement effects relative to the IRS’s standard

selection and audit process for large corporations not included in the CIC program. We therefore

fail to reject our hypothesis.

[Insert Tables 6 & 7 around here]

The results presented in Tables 5 through 7 contrast with results documented at lower

points in the audit probability continuum that an increase in audit probability is associated with

fewer uncertain tax positions. Our results also contrast with the findings in Slemrod et al. (2001)

27

that individual taxpayers claim more uncertain tax positions when facing audit certainty. One

explanation for the differing results is that the individual taxpayers examined by Slemrod et al.

(2001) face only one year of audit certainty, while the corporate taxpayers in our study face

multiple consecutive years of audit certainty (i.e., a repeated game).

To provide additional comfort that we fail to reject our hypothesis because audit certainty

does not affect tax reporting behavior, rather than because our tests lack power to detect an

effect, we undertake a number of additional analyses. First, we re-estimate our determinants

model using the corporate tax return data corresponding to the determinants outlined in IRM

Exhibit 4.46.2-2: total assets and total gross receipts.30 Constructing even two of the point

variables using corporate tax return data could increase the power of our tests because the IRS

uses corporate tax return data to assign points. We estimate a similar area under the ROC curve

(92.70 percent, compared to 94.02 percent in Panel A, Column [1] in Table 3) using this

specification. Further, when we use the propensity score from this model to create our matched

samples, we continue to estimate a statistically insignificant effect of audit certainty on all three

dependent variables relative to both non-assigned and long-assigned firms. Second, we re-

estimate Equations 7 and 8 separately for a sample of firms that were not audited in the year

prior to CIC assignment. This subsample should increase the power of our tests because CIC

assignment likely represents a more substantial audit shock. However, we continue to estimate

no change in tax reporting behavior in this subsample.

Third, we conduct a falsification test where we assume CIC assignment occurred in Year

t-1 rather than in Year t; this test is intended to address the possibility that firms anticipated their

assignment to the program and thus adjusted their tax reporting behavior prior to joining the

program. We continue to estimate a statistically insignificant effect of audit certainty on all three

30 Data to construct the remaining point variables were not made available by the IRS for this project.

28

dependent variables relative to both non-assigned and long-assigned firms. Fourth, our results are

also robust to alternative matching approaches such as one-to-many propensity score matching

and matching based on total points (both one-to-one and one-to-many) rather than on propensity

scores. Finally, we discuss a number of additional alternative specifications in Section 6.

Together with results in Tables 5 through 7, these untabulated tests suggest that audit certainty

does not perceptibly affect tax reporting behavior.

5. Supplemental analysis

The results in Tables 5 through 7 suggest that the CIC program does not have higher

deterrence or enforcement effects relative to the IRS’s standard selection and audit process for

large corporations. In this section, we consider whether the CIC program affects managers’

expectations of future tax payments. We estimate equations (7) and (8) using current-year

additions to the FIN 48 reserve as the dependent variable. Specifically, UTB_CY_ADD

represents expected future cash payments related to uncertain tax positions claimed worldwide in

the current year and equals reserve increases related to tax positions claimed in the current year

(TXTUBPOSINC) divided by worldwide income (PI). Because FIN 48, effective in 2007,

changed the recognition and measurement criteria for recording uncertain tax benefits, we

perform this analysis on fiscal years 2007 through 2011.

We present the results in Table 8. We find that audit certainty does impact current year

additions to the contingent tax reserve. Panel A presents descriptive statistics for UTB_CY_ADD.

In the pooled regression presented in Panel B, we estimate that firms aware of their CIC program

assignment when they file their financial statements report 58 percent more, or approximately

$1.4 million (0.004 coefficient estimate multiplied by $341.6 million average pretax income)

more, current year additions to the contingent tax reserve than do firms not assigned to the CIC

29

program. Panels C and D offer similar results, suggesting that relative to either matched sample,

firms experiencing a change in program status report a higher UTB_CY_ADD than firms not

experiencing a change in status.31 Inferences are unchanged if we include fixed effects or use

total assets as an alternative scalar. Thus, in a variety of specifications, we consistently find that

firms in the CIC program report higher contingent tax reserves.

[Insert Table 8 around here]

The result in Table 5 that the initial tax liability does not change for newly-assigned firms

suggests that the increased reserves do not represent an increase in uncertain tax avoidance.

Instead, plausible explanations include: (i) firms systematically underestimated their likelihood

of sustaining a position prior to CIC assignment and subsequently update their expectations

based on learning new information during the audit process; and/or (ii) firms incorporated audit

likelihood in their determination of reserves prior to CIC assignment (inconsistent with the U.S.

GAAP requirement that firms assume audit certainty with respect to each uncertain position).

6. Robustness tests

We outline a number of additional alternative specifications of the CIC prediction model.

First, because Mills and Newberry (2001) suggest that firms with more than $500 million in

assets are typically included in the CIC program, we re-estimate Table 3 on a sample of firms

with total assets of at least $500 million. Consistent with their suggestion, 25.2 percent of the

firms in this sample are assigned to the CIC program as compared with 19.5 percent of the

sample in Table 3. The area under the ROC curve is also relatively stable in this alternative

specification. Thus, results appear robust to alternative sample specifications regarding firm size.

Further, performing the non-assigned and the long-assigned matches based on this alternative

31 In terms of economic magnitude for Panels C and D, firms experiencing a change in program status report $3.2 million ($10.3 million) more current year additions to the contingent tax reserve after joining the program relative to their matched non-assigned (long-assigned) firms in the same period.

30

sample does not change the results of our hypothesis tests.

To address concerns regarding both separation (that firms assigned to the CIC program

are substantially different from those not assigned to the CIC program) and that the number of

events in our sample is small relative to the sample itself, we also estimate all regressions in

Table 3 using a Firth bias correction.32 Coefficients are qualitatively similar across all six

specifications in Table 3, suggesting that our models are not meaningfully impacted by

separation or small sample size.

We also perform a number of robustness checks on the analyses in Tables 5 through 7.

We first re-estimate the pooled tests on the sample of firms with an estimated score of at least 12;

that is, we use the sample in Column [2] of Table 3. This sample biases for a statistically

significant difference because non-assigned firm-years in this group should be more similar to

the newly-assigned firm-years. We continue to estimate no statistically significant difference in

Fed_Cash_ETR, Adj_Fed_Cash_ETR, or Cash_ETR between newly-assigned firms and

non(long)-assigned firms, but a statistically significant increase in UTB_CY_ADD after the

change in program status. Next, we use an alternative, domestic-only scalar for the two

dependent variables based on U.S. federal tax liability. Though U.S. firms are taxed on

worldwide income, they will only pay U.S. tax on foreign income that is repatriated or deemed

repatriated. For firms that reinvest most of their foreign-source income or that face high foreign

tax rates, this can lead to a mismatch between a domestic-only numerator and worldwide

denominator. Thus, we alternatively define Fed_Cash_ETR and Adj_Fed_Cash_ETR using

domestic pretax income (PIDOM) as the scalar. We continue to estimate no statistically

32 Maximum likelihood estimation in logistic estimation suffers from small-sample bias that is strongly dependent on the number of cases in the less frequent of the two categories (CIC assignment in our case). The Firth method, also known as penalized likelihood, corrects for this bias.

31

significant difference in Fed_Cash_ETR or Adj_Fed_Cash_ETR between newly-assigned firms

and non(long)-assigned firms after the change in program status.

Next, we test a number of alternative match designs. In the tabulated estimates, we create

our matched samples using the propensity scores generated in Column [1] of Table 3 Panel A,

which models the assignment decision and includes the year of assignment as well as additional

years of program participation. In untabulated analysis, we obtain consistent results when we

instead create our control samples using the propensity scores generated in either Column [2],

which includes only firms more likely to be assigned, or in Column [3], which models the initial

assignment. We also match on the propensity score generated in the models in Table 3 Panel B

instead of Table 3 Panel A and find both statistically and economically similar results. Thus,

results appear robust to these alternative matching specifications.

7. Conclusion

This study explores the effect of audit certainty on tax reporting behavior by examining

taxpayers in the IRS Coordinated Industry Case (CIC) Program, a set of taxpayers that consume

a substantial portion of IRS LB&I audit resources. Our results suggest that after entering the CIC

program, firms report similar initial tax liabilities and total tax liabilities as a percentage of

income relative to propensity-matched control firms. In terms of the strategic tax model, our

results suggest that audit certainty does not have greater deterrence or enforcement effects

relative to the IRS’s standard selection and audit process for large corporations not included in

the CIC program. However, we find that newly-assigned CIC taxpayers report higher additions

to the contingent tax reserves relative to propensity score-matched control firms, suggesting that

audit certainty does impact financial reporting for income taxes.

32

Our study makes the following contributions. First, our model of the determinants of CIC

assignment provides researchers without access to confidential corporate tax return data with a

way to estimate whether a firm faces certain audit by the IRS. Second, we provide evidence that

although inclusion in the CIC program is associated with firms’ incentives or ability to avoid

taxes, this substantial resource allocation decision is primarily based on firm size and

complexity. Third, we further our understanding of the strategic game between the taxpayer and

the tax authority by examining the deterrence and enforcement effects of audit certainty. Finally,

our study provides data useful to tax authorities in assessing the cost and benefits of the CIC

program. Our results suggest that the CIC program does not have higher deterrence and

enforcement effects than the IRS’s standard process of auditing large corporations not in the CIC

program. Though increasing revenue is only one of the five factors Scott (1991) outlines as

influencing the development of the CIC program, our finding that audit certainty does not raise

greater revenue relative to the IRS’s standard audit process for large corporations remains an

important finding as the IRS considers, designs and implements new audit approaches.

This research is subject to some caveats. First, we acknowledge that the match between

our treatment and control firms is imperfect. We create multiple matched samples as well as

various matching methods to allay concerns, but to the extent that treatment and control firms are

dissimilar, the parallel trends assumption may not be valid. Second, despite conducting several

analyses to ensure that our failure to reject the null hypothesis is because audit certainty does not

affect tax reporting behavior, we cannot fully rule out the possibility that our null results are

attributed to a lack of power or firms anticipating being assigned to the CIC program prior to

assignment. Finally, the U.S. allocates substantial resources to its tax authority. To the extent that

33

other tax authorities allocate fewer or greater resources for tax enforcement, our results might not

generalize to other certain audit programs.

34

Appendix 1Timeline for CIC program assignment

File Annual Report File Tax Return Year end for Year t for Year t Year end________________|__________|______________________|________|____________

12/31/t 03/01/t+1 09/15/t+1 12/31/t+1

Notes:

The highlighted portion (______) is the typical time period that a firm would be notified regarding CIC assignment for Year t. Thus, the Annual Report and the Tax Return for Year t are filed before the firm becomes aware of its CIC assignment, and the Annual Report and the Tax Return for Year t+1 are filed after the firm becomes aware of its CIC assignment.

The propensity score match is done using Year t data. Post = one for Years t+1 and forward, years under which the Annual Report and the Tax Return are

filed knowing that the firm will be subject to CIC audit, and zero otherwise. Though Years t and forward will certainly be subject to CIC audit as long as the firm is assigned to the

CIC program, any previous open tax years could also be subject to CIC audit.

35

Appendix 2Summary of IRM Exhibit 4.46.2-2

The Internal Revenue Manual outlines criteria for CIC program assignment in Exhibit 4.46.2-2. The assignment is based on seven criteria: 1) Total Assets, 2) Gross Receipts, 3) Operating Entities, 4) Multiple Industry Status, 5) Total Foreign Assets, 6) Total Related Transactions and 7) Foreign Tax. We model our variables and point assignment scheme based on this document.

We are able to use Compustat data and closely follow the variable definition and point assignment system for most of the criteria.

1. Total Assets (Compustat AT, IRS 4.46.2-2 point system)•1 point up to $500 Million in assets;•2 points for assets in the $500 Million to $1 Billion range;•3 points for assets in the $1 Billion to $2 Billion range;•4 points for assets in the $2 Billion to $5 Billion range;•5 points for assets in the $5 Billion to $8 Billion range;•Add 1 point for each additional $3 billion in assets, or fraction thereof, not to exceed 12 points.

2. Gross Receipts (Compustat SALE, IRS 4.46.2-2 point system)•1 point up to $1 Billion in gross receipts;•2 points for gross receipts in the $1 Billion to $2 Billion range;•3 points for gross receipts in the $2 Billion to $3 Billion range;•4 points for gross receipts in the $3 Billion to $5 Billion range;•5 points for gross receipts in the $5 Billion to $10 Billion range; •Add one point for each additional $3 Billion in gross receipts, or fraction thereof, not to exceed 10 points.

3. Operating Entities (geographic segments from Compustat Segments dataset, IRS 4.46.2-2 point system)

•Number of Entities 1 – 1 point;•Number of Entities 2-5 – 3 points;•Number of Entities 6-9 – 5 points;•Number of Entities 10-13 – 7 points;•Number of Entities Over 13 – 9 points.

4. Multiple Industry Status (business segments from Compustat Segments dataset, point system based on point system for Operating Entities because IRM Exhibit is not detailed for this item)

•Number of Industries 1 – 1 point;•Number of Industries 2-5 – 3 points;•Number of Industries 6-9 – 5 points;•Number of Industries 10-13 – 7 points; •Number of Industries Over 13 – 9 points.

5. Total Foreign Assets (we substitute foreign sales for foreign assets; foreign sales percentage from Compustat Segments dataset times Compustat SALE; modified point system)

•1 point up to $500 Million in foreign sales;•2 points for foreign sales in the $500 Million to $1 Billion range;•3 points for foreign sales in the $1 Billion to $1.5 Billion range;•4 points for foreign sales in the $1.5 Billion to $2.5 Billion range;•5 points for foreign sales in the $2.5 Billion to $5 Billion range;•Add 1 point for each additional $1.5 Billion in foreign sales, or fraction thereof, not to exceed 8 points.

We are unable to create a publicly available proxy for criteria 6) Total Related Transactions.

36

7. Foreign Tax (Compustat TXFO, IRS 4.46.2-2 point system)•1 point up to $7 Million in foreign tax;•2 points for foreign tax in the $7 Million to $100 Million range;•3 points for foreign tax in the $100 Million to $200 Million range;•Add 1 point for each additional $200 million of foreign tax, or fraction thereof, not to exceed 8 points.

37

Appendix 3Variable definitions

AssetPoints =

GrossReceiptsPoints =

GeoSegPoints =

BusSegPoints =

ForeignSalesPoints =

ForeignTaxPoints =

Leverage =

R&D =

CapInt =

ExcessStockBen =

NOL =

Fed_Cash_ETR =

Adj_Fed_Cash_ETR =

Cash_ETR =

UTB_CY_ADD =

CICParticipationInd =

∆Firm =

Post =

One if Compustat TLCF is greater than 0, and zero otherwise

Net property, plant and equipment (Compustat PPENT) divided by total assets (Compustat AT)

Number of points assigned for revenues (Compustat SALE) based on IRM Exhibit 4.46.2-2; See Appendix 2 for specific point ranges

Number of points assigned for business segments (Compustat Segments dataset) based on IRM Exhibit 4.46.2-2; See Appendix 2 for specific point ranges

Number of points assigned for foreign sales (% of foreign sales from Compustat segmerged * SALE) based on IRM Exhibit 4.46.2-2; See Appendix 2 for specific point ranges

One if Compustat TXBCOF is greater than 0, and zero otherwise

One for both the CIC firm and its matched firm for all years after the change in CIC program status

Total tax reported on Page 1 Line 31 of Form 1120 divided by pretax income (Compustat PI)

Total cash taxes paid (Compustat TXPD) divided by pretax income (Compustat PI)

One for firms that experience a change in their CIC program assignment status during our sample period and zero otherwise

Additions to tax contingency reserves for positions claimed in the current year (Compustat TXTUBPOSINC) divided by pretax income (Compustat PI)

One if a firm is aware of CIC program assignment during the current year, and zero otherwise

(Total tax reported on Page 1 Line 31 of Form 1120 plus settlements paid to IRS) divided by pretax income (Compustat PI)

Number of points assigned for foreign taxes (Compustat TXFO) based on IRM Exhibit 4.46.2-2; See Appendix 2 for specific point ranges

Long-term debt (Compustat DLTT) divided by total assets (Compustat AT)

Number of points assigned for total assets (Compustat AT) based on IRM Exhibit 4.46.2-2; See Appendix 2 for specific point ranges

Research and development expenses (Compustat XRD) divided by revenues (Compustat SALE)

Number of points assigned for geographic segments (Compustat Segments dataset) based on IRM Exhibit 4.46.2-2; See Appendix 2 for specific point ranges

38

References

Allingham, M. G., and A. Sandmo. 1972. Income Tax Evasion: A Theoretical Analysis. Journal of Public Economics 1(3-4): 323-338.

Beck, P., and P. Lisowsky. 2013. Tax Uncertainty and Voluntary Real-time Tax Audits. The Accounting Review 89(3): 867-901.

DeAngelo, H., and R. Masulis. 1980. Optimal Capital Structure under Corporate and Personal Taxation. Journal of Financial Economics 8(1): 3-29.

DeBacker, J., B. T. Heim, A. Tran, and A. Yuskavage. 2015. Legal enforcement and corporate behavior: An analysis of tax aggressiveness after an audit. Journal of Law and Economics 58: 291-324.

De Simone, L., R. Sansing, and J. Seidman. 2013. When are Enhanced Relationship Tax Compliance Programs Mutually Beneficial? The Accounting Review 88(6): 1971-1991.

Graetz, M. J., J. F. Reinganum, and L. L. Wilde. 1986. The Tax Compliance Game: Toward an Interactive Theory of Law Enforcement. Journal of Law, Economics, & Organization 2(1): 1-32.

Graham, J., M. Lang, and D. Shackelford. 2004. Employee Stock Options, Corporate Taxes, and Debt Policy. Journal of Finance 59: 1585-1618.

Hanlon, M., L. Mills, and J. Slemrod. 2007. An Empirical Examination of Corporate Tax Noncompliance. In Taxing Corporate Income in the 21st Century, ed. A. J. Auerbach, J. R. Hines Jr., and J. Slemrod, 171-210. Cambridge, UK: Cambridge University Press.

Ho, D., K. Imai, G. King, and E. Stewart. 2007. Matching as Nonparametric Preprocessing for Reducing Model Dependence in Parametric Causal Inference. Political Analysis 15: 199-236.

Hoopes, J. L., D. Mescall, and J. A. Pittman. 2012. Do IRS Audits Deter Corporate Tax Avoidance? The Accounting Review 87(5): 1603-1639.

Hosmer, D., S. Lemeshow, and R. Sturdivant. 2013. Applied Logistic Regression. Wiley Series in Probability and Statistics, 3rd ed. New York: John Wiley & Sons Inc.

Lennox, C., W. Li, J. Pittman, and Z. Wang. 2015. The Determinants and Consequences of Tax Audits: Some Evidence from China. Working paper, Nanyang Technological University, Fuzhou University, Memorial University of Newfoundland, and Shanghai University of Finance and Economics.

Mills, L. F. 1998. Book-Tax Differences and Internal Revenue Service Adjustments. Journal of Accounting Research 36(2): 343-356.

39

Mills, L. F., and K. J. Newberry. 2001. The Influence of Tax and Nontax Costs on Book-Tax Reporting Differences: Public and Private Firms. Journal of the American Taxation Association 23(1): 1-19.

Mills, L. F., and R. C. Sansing. 2000. Strategic Tax and Financial Reporting Decisions: Theory and Evidence. Contemporary Accounting Research 17(1): 85-106.

Rhoades, S. C. 1999. The Impact of Multiple Component Reporting on Tax Compliance and Audit Strategies. The Accounting Review 74(1): 63-85.

Sansing, R. C. 1993. Information Acquisition in a Tax Compliance Game. The Accounting Review 68(4): 874-884.

Scott, P. K. 1991. Coping with the new IRS exam initiatives. William and Mary Tax Conference Proceedings. Found at: http://heinonline.org/HOL/Page?handle=hein.journals/antcwilm31&div=12&g_sent=1&collection=journals

Slemrod, J., M. Blumenthal, and C. Christian. 2001. Taxpayer Response to an Increased

Probability of Audit: Evidence from a Controlled Experiment in Minnesota. Journal of Public Economics 79(3): 455-483.

40

TABLE 1Sample derivation

Panel A: Sample selection

Publicly-traded firm-years from 2000 to 2011 with >=$250M in TaxReturnAssets 34,379

Less: observations not matched with Compustat data (2,057)

Less: observations missing dependent or explanatory variables (3,611)

Less: observations missing one year lag and/or one year lead (5,617)

Final sample 23,094

Less: firm-years prior to 2007 (12,092)

Sample for supplemental analysis 11,002

Firms entering the CIC program during sample period 405

Panel B: Number of observations by year

Year Firm-years % CIC All % CIC Firm-years % CIC2000 2,850 10.2% 1,297 15.2% 48 11.9%2001 2,826 12.4% 1,517 16.7% 55 13.6%2002 2,803 15.9% 1,604 22.5% 86 21.2%2003 2,781 17.6% 1,643 24.1% 49 12.1%2004 2,865 16.9% 1,662 23.9% 33 8.1%2005 2,891 17.5% 2,190 19.8% 44 10.9%2006 2,922 17.4% 2,179 19.6% 26 6.4%2007 2,906 17.3% 2,196 19.6% 20 4.9%2008 2,875 17.3% 2,226 19.3% 15 3.7%2009 2,876 16.6% 2,216 18.7% 11 2.7%2010 2,901 15.6% 2,183 18.3% 13 3.2%2011 2,883 13.8% 2,181 16.2% 5 1.2%Total 34,379 23,094 405

Initial sample # CIC JoinersCIC Prediction

Notes:

This table presents the sample derivation process. Panel A provides the sample selection and Panel B provides the number of observations by year. TaxReturnAssets equals total assets reported on Page 1 of the Form 1120.

41

TABLE 2Descriptive statistics for CIC prediction model

Panel A: Descriptive statistics for CIC firm-years

Variable N Median Mean SDAssetPoints 4,493 6.0 7.4 3.4GrossReceiptsPoints 4,493 5.0 5.6 2.7GeoSegPoints 4,493 3.0 2.7 1.5BusSegPoints 4,493 3.0 2.5 1.4ForeignSalesPoints 4,493 2.0 3.4 2.9ForeignTaxPoints 4,493 2.0 2.2 1.7Leverage 4,493 0.244 0.263 0.178R&D 4,493 0.000 0.030 0.062CapInt 4,493 0.205 0.273 0.229ExcessStockBen 4,493 0.000 0.280 0.449NOL 4,493 0.000 0.387 0.487

Panel B: Descriptive statistics for Non-CIC firm-years

Variable N Median Mean SDAssetPoints 18,601 2.0 *** 2.8 *** 1.8GrossReceiptsPoints 18,601 1.0 *** 1.8 *** 1.4GeoSegPoints 18,601 1.0 *** 2.1 *** 1.3BusSegPoints 18,601 1.0 *** 1.9 *** 1.2ForeignSalesPoints 18,601 1.0 *** 1.3 *** 0.9ForeignTaxPoints 18,601 1.0 *** 1.2 *** 0.5Leverage 18,601 0.205 *** 0.249 *** 0.221R&D 18,601 0.000 *** 0.026 *** 0.069CapInt 18,601 0.146 *** 0.236 *** 0.245ExcessStockBen 18,601 0.000 0.287 0.452NOL 18,601 0.000 *** 0.360 *** 0.480

Notes:

***, **, * denotes two-tailed significance at 0.01, 0.05, and 0.10, respectively.

This table presents descriptive statistics for variables in the CIC prediction model. Our sample period includes fiscal years 2000 through 2011. Panel A presents descriptive statistics for CIC firm-years and Panel B presents descriptive statistics for Non-CIC firm-years. The Points variables are based on the point allocations outlined in the Internal Revenue Manual. See Appendix 3 for variable definitions.

42

TABLE 3CIC prediction model

Panel A: CIC prediction model with IRM factors

Constant -5.822 *** -4.064 *** -6.031 ***(0.091) (0.115) (0.152)

AssetPoints 0.339 *** 0.261 *** 0.194 ***(0.012) (0.012) (0.024)[0.025] [0.040] [0.004]

GrossReceiptsPoints 0.502 *** 0.365 *** 0.345 ***(0.018) (0.018) (0.031)[0.038] [0.055] [0.006]

GeoSegPoints 0.221 *** 0.066 *** 0.190 ***(0.021) (0.022) (0.039)[0.017] [0.010] [0.004]

BusSegPoints 0.154 *** 0.063 *** 0.110 ***(0.019) (0.019) (0.041)[0.012] [0.009] [0.002]

ForeignSalesPoints 0.052 ** 0.118 *** -0.070(0.023) (0.021) (0.043)[0.004] [0.018] [-0.001]

ForeignTaxPoints 0.224 *** 0.115 *** 0.010(0.047) (0.042) (0.081)[0.017] [0.017] [0.000]

Additional sample cuts

NPseudo R-squaredArea under ROC curve

NoneCIC years after initial

assignmentPoints < 12

10,90745.75%

94.02% 86.51%

[1] [3]

CICFirm = 1 if firm is assigned to CIC

program during current year

23,094 19,00659.71% 17.55%

CICFirm = 1 if firm is initially assigned to CIC program during

current year

CICFirm = 1 if firm is assigned to CIC

program during current year[2]

86.24%

Notes: ***, **, * denotes two-tailed significance at 0.01, 0.05, and 0.10, respectively.

This table presents the results from estimating a logistic regression of CIC program assignment. Our sample period includes fiscal years 2000 through 2011. Panel A models CIC assignment as a function of factors outlined in the Internal Revenue Manual, and Panel B models CIC assignment as a function of factors outlined in the Internal Revenue Manual and firm attributes. See Appendix 3 for variable definitions. Standard errors are reported in parentheses and marginal effects are reported in brackets.

43

TABLE 3 (continued)CIC prediction model

44

Panel B: CIC prediction model with IRM factors and firm attributes

Constant -5.984 *** -4.114 *** -5.790 ***(0.105) (0.130) (0.176)

AssetPoints 0.338 *** 0.252 *** 0.194 ***(0.012) (0.012) (0.025)

GrossReceiptsPoints 0.533 *** 0.408 *** 0.359 ***(0.019) (0.019) (0.032)

GeoSegPoints 0.184 *** 0.028 0.167 ***(0.021) (0.023) (0.041)

BusSegPoints 0.167 *** 0.072 *** 0.104 **(0.019) (0.019) (0.041)

ForeignSalesPoints 0.041 * 0.101 *** -0.086 *(0.023) (0.022) (0.044)

ForeignTaxPoints 0.240 *** 0.122 *** 0.026(0.048) (0.043) (0.082)

Leverage 0.112 0.064 -0.498 *(0.132) (0.142) (0.285)

R&D 4.258 *** 5.223 *** 2.654 ***(0.351) (0.434) (0.759)

CapInt 0.274 ** 0.097 0.313(0.114) (0.120) (0.239)

ExcessStockBen -0.183 *** -0.237 *** -1.076 ***(0.054) (0.056) (0.155)

NOL -0.100 * -0.172 *** -0.045(0.052) (0.053) (0.114)

Additional sample cuts

NPseudo R-squaredArea under ROC curve

10,90747.13%

23,094 19,00660.33%

86.90%

[3][2]

NoneCIC years after initial

assignmentPoints < 12

[1]

94.22% 86.42%19.63%

CICFirm = 1 if firm is assigned to CIC

program during current year

CICFirm = 1 if firm is initially assigned to CIC program during

current year

CICFirm = 1 if firm is assigned to CIC

program during current year

TABLE 4Univariate analysis of newly-assigned firms, non-assigned firms, and long-assigned firms

45

Panel A: Descriptive statistics in assignment year for newly-assigned firms and non-assigned firms

Median Mean SD Median Mean SDAssetPoints 4.0 5.4 2.6 3.0 *** 3.7 *** 1.9GrossReceiptsPoints 4.0 4.1 2.3 2.0 *** 2.6 *** 1.5GeoSegPoints 3.0 2.6 1.4 3.0 2.6 1.3BusSegPoints 3.0 2.4 1.4 3.0 2.4 1.3ForeignSalesPoints 1.0 2.4 2.1 1.0 *** 1.5 *** 1.0ForeignTaxPoints 2.0 1.7 1.0 1.0 *** 1.5 *** 0.5Leverage 0.241 0.254 0.182 0.242 0.260 0.199R&D 0.000 0.032 0.066 0.000 0.055 *** 0.113CapInt 0.222 0.278 0.225 0.205 0.267 0.233ExcessStockBen 0.000 0.131 0.338 0.000 *** 0.049 *** 0.217NOL 0.000 0.385 0.487 0.000 0.346 0.476

Panel B: Descriptive statistics in assignment year for newly-assigned firms and long-assigned firms

Median Mean SD Median Mean SDAssetPoints 4.0 5.3 2.7 8.5 *** 8.5 *** 3.3GrossReceiptsPoints 3.0 3.7 2.2 5.0 *** 6.1 *** 2.5GeoSegPoints 3.0 2.6 1.4 3.0 2.5 1.4BusSegPoints 3.0 2.3 1.4 3.0 2.4 1.5ForeignSalesPoints 1.0 2.5 2.1 2.0 *** 3.4 *** 2.8ForeignTaxPoints 2.0 1.7 1.1 2.0 ** 2.1 ** 1.5Leverage 0.206 0.230 0.172 0.239 * 0.261 * 0.179R&D 0.000 0.037 0.070 0.000 * 0.025 * 0.055CapInt 0.157 0.244 0.229 0.169 0.242 0.226ExcessStockBen 0.000 0.250 0.434 0.000 0.278 0.449NOL 0.000 0.478 0.501 0.000 ** 0.367 ** 0.483

Newly-assigned firms (n=405) Non-assigned firms (n=405)

Newly-assigned firms (n=180) Long-assigned firms (n=180)

Panel C: Descriptive statistics for Tax measures

N Median Mean SD N Median Mean SD N Median Mean SDFed_Cash_ETR 23,094 0.081 0.135 0.178 4,310 0.061 0.119 0.168 1,798 0.070 0.121 0.170Adj_Fed_Cash_ETR 23,094 0.086 0.137 0.182 4,310 0.068 0.123 0.173 1,798 0.080 0.125 0.174Cash_ETR 23,094 0.211 0.204 0.459 4,310 0.206 0.205 0.459 1,798 0.218 0.213 0.454

Newly-assigned and Long-assigned samplePooled sample

Newly-assigned and Non-assigned sample

Notes:

***, **, * denotes two-tailed significance at 0.01, 0.05, and 0.10, respectively.

This table presents results comparing newly-assigned CIC firms to non-assigned firms and long-assigned firms. Our sample period includes fiscal years 2000 through 2011. Panel A presents univariate differences between newly-assigned CIC firms and non-assigned firms in the year of CIC assignment (Year t). Panel B presents univariate differences between newly-assigned CIC firms and long-assigned firms in the year of CIC assignment (Year t). Panel C presents descriptive statistics for Fed_Cash_ETR, Adj_Fed_Cash_ETR, and Cash_ETR. See Appendix 3 for variable definitions.

46

TABLE 5CIC Participation/Assignment and Initial Federal Tax Rates

Variable

Intercept 0.254 *** 0.223 *** 0.219 *** 0.215 *** 0.212 *** 0.221 *** 0.214 ***

(48.84) (15.34) (12.61) (20.51) (10.14) (8.21) (11.49)

CICParticipationInd -0.002(-0.45)

Post -0.001 -0.008 -0.008 0.005 0.015 0.013(-0.20) (-1.13) (-1.14) (0.46) (1.33) (1.17)

∆Firm 0.006 0.002(0.73) (0.11)

Post*∆Firm 0.006 -0.003(0.65) (-0.18)

AssetPoints -0.003 *** -0.002 0.004 ** 0.000 0.002 -0.006 *** -0.003(-2.76) (-1.03) (2.02) (0.24) (0.66) (-2.70) (-1.30)

GrossReceiptsPoints 0.011 *** 0.017 *** 0.020 *** 0.017 *** 0.010 ** 0.009 *** 0.008 ***

(7.51) (6.62) (7.41) (9.18) (2.24) (2.91) (2.94)

GeoSegPoints -0.015 *** -0.010 *** -0.014 *** -0.014 *** -0.012 *** -0.002 -0.009 ***

(-11.97) (-3.64) (-4.90) (-7.00) (-2.94) (-0.33) (-2.94)

BusSegPoints -0.002 -0.002 0.004 0.001 0.002 -0.010 ** -0.003(-1.46) (-0.66) (1.45) (0.27) (0.51) (-2.53) (-0.94)

ForeignSalesPoints -0.008 *** -0.015 *** -0.026 *** -0.016 *** -0.009 ** -0.009 *** -0.007 ***

(-4.38) (-5.91) (-7.85) (-8.22) (-2.47) (-3.61) (-3.47)

ForeignTaxPoints -0.0110 *** -0.011 *** -0.030 *** -0.014 *** -0.016 *** -0.007 ** -0.010 ***

(-4.43) (-3.13) (-4.78) (-4.82) (-3.45) (-2.15) (-3.99)

Leverage -0.113 *** -0.091 *** -0.084 *** -0.083 *** -0.142 *** -0.045 -0.097 ***

(-13.25) (-4.20) (-4.86) (-6.10) (-3.71) (-1.23) (-3.65)

R&D -0.365 *** -0.319 *** -0.319 *** -0.322 *** -0.304 *** -0.238 * -0.309 ***

(-16.28) (-6.65) (-11.02) (-13.94) (-5.30) (-1.74) (-5.28)

CapInt -0.097 *** -0.069 *** -0.087 *** -0.079 *** -0.061 ** -0.074 *** -0.057 ***

(-13.49) (-4.02) (-5.50) (-6.75) (-2.45) (-2.95) (-3.29)

ExcessStockBen 0.042 *** 0.046 *** 0.034 *** 0.041 *** 0.046 *** 0.024 * 0.034 ***

(13.02) (4.69) (3.44) (5.77) (3.65) (1.90) (3.70)

NOL -0.052 *** -0.048 *** -0.025 *** -0.038 *** -0.049 *** -0.013 -0.030 ***

(-15.33) (-6.90) (-3.97) (-7.87) (-4.61) (-1.08) (-3.85)

NR-squared

23,094 2,155 2,155

Panel A: Pooled

Panel B: Changes (Newly-assigned & Non-assigned)

14.90%4,310

14.59% 19.77% 16.89% 16.13%

∆FirmsNon-

∆Firms All Firms

899 899 1,7987.31% 10.21%

All Firms

Panel C: Changes (Newly-assigned & Long-assigned)

∆FirmsNon-

∆Firms All Firms

Notes:***, **, * denotes two-tailed significance at 0.01, 0.05, and 0.10, respectively.

This table presents results on the relation between CIC participation/assignment and Fed_Cash_ETR. Our sample period includes fiscal years 2000 through 2011. Panel A presents the results from estimating a pooled multivariate OLS regression and Panel B (C) presents the results from estimating a changes model for newly-assigned firms relative to non-assigned (long-assigned) firms. See Appendix 3 for variable definitions. t-statistics are reported in parentheses.

47

TABLE 6CIC Participation/Assignment and Adjusted Federal Tax Rates

Variable

Intercept 0.257 *** 0.239 *** 0.225 *** 0.223 *** 0.218 *** 0.231 *** 0.223 ***

(48.98) (15.51) (12.50) (20.33) (10.24) (8.45) (11.76)

CICParticipationInd 0.000(0.07)

Post -0.003 -0.008 -0.008 0.004 0.015 0.012(-0.36) (-1.11) (-1.13) (0.36) (1.27) (1.08)

∆Firm 0.011 0.000(1.33) (-0.02)

Post*∆Firm 0.005 -0.002(0.50) (-0.14)

AssetPoints -0.003 *** -0.003 0.005 ** 0.000 0.002 -0.006 ** -0.002(-2.85) (-1.55) (2.13) (-0.07) (0.62) (-2.54) (-1.22)

GrossReceiptsPoints 0.012 *** 0.018 *** 0.019 *** 0.017 *** 0.010 ** 0.009 *** 0.008 ***

(7.58) (6.86) (6.93) (9.07) (2.26) (2.62) (2.74)

GeoSegPoints -0.016 *** -0.011 *** -0.014 *** -0.015 *** -0.012 *** -0.002 -0.009 ***

(-11.96) (-3.82) (-4.93) (-7.11) (-2.79) (-0.41) (-2.89)

BusSegPoints -0.002 -0.002 0.004 0.000 0.002 -0.010 *** -0.003(-1.31) (-0.79) (1.38) (0.13) (0.49) (-2.62) (-1.04)

ForeignSalesPoints -0.008 *** -0.014 *** -0.026 *** -0.015 *** -0.009 ** -0.009 *** -0.007 ***

(-4.34) (-4.83) (-7.80) (-6.93) (-2.48) (-3.52) (-3.42)

ForeignTaxPoints -0.012 *** -0.013 *** -0.030 *** -0.015 *** -0.016 *** -0.007 ** -0.010 ***

(-4.69) (-3.15) (-4.75) (-4.77) (-3.34) (-2.20) (-4.00)

Leverage -0.115 *** -0.096 *** -0.086 *** -0.086 *** -0.146 *** -0.043 -0.098 ***

(-13.27) (-4.31) (-4.88) (-6.13) (-3.74) (-1.17) (-3.63)

R&D -0.365 *** -0.329 *** -0.324 *** -0.329 *** -0.308 *** -0.236 * -0.311 ***

(-15.83) (-6.47) (-10.88) (-13.66) (-5.04) (-1.70) (-5.10)

CapInt -0.096 *** -0.077 *** -0.086 *** -0.083 *** -0.063 ** -0.074 *** -0.057 ***

(-13.17) (-4.32) (-5.32) (-6.87) (-2.48) (-2.88) (-3.26)

ExcessStockBen 0.041 *** 0.043 *** 0.034 *** 0.039 *** 0.049 *** 0.022 * 0.034 ***

(12.65) (4.24) (3.30) (5.35) (3.74) (1.75) (3.64)

NOL -0.053 *** -0.050 *** -0.026 *** -0.038 *** -0.051 *** -0.014 -0.031 ***

(-15.37) (-6.82) (-3.99) (-7.82) (-4.64) (-1.07) (-3.86)

NR-squared 9.96%

1,79814.50% 14.11% 19.30% 16.50% 15.75% 7.08%23,094 2,155 2,155 4,310 899 899

All Firms ∆FirmsNon-

∆Firms All Firms ∆FirmsNon-

∆Firms All Firms

Panel A: Pooled

Panel B: Changes (Newly-assigned & Non-assigned)

Panel C: Changes (Newly-assigned & Long-assigned)

Notes:***, **, * denotes two-tailed significance at 0.01, 0.05, and 0.10, respectively.

This table presents results on the relation between CIC participation/assignment and Adj_Fed_Cash_ETR. Our sample period includes fiscal years 2000 through 2011. Panel A presents the results from estimating a pooled multivariate OLS regression and Panel B (C) presents the results from estimating a changes model for newly-

48

assigned firms relative to non-assigned (long-assigned) firms. See Appendix 3 for variable definitions. t-statistics are reported in parentheses.TABLE 7CIC Participation/Assignment and Cash Effective Tax Rates

Variable

Intercept 0.268 *** 0.286 *** 0.239 *** 0.276 *** 0.244 *** 0.335 *** 0.272 ***

(25.04) (7.00) (4.33) (8.97) (4.77) (3.95) (4.96)

CICParticipationInd 0.002(0.13)

Post 0.016 -0.011 -0.009 0.051 0.054 0.056 *

(0.78) (-0.52) (-0.43) (1.57) (1.63) (1.75)∆Firm 0.002 0.026

(0.10) (0.69)

Post*∆Firm 0.021 -0.004(0.76) (-0.09)

AssetPoints -0.008 *** -0.016 *** -0.002 -0.011 *** -0.014 ** -0.019 *** -0.017 ***

(-4.01) (-3.29) (-0.44) (-3.22) (-2.09) (-2.61) (-3.35)

GrossReceiptsPoints 0.011 *** 0.024 *** 0.020 *** 0.022 *** 0.022 ** 0.019 ** 0.021 ***

(4.17) (3.75) (3.03) (4.83) (2.19) (2.00) (3.08)

GeoSegPoints -0.004 0.005 -0.004 -0.001 0.019 0.011 0.015 *

(-1.16) (0.51) (-0.43) (-0.23) (1.54) (0.89) (1.79)

BusSegPoints -0.003 -0.008 0.013 * -0.001 -0.001 -0.030 *** -0.015 **

(-1.09) (-1.18) (1.79) (-0.16) (-0.11) (-2.93) (-2.11)

ForeignSalesPoints -0.007 ** -0.005 -0.024 *** -0.010 -0.005 -0.016 ** -0.012 **

(-2.00) (-0.56) (-2.61) (-1.60) (-0.50) (-2.08) (-2.01)

ForeignTaxPoints 0.0290 *** 0.021 * 0.031 0.026 *** 0.013 0.021 ** 0.019 **

(5.61) (1.83) (1.37) (2.67) (0.84) (2.03) (2.21)

Leverage -0.107 *** -0.061 -0.159 *** -0.108 ** -0.138 -0.038 -0.107(-5.67) (-0.8) (-2.75) (-2.36) (-1.10) (-0.31) (-1.22)

R&D -0.604 *** -0.817 *** -0.378 *** -0.548 *** -0.493 *** -0.527 -0.573 ***

(-11.42) (-4.63) (-3.40) (-6.34) (-2.69) (-1.21) (-3.04)

CapInt -0.111 *** -0.103 ** -0.144 *** -0.126 *** -0.044 -0.211 *** -0.115 **

(-7.41) (-2.16) (-3.24) (-3.88) (-0.67) (-3.28) (-2.54)

ExcessStockBen 0.056 *** 0.039 0.060 ** 0.047 ** 0.005 0.025 0.007(7.97) (1.50) (1.98) (2.36) (0.15) (0.74) (0.27)

NOL -0.059 *** -0.093 *** -0.070 *** -0.082 *** -0.115 *** 0.014 -0.053 **

(-7.74) (-4.42) (-3.37) (-5.61) (-3.80) (0.39) (-2.33)

NR-squared

Panel A: Pooled

Panel B: Changes (Newly-assigned & Non-assigned)

Panel C: Changes (Newly-assigned & Long-assigned)

All Firms ∆FirmsNon-

∆Firms All Firms ∆FirmsNon-

∆Firms All Firms

23,094 2,155 2,155 4,310 899 8992.50% 3.36% 2.90% 3.08% 2.76% 2.47% 2.15%

1,798

Notes: ***, **, * denotes two-tailed significance at 0.01, 0.05, and 0.10, respectively.

This table presents results on the relation between CIC participation/assignment and Cash_ETR. Our sample period includes fiscal years 2000 through 2011. Panel A presents the results from estimating a pooled multivariate OLS regression and Panel B (C) presents the results from estimating a changes model for newly-

49

assigned firms relative to non-assigned (long-assigned) firms. See Appendix 3 for variable definitions. t-statistics are reported in parentheses.

50

TABLE 8CIC Participation/Assignment and Management Expectations of Future Tax Outcomes

Panel A: Descriptive statistics for UTB_CY_ADD

N Median Mean SD N Median Mean SD N Median Mean SDUTB_CY_ADD 11,002 0.000 0.007 0.021 970 0.000 0.009 0.025 822 0.000 0.012 0.026

Pooled sample Non-assigned sample Long-assigned sample

Variable

Intercept -0.001 * 0.008 * 0.002 0.005 0.004 0.003 0.008(-1.74) (1.75) (0.57) (1.41) (0.76) (0.34) (1.58)

CICParticipationInd 0.004 ***

(3.21)Post 0.008 *** -0.001 -0.001 0.008 *** 0.001 0.001

(3.48) (-0.47) (-0.39) (2.78) (0.43) (0.28)∆Firm 0.001 -0.009 **

(0.24) (-2.26)

Post*∆Firm 0.010 *** 0.008 *

(2.74) (1.82)AssetPoints 0.000 *** -0.001 *** 0.000 -0.001 *** -0.001 * -0.001 -0.001 ***

(-3.16) (-2.79) (-0.53) (-2.84) (-1.93) (-1.38) (-2.59)

GrossReceiptsPoints 0.001 ** 0.001 0.001 0.000 0.000 0.000 0.000(2.49) (0.70) (1.59) (0.47) (0.12) (0.39) (0.33)

GeoSegPoints 0.001 *** 0.001 0.001 0.001 0.002 * 0.004 *** 0.003 ***

(4.96) (1.39) (0.58) (1.04) (1.73) (3.32) (3.51)

BusSegPoints 0.000 0.001 0.000 0.001 0.001 0.000 0.000(-0.04) (0.55) (0.34) (1.01) (0.77) (-0.16) (0.51)

ForeignSalesPoints 0.000 0.000 0.002 0.001 0.000 0.000 0.000(0.80) (0.03) (1.57) (1.59) (-0.01) (0.25) (0.54)

ForeignTaxPoints 0.002 *** 0.000 -0.002 0.000 0.000 -0.001 -0.001(2.71) (0.37) (-0.9) (-0.45) (0.29) (-0.97) (-0.80)

Leverage -0.001 -0.007 -0.003 -0.004 -0.001 0.026 ** 0.011(-0.63) (-1.15) (-0.75) (-1.14) (-0.15) (2.26) (1.63)

R&D 0.019 *** 0.063 ** 0.024 0.038 *** 0.077 *** 0.025 0.051 **

(2.77) (2.32) (1.44) (2.59) (2.81) (0.70) (2.37)

CapInt -0.002 * -0.010 ** -0.001 -0.004 -0.006 -0.013 ** -0.007 **

(-1.93) (-1.99) (-0.25) (-1.53) (-1.15) (-2.22) (-2.06)

ExcessStockBen 0.003 *** -0.002 0.001 0.000 -0.001 0.003 0.001(5.22) (-0.90) (0.60) (-0.10) (-0.40) (0.91) (0.43)

NOL 0.002 *** -0.002 -0.002 -0.002 -0.004 0.004 0.000(3.56) (-0.76) (-1.12) (-1.35) (-1.14) (1.19) (0.04)

NR-squared

Panel B: Pooled

Panel C: Changes (Newly-assigned & Non-assigned)

Panel D: Changes (Newly-assigned & Long-assigned)

All Firms ∆FirmsNon-

∆Firms All Firms ∆FirmsNon-

∆Firms All Firms

11,002 485 485 970 411 4116.30% 6.87% 0.63% 7.51% 8.29% 8.78% 7.79%

822

Notes:***, **, * denotes two-tailed significance at 0.01, 0.05, and 0.10, respectively.

51

This table presents results on the relation between CIC participation/assignment and UTB_CY_ADD. This analysis includes fiscal years 2007 through 2011. Panel A presents the results from estimating a pooled multivariate OLS regression and Panel B (C) presents the results from estimating a changes model for newly-assigned firms relative to non-assigned (long-assigned) firms. See Appendix 3 for variable definitions. t-statistics are reported in parentheses.

52