Corporate Taxpayer Responses to Size-Based Enforcement and

Corporate Taxpayer Responses to Size-Based Enforcement and

Disclosure Thresholds

Jason DeBacker

University of South Carolina

Jason.DeBacker@moore.sc.edu

Erin Towery

University of Georgia

etowery@uga.edu

Bibek Adhikari

Amazon Prime Science

bibek27@gmail.com

December 2025

ABSTRACT

This study employs administrative tax return and audit data to examine the effects of three Internal

Revenue Service (IRS) enforcement policies focused on large corporations. The threshold for IRS

Large Business and International Division monitoring increased from $5 million in 2000 and 2001 to

$10 million in assets starting in 2002. Beginning in the 2004 tax year, the IRS requires C corporations

with at least $10 million in assets to file Schedule M-3, which reconciles reported incomes in financial

statements and tax returns. In the same year, we observe a shift in audit policy resulting in a discrete

jump in audit rates at the $10 million asset threshold. We find that C corporations strategically bunch

below the $10 million asset threshold in most years from 2004 to 2010. Using variation in audit rates

around the $10 million threshold over time and the staggered Schedule M-3 implementation dates for

C corporations and S corporations, our evidence collectively suggests C corporations bunch below

the threshold primarily to avoid higher audit rates. We implement a triple difference estimator to show

that the enforcement notch at $10 million in assets has persistent effects on corporation size.

Keywords: Corporate income tax, tax avoidance, IRS, audit examinations, enforcement

JEL Classifications: H25, H26, H32, L25

We thank John Guyton, Pat Langetieg, and Annette Portz for their helpful comments. We are also grateful to Youssef

Benzarti, Katarzyna Bilicka, Anne Burton, Andy Garin, Timothy Harris, Baban Hasnat, Sean Higgins, Christos

Kotsogiannis, Dmitri Koustas, Hang Nguyen, Amanda Ross, Luca Salvadori, Kenneth Tester, Eric Zwick, and participants

at the Mannheim Taxation Conference, Online Public Finance Seminar, Shadow Economy, Tax Behaviour, and

Institutions Conference, National Tax Association Conference, Southern Economic Association Conference, and

seminars at Illinois State University, University of Exeter Tax Administration Research Centre, University of Hawaii,

Univesity of South Carolina, University of Texas at Arlington, University of Tennessee-Knoxville, and the U.S. Treasury’s

Office of Tax Analysis. The views expressed here are those of the authors and do not necessarily reflect the official view

of the Internal Revenue Service. This project was conducted through the Joint Statistical Research Program of the

Statistics of Income Division of the IRS. All data work for this project involving confidential taxpayer information was

done on IRS computers, by IRS employees, and at no time was confidential taxpayer data ever outside of the IRS

computing environment. DeBacker and Towery are IRS employees under an agreement made possible by the

Intragovernmental Personnel Act of 1970 (5 U.S.C. 3371-3376).

1. Introduction

In the U.S. and around the world, a small number of large firms generate a majority of taxable

business income, making them an important source of government revenue. Large firms also

withhold and remit employee income taxes, employee payroll taxes, and customer sales taxes, meaning

they play a crucial role in governmental tax collection as well (Slemrod and Velayudhan 2018). At the

same time, governments lose large sums of revenue due to businesses under-reporting their income.

In the U.S. alone, the tax gap — the difference between what businesses should pay if they fully

complied with the tax laws and what they pay in a timely manner — averages around $45 billion

annually (IRS, 2023).

Given the importance of large firms, tax authorities worldwide have implemented

enforcement initiatives specifically intended to encourage tax compliance by large corporations (Baer

2002; Baum et al. 2017). Despite the widespread use of size-based tax enforcement strategies, there

is surprisingly little evidence evaluating how businesses respond to such enforcement strategies.

Using administrative tax return and audit examination data for the universe of U.S.

corporations, we investigate how U.S. corporations respond to three important size-based

enforcement strategies initiated by the Internal Revenue Service (IRS). First, the IRS Large Business

and International (LB&I) Division monitors tax compliance for large businesses, which are defined as

those with at least $5 million in assets in 2000-2001 and $10 million in assets thereafter.1 Second,

beginning with 2004 tax years, the IRS required C corporations with at least $10 million in assets to

include Schedule M-3 with their federal income tax return.2 This schedule provides detailed

1 The predecessor to the IRS LB&I Division was the IRS Large and Mid Sized Business (LMSB) Division. Prior to 2002,

the LMSB Division monitored tax compliance for businesses with at $5 million in total assets. The IRS increased the

LMSB threshold to $10 million in assets in 2002. In 2010, the IRS reorganized the LMSB Division and changed the name

to the Large Business and International (LB&I) Division.

2 C corporations, governed by Subchapter C of the Internal Revenue Code, are taxable entities for U.S. tax purposes. Many

medium and large businesses, including almost all publicly traded companies, are organized as C corporations. However,

most of the firms relevant to this study are privately held C corporations without significant international presence. In

1

information about differences between pretax book income reported in the corporation’s financial

statements and taxable income reported in their federal income tax return. Third, using confidential

audit examination data, we observe a large and discontinuous increase in the audit rates for C

corporations with at least $10 million in assets beginning in tax year 2004. We exploit the discontinuity

in these enforcement policies around the $5 and $10 million asset thresholds to examine corporate

taxpayer responses to size-based thresholds.

Corporate taxpayers could potentially manipulate their reported assets to fall below the sizebased asset threshold if they perceive that enforcement and compliance costs increase above the

threshold. Using the bunching estimator from Kleven and Waseem (2013) on a series of annual crosssections derived from the universe of C corporation tax returns for tax years 2000 to 2010, we find

the following. First, we observe little evidence of excess bunching behavior around the $10 million

LMSB/LB&I asset threshold in 2002-2003, the first two years after the IRS increased the threshold

from $5 million in assets to $10 million in assets but without a meaningful increase in audit rates. We

also observe no evidence of bunching to avoid the LMSB/LB&I threshold in 2000-2001. Second, we

find evidence of economically significant excess bunching beginning in 2004-2005, the first two years

after the IRS began requiring Schedule M-3 and substantially increased audit rates for C corporations

above the $10 million asset threshold. Specifically, C corporations near the asset threshold reduce

their reported assets by about 0.5 million (about 5 percent of their reported assets) in 2004-2005.

Our finding of excess bunching below the $10 million asset threshold in the first two years

after the introduction of Schedule M-3 and the discrete jump in audit rates suggests that firms could

be bunching to avoid providing Schedule M-3 disclosures and/or to avoid higher audit rates.

Therefore, we next use the population of S corporations to investigate whether the Schedule M-3 or

2001, only about 3.8 percent of C corporations reported assets more than $5 million and just 2.5 percent of C corporations

reported assets more than $10 million.

2

changes in IRS audit rates around the $10 million asset threshold explain the bunching behavior.

Specifically, we estimate bunching behavior by S corporations with at least $10 million in assets

immediately after they were required to file Schedule M-3 beginning in 2006 (i.e., two years after

Schedule M-3 became effective for C corporations)3 If Schedule M-3 explains the bunching behavior

we observe for C corporations, we expect to also observe bunching behavior for S corporations

around the $10 million asset threshold in 2006. However, we observe no bunching behavior around

the $10 asset threshold for S Corporations, which suggests that Schedule M-3 does not explain the

bunching behavior we observe for C corporations. In contrast, we find that the degree of bunching

behavior in C Corporations is directly associated with differential audit rates at the $10 million asset

threshold; the elasticity of reported assets with respect to the audit rate is approximately 0.37,

suggesting that a one percentage point increase in audit rate around the threshold is associated with

firms reducing reported assets by an additional 0.37 percent. We observe no significant bunching

behavior in years where audit rates are similar just above and below the threshold. These analyses

collectively suggest that the bunching behavior we observe reflects C corporations responding to

higher audit rates at the $10 million asset threshold rather than to Schedule M-3 disclosure

requirements.

In our final two sets of analyses, we investigate the role of tax aggressiveness in bunching

behavior and whether C corporations’ responses to the thresholds persist over time. Despite higher

audit rates and bunching around the $10 million threshold, we find no evidence that C corporations

3 Unlike C corporations, S corporations elect to pass corporate income through to their shareholders, which avoids the

double taxation of corporate income. However, relative to C corporations, S corporations are more restrictive in terms of

the number of shareholders, the citizenship of shareholders, and the classes of stock offered. Yagan (2015) provides

evidence that C corporations and S corporations of the same age compete in the same narrow industries and at the same

scales, with the exception of the very largest corporations, which are all publicly traded C corporations far above the

LMSB/LB&I thresholds relevant for this study. However, recent studies have documented that S corporations are

monitored much less by the IRS than C corporations and thus likely respond to tax enforcement policies differently

(Adhikari et al., 2021; Guyton et al., 2021). Therefore, following Yagan (2015), we use S corporations as a plausible

comparison group for C corporations.

3

in the bunching region claim more aggressive tax positions, as measured by effective tax rates (ETRs).

To investigate the persistence of C corporations’ responses to the asset thresholds, we estimate a triple

differences model, comparing C and S corporations of similar size before and after 2004. The

estimates show that C corporations who are below the $10 million threshold before 2004 have total

assets that are $1 million dollars lower than similarly sized S corporations in the years after 2004.

Overall, our findings are consistent with corporations bunching below the asset threshold to avoid

higher perceived audit rates and the direct costs associated with undergoing an audit, which is

consistent with models of tax salience and risk-as-feelings (Chetty et al. 2009; Sandmo 2012; Bergolo

et al. 2023).

Our study contributes to the literature in multiple ways. First, prior studies that have employed

bunching techniques focus mostly on taxpayer responses to ‘notches’ that create discrete changes in

tax liability due to a change in tax rates or a change in tax regime above a certain threshold.4 However,

there are very few studies estimating bunching responses generated by a discrete change in an

enforcement regime. Two exceptions are Almunia and Lopez-Rodriguez (2018), who study corporate

taxpayer bunching behavior in response to the revenue-based large business threshold in Spain, and

Tennant and Tracey (2019), who study the sales-based large business threshold in Jamaica. The

discrete change in monitoring efforts without an accompanying change in tax rates or tax regimes

allows the authors to examine bunching responses generated by tax enforcement policies in isolation.

However, because these enforcement policies are based on revenue or sales, they increase firms’

incentives to under-report revenue or sales as doing so can reduce both the tax liability and the

4 The change is tax liability is most commonly created by a change in tax rates or a change in tax regime (e.g., transitioning

from a sales tax to a Value-Added Tax) above a specified threshold. For instance, Saez (2010) and Chetty et al. (2011)

study bunching at kink points where the marginal tax rate increases, and Kleven and Waseem (2013) study bunching

around notches where the average tax rate increases discretely. With respect to business income taxation, Onji (2009) and

Harju et al. (2019) study firms’ bunching responses to avoid the VAT regime, Best et al. (2015) study firms’ bunching

responses to the threshold where the tax base switches from turnover to profit, and Bachas and Soto (2021) study firms’

bunching responses to certain profit thresholds where the average corporate tax rate increases discretely.

4

probability of audit. In contrast, the U.S. enforcement thresholds we examine are based on assets.

Because under-reporting of assets does not directly reduce tax liability, we exploit a unique opportunity

to cleanly estimate the corporate responses to changes in enforcement efforts and disclosure

requirements separate from other incentives to report lower assets.

Second, size-based thresholds in previous studies only affect the enforcement efforts of the

tax authorities. However, our size threshold applies to both enforcement (e.g., audit rates) and

disclosure requirements (e.g., Schedule M-3), but with changes in the threshold value, implementation

year, and across C corporations and S corporations. We use these variations to provide evidence on

the impact of higher enforcement, greater disclosure requirements, and the interaction between higher

enforcement and greater disclosure (Harju et al. 2019). In doing so, we complement studies such as

Basri et al. (2021) and Small and Brown (2020), who find that increased tax enforcement for medium

and large business can have significant effects on tax revenue.

Finally, previous corporate bunching studies do not employ tax audit examination data, so

they are unable to document actual changes in the audit rate between taxpayers just above and below

the threshold. We observe the audit rate differentials using confidential IRS audit examination data

around the $10 million asset threshold. We report substantial variation over the years in the audit

rates between firms just above and just below the threshold, and we find that the degree of bunching

behavior is directly associated with the changes in the audit rate. This could help explain why Almunia

and Lopez-Rodriguez (2018) find that Spanish firms bunch just below the enforcement threshold of

€6 million, while Tennant and Tracey (2019) find that Jamaican firms do not bunch below the

enforcement threshold of J$500 million.

2. Background Information

5

This section provides relevant background information for the IRS LMSB/LB&I Division,

the Schedule M-3, and IRS’s enforcement efforts focused on large corporations over time.

2.1 Large and Mid-Sized Business (LMSB)/Large Business and International

(LB&I) Division

Prior to 2000, the IRS primarily categorized taxpayers by geographic region, regardless of size

and regardless of whether the taxpayer was an individual taxpayer or a business taxpayer.5 The Internal

Revenue Service Restructuring and Reform Act of 1998 mandated the modernization of the IRS’s

organizational structure in order to meet the needs of taxpayers and to provide them with better

customer service. In 2000, the IRS responded to the Congressional mandate by reorganizing its

organizational structure into four divisions based on taxpayer type and taxpayer size: (i) Wage and

Investment, (ii) Small Business/Self-Employed, (iii) Large and Mid-Size Business, and (iv) TaxExempt and Government Entities.

Although the Large and Mid-Sized Business (LMSB) Division initially monitored business

taxpayers with at least $5 million in total assets, the IRS increased the LMSB threshold to $10 million

in total assets in 2002. Corporations with $10 million or more in assets are among the 2% largest

corporations. In 2010, the IRS reorganized the LMSB Division and changed its name to the Large

Business and International (LB&I) Division to reflect the increased focus on international tax

compliance issues (Laughlin and Allison 2001; Journal of Accountancy 2010). The LB&I Division has

more resources per taxpayer for enforcement efforts and can therefore potentially audit a higher

fraction of the taxpayers under its purview.

2.2 Schedule M-3

5 A notable exception is the Coordinated Industry Case Program (formerly known as the Coordinated Examination

Program), which was created in 1966 to audit the largest and most complex multinational corporations. In 2019, the IRS

replaced the Coordinated Industry Case program with the Large Corporate Compliance program.

6

A taxpayer’s book income reported in its financial statements often differs from taxable income

reported on its tax return because the former is governed by generally accepted accounting principles

(GAAP), while the latter is governed by tax law. In the 2000s, the Treasury expressed concerns that

large differences between book income and taxable income (referred to as book-tax differences) were

likely indicative of tax sheltering activity. Prior to 2004, corporate taxpayers reconciled their book

income to their taxable income on Schedule M-1, Reconciliation of Income (Loss) Per Books with Income Per

Return. However, Schedule M-1 was not sufficiently granular to aid the IRS in assessing tax compliance

risk, which suggested the need for an expanded book-tax difference reconciliation (Mills and Plesko,

2003). Thus, in 2004, the IRS created Schedule M-3, Net Income (Loss) Reconciliation for Corporations with

Total Assets of $10 Million or More, to better understand book-tax differences. The evidence on whether

the implementation of Schedule M-3 affected corporate tax avoidance is somewhat mixed (Donohoe

and McGill 2011; Hope et al. 2013; Henry et al. 2016).

Schedule M-3 became effective for tax years ending on or after December 31, 2004 for C

corporations with at least $10 million of total assets. S corporations were required to file Schedule M3 for tax years ending on or after December 31, 2006. Schedule M-3 is comprised of three parts. Part

1 provides financial information and reconciles worldwide book income per the corporation’s financial

statements with the book income of entities reported on the consolidated Form 1120. Part 2 details

specific income-related book-tax differences and Part 3 details specific expense-related book-tax

differences; Parts 2 and 3 also require firms to disclose whether each book-tax difference is classified

as permanent or temporary.6 Corporations are only required to complete Part 1 of Schedule M-3 in

the first year they file Schedule M-3.

6 Permanent book-tax differences represent items that are reported in book income or taxable income, but not both.

Temporary book-tax differences represent items that are reported in both book income and taxable income, but in

different reporting periods.

7

There have been few changes to Schedule M-3 since its introduction. In 2008, Part 1 of Schedule

M-3 was expanded to reconcile the assets and liabilities included in the worldwide financial statements

with the assets and liabilities included in the consolidated tax return. Finally, beginning in 2014,

corporations with at least $10 million in total assets but less than $50 million in total assets were only

required to complete Part 1 of Schedule M-3 and could complete Schedule M-1 in lieu of Parts 2 and

3 of Schedule M-3.

2.3 IRS Audit Rates for Large Corporations

The IRS reports audit examination statistics for up to thirteen groups of C corporations

ordered by asset size, but only aggregate audit statistics for all S corporations in its Databook, an

annual publication released every March. Among other things, the Databook reports audit rates, the

percentage of tax return audits that did not result in proposed changes to the taxpayer’s reported tax

liability, and the magnitude of additional tax liabilities recommended after audit for the previous fiscal

year.7 These statistics enable corporations to observe information about the IRS audit activity in their

respective size categories, suggesting that they can be aware of any changes to audit activity on a timely

basis. Indeed, in a survey of managers, Hoopes et al. (2012) find that a substantial number of managers

use historical data provided by the tax authority to gauge tax enforcement.

Our administrative data allow us to have a more granular view of IRS audit policies. Using IRS

audit data, we document a discrete jump in the probability of audit for C corporations with assets of

$10 million of more beginning in tax year 2004. Figure 1 shows this change in audit policy from 20022003 to 2004-2005. The differential in audit rates between corporations just above and just below the

7 Although the size categories have changed over time, the thirteen categories generally include: no reported assets, assets

below $250 thousand, assets between $250 thousand and $1 million, assets between $1 million and $5 million, assets

between $5 million and $10 million, assets between $10 million and $50 million, assets between $50 million and $100

million, assets between $100 million and $250 million, assets between $250 million and $500 million, assets between $500

million and $1 billion, assets between $1 billion and $5 billion, assets between $5 billion and $20 billion, and assets above

$20 billion.

8

$10 million asset threshold grew from 7.5 percentage points in 2002-2003 to 13.7 percentage points

in 2004-2005.8

Figure 1: Audit Rates Among C Corporations with Assets of $7 to $13 Million, Tax Years

2002-2005

Notes: This figure shows the mean audit rates (connected black dots) and the 95% confidence interval (shaded

areas) around these means for C corporations with assets between $7 million and $13 million. The data are grouped into

$200,000 bins. The vertical red line indicates $10 million in assets. The series with circle markers is for years 2002 and

2003 and the series with triangle markers is for years 2004 and 2005.

The degree to which audit rates increase around this threshold varies over time (see Figure 4

and Table A.3.1 for the year-by-year differentials). We exploit this variation to disentangle the effects

of the changes in disclosure and enforcement around $10 million that both appear in tax year 2004.

3. Empirical Approach

We investigate whether corporations respond to additional enforcement or disclosure by

8 Recall that the LMSB/LB&I threshold changed from $5 million in assets to $10 million in 2002.

discontinuity in audit rates at $5 million before 2002.

9

We do not observe a

examining whether C corporations bunch just below asset thresholds to avoid additional disclosure,

stronger enforcement, or both.9 To quantify C corporations’ responses to the notch created by these

thresholds, we use the bunching techniques developed by Kleven and Waseem (2013) and applied in

Almunia and Lopez-Rodriguez (2018). Kleven and Waseem (2013) extend the classic bunching

estimator of Saez (2010) and Chetty et al. (2011) that was developed in the context of kinks so that it

can be used in the presence of notches. This bunching approach compares the observed distribution

of reported assets to the counterfactual distribution that would have been observed in the absence of

the notch. To estimate the counterfactual distribution, we fit a flexible polynomial to the observed

distribution, excluding data in the interval around the threshold. We determine the interval to exclude

by dividing the data into small bins of width w and then estimate the local polynomial regression of

the following form:

𝑎𝑢𝑏

𝑞

𝑖

𝐹𝑗 = ∑ β𝑖 . (𝑎𝑗 ) + ∑ γ𝑘 . 1(𝑎𝑗 = 𝑘) + ϵ𝑗 ,

𝑖=0

(1)

𝑘=𝑎𝑙𝑏

where 𝐹𝑗 is the number of firms in asset bin 𝑗, 𝑞 is the order of the polynomial, 𝑎𝑗 is the midpoint of

the bin 𝑗, 𝑎𝑙𝑏 is the lower bound of the excluded region, 𝑎𝑢𝑏 is the upper bound of the excluded

region, and 𝛾𝑘 are the intercept shifters for each of the bins in the excluded interval.

The predicted values from equation (1) represent the counterfactual bin counts, and we omit

the intercept shifters in the excluded interval to ensure that the counterfactual distribution is smooth

around the threshold. The following equation provides the estimated counterfactual distribution.

𝑞

̂𝑗 = ∑ β̂𝑖 . (𝑎𝑗 )𝑖

𝐹

𝑖=0

9 Appendix A.1 provides a theoretical framework that would predict bunching behavior in this context.

10

(2)

We then estimate the excess mass due to bunching (B) on the left side of the asset threshold

using equation (3) and the missing mass (H) on the right side of the asset threshold using equation (4).

𝑎𝐿

̂𝑗 ) ≥ 0

𝐵̂ = ∑ (𝐹𝑗 − 𝐹

(3)

𝑗=𝑎𝑙𝑏

𝑎𝑢𝑏

̂ = ∑ (𝐹

̂𝑗 − 𝐹𝑗 ) ≥ 0

𝐻

(4)

𝑗=𝑦 𝐿

We determine the lower and upper bounds of the excluded region by requiring the area under

the counterfactual distribution and the area under the observed distribution to be equal, implying that

the bunching mass area (B) must be equal to the missing mass area (H). This assumption implies that

all responses to the thresholds are on the intensive margin, meaning corporations do not exit or split

into multiple smaller corporations to avoid crossing the thresholds (Kleven and Waseem 2013).

To quantify the bunching estimates, we first choose the lower bound (𝑎𝑙𝑏 ) at the bin where

the density of reported assets switches from decreasing to increasing due to the bunching response.

Next, we set the upper bound at 𝑎𝑢𝑏 ≈ 𝑎𝐿 and iteratively estimate equation (1), increasing the value

of 𝑎𝑢𝑏 by a small amount after each iteration. This procedure iterates until reaching the value of 𝑎𝑢𝑏

such that estimated bunching mass (i.e., 𝐵̂) is approximately equal to the estimated missing mass

(i.e., 𝐻̂).

To conduct inference, we estimate the standard errors using bootstrapping techniques. We

generate a large number of asset distributions by randomly resampling the residuals from equation (1)

with replacement, which produces counterfactual densities based on resampled distribution. We then

use the counterfactual densities to test the null hypothesis that there is no excess mass at the notch

relative to the counterfactual distribution.

11

4. Data

Identifying excess mass using the bunching estimator requires a sufficiently large number of

observations near the threshold of the notch. We overcome this hurdle by using administrative data

from the IRS that includes the universe of C corporation income tax returns. In our main bunching

analysis, we use total assets as reported on Form 1120, Box D. For other analyses, we gather data on

income and deduction items, taxes paid, and other items on the tax return. In addition to the corporate

tax return data, we also obtain audit examination data from the IRS to compute audit rates and to

obtain audit adjustment amounts.

Our research design exploits (i) variation in the cross-section by comparing the distribution of

corporations on either side of the asset threshold, and (ii) variation in the time series by comparing

the bunching estimator across years. Recall that the LMSB/LB&I threshold changed from $5 million

to $10 million in 2002, the Schedule M-3 became effective for C corporations in 2004 and for S

corporations in 2006, and IRS audit policies vary across time. We therefore examine corporate tax

returns for tax years 2000 through 2010. From these data, we drop Regulated Investment Companies

(RICs) and Real Estate Investment Trust Funds (REITs) because they face different tax incentives

than traditional C corporations. This yields a sample of 9,684,144 C corporation-year observations.

We focus the bunching estimator on C corporations near the threshold of $10 million in total assets.

Our data contain 129,703 corporation-year observations with total assets between $7 million and $13

million.

Table 1 provides descriptive statistics for C corporations with assets between $7 and $13

million, as well as for those below the threshold (i.e., $7 million to $10 million) and those above the

threshold (i.e., $10 million to $13 million). The average C corporation reports $19 million in gross

receipts, $5.4 million in deductions, and $0.77 million of net income. When comparing C corporations

just above and below the threshold, we observe that corporations just above the $10 million threshold

12

have modestly larger values for the “corporation size” variables. They also pay larger total taxes (i.e.,

$0.25 million) compared to corporations below the threshold (i.e., $0.18 million). IRS audit rates

increase from 6 percent below the threshold to 15 percent above the threshold. Return on assets,

effective tax rates, the share of foreign-controlled corporations, and the share of multinational

corporations are very similar above and below the threshold. Similarly, there are no meaningful

differences in the two-digit industry composition between C corporations just below and just above

the threshold.

To summarize, C corporations just above the threshold are, unsurprisingly, modestly larger

than the firms below the threshold, but other economic characteristics are very similar. However, IRS

scrutiny is much higher for C corporations above the threshold than for C corporations below the

threshold.

13

Table 1: Means of Variables of Interest for C Corporations Near the $10 Million Asset

Threshold, Tax Years 2000-2010

Notes: This table provides means of variables of interest for C corporations with assets between $7 and $13 million as

well as for those below the threshold (i.e., $7 million to $10 million) and those above the threshold (i.e., $10 million to

$13 million). The data is pooled across tax years 2000 to 2010. The return on assets has been winsorized at the 1% level

to reduce the effects of outliers.

5. Main Results

5.1 Bunching Responses to LMSB/LB&I monitoring

In 2002, the LMSB/LB&I threshold increased from $5 million in total assets to $10 million in

total assets. We expect monitoring efforts to be higher for firms under the purview of the

LMSB/LB&I Division because enforcement resources per taxpayer tend to be larger among

LMSB/LB&I taxpayers. Theory predicts that corporations will respond to a notch in enforcement

14

effort by reporting lower assets to avoid the additional monitoring. If the LMSB/LB&I Division’s

increased monitoring is costly for C corporations, we expect to see a behavioral response from the

corporations subject to the increased scrutiny they face when under the purview of the LMSB/LB&I

Division. We test this prediction by examining the distribution of corporations around the $10 million

asset threshold. Because all corporations with at least $5 million of assets are under the purview of

the LMSB/LB&I Division in 2000-2001, our model does not predict bunching around $10 million in

those years.

Figure 2 presents the results of this analysis, where we narrow our sample to observations with

total assets between $7 million and $13 million when considering bunching around the $10 million

threshold and $2 to $8 million when considering bunching around the $5 million threshold. The top

left panel shows the distribution of corporations for tax years 2000-2001 and the top right panel shows

the distribution for tax years 2002-2003. These panels show no evidence of bunching in either the

2000-2001 or 2002-2003 period around $10 million. The black lines represent the actual distribution

of corporations and the blue lines represent the counterfactual distribution of corporations (i.e., what

the distribution would be expected to look like if there were not a $10 million asset notch). The actual

and counterfactual distributions overlap entirely suggesting that corporations did not respond to the

$10 million asset threshold, either before or after it became the LMSB/LB&I Division threshold.

Estimates of the excess mass, b, are near zero and statistically insignificant both in 2000-2001 and

2002-2003. This evidence is inconsistent with corporations manipulating their assets to avoid being

monitored by the LMSB/LB&I Division. The figure in the bottom panel shows the actual and

counterfactual distributions around the $5 million LB&I threshold in 2000-2001. Again, we find no

evidence of bunching around the LB&I threshold even when it was at this lower value.

15

Figure 2: Distribution of C Corporations Around $10 Million in Assets, Tax Years 2000-2003

2000-2001 (Before $10 Million LB&I)

2002-2003 (After $10 Million LB&I)

2000-2001, Centered at $5 Million LB&I Threshold

Notes: This figure shows the reported distribution of assets (connected black dots) and the estimated counterfactual

(smooth blue line) for C corporations with assets between $7 million and $13 million, or between $2 and $8 million. The

counterfactual distribution is obtained by fitting a flexible polynomial of order 4, excluding data in the interval around the

threshold. The data are grouped into $50,000 bins. The vertical red dashed line indicates $10 million in assets in the top

two panels and $5 million in the bottom panel. The vertical blue dotted line indicates the area to the left of the threshold

used in the estimator. The figure on the top left is for the years before the LMSB/LB&I threshold increased to $10 million

and the figure on the top right is for the years after the LMSB/LB&I threshold increased to $10 million. The bottom

figure is for 2000-2001 when the LMSB/LB&I threshold was at $5 million.

16

5.2 Bunching Responses to Additional Disclosure Requirements and Increased

Audit Intensity

As previously discussed, C corporations with at least $10 million of assets were required to

complete Schedule M-3 beginning with the 2004 tax year. In addition, we have observed a change in

audit strategy beginning in tax year 2004 (see Figure 1). Therefore, beginning in tax year 2004, C

corporations were facing both a discrete change in reporting requirements (with Schedule M-3) and a

discrete change in audit intensity right at the $10 million threshold.

To understand whether corporations are responding to these reporting and enforcement

initiatives, we employ the bunching estimator described previously. Figure 3 presents the distribution

of corporations in 2002-2003 (which we show above but repeat here for comparison) and 2004-2005.

We observe no evidence of bunching in the 2002-2003 distribution, but we observe an increased

number of corporations reporting assets just below $10 million and a missing mass just above $10

million in the 2004-2005 distribution.10 The estimate of the excess mass, b, in 2004-2005 is 0.509 and

it is significant at the 1 percent level. In other words, corporations near the $10 million asset threshold

reduce their reported assets by about $0.5 million (~5 percent), which represents an economically

significant response to changes in enforcement and disclosure at the threshold.

10 Schedule M-3 is more salient to corporations than LMSB/LB&I designations because corporations do not need to take

any action or file any forms to be part of the LMSB/LB&I. However, after Schedule M-3's introduction, C corporations

above the threshold need to file Schedule M-3 to comply with the tax law. This could also lead to higher bunching after

Schedule M-3 introduction relative to just LMSB/LB&I designation.

17

Figure 3: Distribution of C Corporations Around $10 Million in Assets, Tax Years 2002-2005

2002-2003 (Before M-3)

2004-2005 (After M-3)

Notes: This figure shows the reported distribution of assets (connected black dots) and the estimated counterfactual

(smooth blue line) for C corporations with assets between $7 million and $13 million. The counterfactual distribution is

obtained by fitting a flexible polynomial of order 4, excluding data in the interval around the threshold. The data are

grouped into $50,000 bins. The vertical red dashed line indicates $10 million in assets. The vertical blue dotted lines indicate

the area to the left of the threshold used in the estimator and the estimated area to the right of the threshold with the

missing mass. The figure on the left is for the years before Schedule M-3 was introduced and the figure on the right is for

the years after Schedule M-3 was introduced.

The manipulation of assets to remain under the $10 million threshold suggests that

corporations are trying to avoid more reporting or an increased risk of audit – or both. Given the

simultaneity of these changes, this cross-sectional estimator cannot identify which mechanism is

driving the response. Thus, in the next section, we conduct tests to better understand the mechanism.

6. Mechanisms: What are Corporations Responding to?

In this section, we conduct additional analysis to better understand why C corporations bunch below

the $10 million threshold in tax years 2004 and 2005.

6.1 Bunching to avoid audit risk

18

Figure 1 shows a change in audit policy in 2004-2005, with a much larger jump in audit rates

at $10 million in assets after 2004. In Figure 4, we report these audit rate differentials in each year

2000 to 2010 (dashed lined with triangle marker), along with the estimate of excess bunching in those

years (solid line with circle markers).11

Figure 4: Bunching and audit differentials by year, 2000-2010

The audit rate differential for C corporations just above versus just below the $10 million asset

threshold peaked in 2004 at 17 percentage points, remained high through 2008 (14.4 pp), and then fell

after 2009 (4 pp). These fluctuations in audit policy are somewhat correlated with IRS resources over

time; the IRS Databook documents the enforcement budget increasing by more than 20% from 2000

to 2010 and falling sharply thereafter. However, we are the first to document the abrupt change in

audit rates at the $10 million threshold, and current IRS personnel to whom we have spoken are

unsure of the reason for why the 2004-2008 period saw such a notch in audit probability at $10 million.

11 Table A.3 includes the point estimates and standard errors for the data in Figure 4.

19

Figure 4 shows that, regardless of the reason for the change in audit policy over time,

corporations are responding to it. In years with higher audit rate differentials, we see stronger

behavioral responses from corporations. And when audit rate differentials become small again in

2009-2010, the bunching behavior disappears. This time series of bunching estimates and audit rate

differentials is therefore strongly suggestive of corporations responding to higher audit rates, rather

than the additional disclosure requirements of Schedule M-3, which do not vary during the 2004-2010

period.

6.2 Bunching to Avoid Schedule M-3 Disclosure

In this subsection, we extend our analysis to S corporations to test whether they are responding

to Schedule M-3 requirements. If they are not, it further strengthens the suggestive evidence above

that C corporation behavior post-2004 is driven by increased audit risk. S corporations with at least

$10 million of assets were required to file Schedule M-3 with their returns beginning with 2006 tax

years (i.e., two years later than C corporations). Moreover, unlike C corporations, S corporations’

audit rates are very low (below 5 percent) and exhibit no discontinuity at $10 million. Because S

corporations have consistently low audit rates over time, any bunching behavior at the $10 million

threshold after 2005 can be attributable to Schedule M-3.

Table 2 summarizes the results for S corporations by presenting the estimate of excess

bunching and audit rate differentials for S corporations from 2000-2011. We report these statistics in

2-year bins for increased precision.12 In all years the audit rate differential is quite small, peaking at

1.37 percentage points in 2008-2009. These years are also the only year with an economically

meaningful and statistically significant estimate of excess bunching. In 2006-2007, the years

immediately after Schedule M-3 was required among S corporations with $10 million of more in assets,

12

There are fewer than half as many S corporations around $10 million in assets as compared to C corporations.

20

and in 2010-2011, we observe no behavioral responses among S corporations to the $10 million

threshold. Again, these patterns suggest that audit risk is the salient feature to corporate taxpayers.

Table 2: S Corporation Bunching Estimates and Audit Rate Differentials

Years

2000-2001

2002-2003

2004-2005

2006-2007

2008-2009

2010-2011

b

Std. Err.

Audit Rate

Differential (pp)

0.125

0.124

0.096

0.079

0.111

0.112

1.13

-0.60

0.91

-0.18

1.37

0.14

0.038

-0.100

0.106

0.103

0.276

0.043

6.3 Are Corporations Under-reporting Assets to Increase Tax Aggressiveness?

Thus far, we have documented excess bunching among C corporations in response to the $10

million asset threshold in tax years 2004-2008 and we have provided evidence suggesting that this is

likely due to a discontinuous jump in the audit rate at the threshold. We now turn to the question of

whether C corporations are reporting assets that subject them to lower monitoring efforts in order to

claim more aggressive tax positions, which lower their tax liabilities.

To answer this question, we consider how a C corporation’s effective tax rate (ETR) varies

around the $10 million asset threshold. This will provide visual evidence of whether C corporations

that face lower monitoring efforts report lower ETRs because they are claiming more aggressive

positions on their tax returns. We follow Plesko (2003) and define the ETR as the ratio of taxes paid

to taxable income.13

13 More specifically, we define ETR as the ratio of taxes paid before net operating losses, to taxable income before net

operating losses and special deductions plus deductions for taxes paid, interest expense, depreciation. There are broader

ETR measures that use book income, but this measure has the advantage of being comprised of only items on Form 1120

and is therefore available for all C corporations, not just larger corporations required to complete Schedule M-3.

21

Figure 5 plots effective tax rates by reported assets for years 2002-2003 and 2004-2005. Recall

that the audit rate differentials between small and large corporations peaked in 2004-2005 and there

was significant excess bunching in those years. Despite this result, ETRs do not show any

discontinuous jump at the $10 million asset threshold and are quite similar between small and large

corporations within this range of assets.

We interpret this as evidence against the hypothesis that C corporations are reporting lower

assets to avoid monitoring to take aggressive tax positions. This result is not inconsistent with the

economic theory. The model we present in the Appendix A.1 predicts that both under-reporting

assets and under-reporting income increase the effectiveness of the enforcement technology (e.g., by

leaving paper trails that raise red flags to the IRS). Therefore, if the audit rate sharply increases above

the $10 million threshold, then it is possible for C corporations to prioritize under-reporting assets to

fall below the $10 million threshold and reduce under-reporting of income to reduce the probability

of raising red flags. Further, there are non-trivial expected costs of being audited even for corporations

with average levels of tax aggressiveness. Consider the direct costs of compliance with an audit: a

typical audit of corporations of around $10 million in assets takes 67 hours by an IRS agent. The

taxpayer’s hours spent on audits are probably at least as large. If we take the conservative stance that

the taxpayers hours to comply with the audit are equal to the auditors time and that each hour is worth

$100 to the corporation, then the costs of complying with an audit are about $6,700. Moreover, C

corporations of this size have an average adjustment to tax liability of approximately $50,000. Because

the average audit rate differential at $10 million is roughly 10 percentage points from 2004-2007, this

means the expected increase in tax liability from moving past the threshold is around $5,670. Taking

the expected time cost and additional tax liability together, there are significant benefits of avoiding

higher audit rates, even for C corporations that are not pushing the envelope with tax avoidance.

22

Figure 5: Effective Tax Rates (ETRs) among C Corporations with Assets of $7

to $13 Million, Tax Years 2002-2005

2002-2003

2004-2005

Notes: This figure shows the mean effective tax rates (connected black dots) and the 95% confidence interval (dashed gray

lines) around these means for C corporations with assets between $7 million and $13 million. The data are grouped into

$200,000 bins. The vertical red line indicates $10 million in assets. The figure on the left is for the years 2002 and 2003

and the figure on the right is for the years 2004 and 2005.

6.4 Do lower reported assets represent a real or reporting response?

It is difficult to infer whether changes in reported assets are real or reporting responses using tax

return data. However, we perform the following untabulated analysis to find suggestive evidence of

how corporations are responding to changes in enforcement. First, we test for differences in

balance sheet items that may be more easily misreported (such as intangibles) among C corporations

in the bunching region. Our data (IRS research database) do not record every line item from

Schedule L and excludes intangibles, but in tests of each of those that are reported, we found no

differences in the amounts reported relative to total assets between C corporations in the bunching

23

region and those outside of it.14 Second, we examine differences in the return on assets among C

corporations in the bunching region versus those outside of it. If the responses are pure reporting

responses, then we should observe a higher return on reported assets for corporations in the

bunching region (so long are these corporations aren’t underreporting earnings by at least as much

as they underreport assets), who are reporting fewer assets but generate earnings from a larger

amount of true assets. We find no evidence of higher ROA among corporations in the bunching

region, suggesting that the bunching behavior we observe is most consistent with a real response; C

corporations accumulating fewer assets to avoid the higher audit rates at $10 million in assets. We

reiterate that our evidence is only suggestive. Future research could pair tax return data with other

firm-level micro-data, such as administrative data from the US Census, to better understand these

responses.

7. Persistent Impacts on Corporate Growth

To understand whether the bunching behavior we observe represents a one-time reporting response

by C corporations or a salient threshold that has persistent effects on firm size, we estimate a triple

differences model that compares C and S corporations just above and below the $10 million threshold

in 2003 in the periods 2000-2003 and 2004-2010. Using a balanced panel of C and S corporations

from 2000-2010 that had between $8m and $12m in assets in 2003, we estimate the following triple

difference model:

𝑌𝑖𝑡 = 𝛽1 𝑃𝑜𝑠𝑡𝑡 ∗ (𝐴𝑠𝑠𝑒𝑡𝑠 < $10𝑚)𝑖 + 𝛽2 𝑃𝑜𝑠𝑡𝑡 ∗ 𝐶 − 𝑐𝑜𝑟𝑝𝑜𝑟𝑎𝑡𝑖𝑜𝑛𝑖 + 𝛽3 𝑃𝑜𝑠𝑡𝑡

∗ (𝐴𝑠𝑠𝑒𝑡𝑠 < $10𝑚)𝑖 ∗ 𝐶 − 𝑐𝑜𝑟𝑝𝑜𝑟𝑎𝑡𝑖𝑜𝑛𝑖 + 𝛼𝑖 + 𝛿𝑡 + 𝜀𝑖𝑡

The variable 𝑃𝑜𝑠𝑡𝑡 is an indicator variable that equals one for years after 2003, 𝐶 − 𝑐𝑜𝑟𝑝𝑜𝑟𝑎𝑡𝑖𝑜𝑛𝑖 is

14 Our data include records for Schedule L lines 7, 10b, 18d, 19d, 21b, 21d, 22b, 25b, and 28d.

24

an indicator variable equal to one if the corporation is a C corporation and zero if the corporation is

an S corporation, and (𝐴𝑠𝑠𝑒𝑡𝑠 < $10𝑚)𝑖 is an indicator variable equal to one if assets are below $10

million in 2003. The parameters 𝛼𝑖 and 𝛿𝑡 are corporation and year fixed effects, respectively. The

coefficient of interest is 𝛽3 , which is the mean difference in assets in 2004-2010 period between C and

S corporations with less than $10 million in assets in 2003. This coefficient would represent the effects

of policies which had differential impacts on smaller (<$10 million in assets) C and S corporations in

the 2004-2010 period, such as Schedule M-3 reporting and differentials in audit rates.

We estimate this model above for two outcome variables: total assets and net income.15 Table

3 presents estimates of the coefficients 𝛽1 , 𝛽2 , and 𝛽3 . Focusing on the coefficient on interest, 𝛽3 , we

find that small C corporations assets are about $980,000 lower in the 2004-2010 period than similarly

sized S corporations. Thus, consistent with the bunching estimators in our main analysis, we observe

a significant effect on the size (measured by assets) of C corporations because of policies affecting

those corporations above $10 million in assets. We do not find any systematic differences in net

income between C and S corporations. The latter result suggests that there are no differences in

reported income among smaller C and S corporations.

15 We considered other outcomes, such as audit rates and proposed adjustments to income upon audit, but there are too

few S corporations audited in our balanced panel to be able to identify the parameters in that model. In addition, we

cannot consider effects on taxable income or tax liability because S corporations are pass-through businesses and thus do

not pay income taxes at the entity level.

25

Table 3: DDD Estimates

Post x (Assets <$10m)

Post x C-Corporation

Post x (Assets <$10m) x Ccorporation

Entity FE

Year FE

Total

Assets

0.356***

(0.082)

1.872***

(0.212)

Net

Income

-0.010*

(0.223)

0.007***

(0.052)

-0.979***

(0.245)

0.007

(0.062)

Yes

Yes

Yes

Yes

Obs

155,056

155,056

R-sq

0.259

0.475

Balanced panel of S and C corporations from 2000

through 2010. Post refers to 2004-2010. The indicator

variable (Assets > $10m) refers to total assets of the

corporation in 2003. Standard errors are clustered at the

entity level. *** p < 0.01, ** p < 0.05, * p < 0.10.

Beginning in 2006, S corporations with $10 million in assets were required to complete Schedule M3. To understand if this has a differential impact on S versus C corporations in the post-period, we

estimate an event-study, triple difference model:

𝑌𝑖𝑡 =

𝛽1𝑗 ∗ (𝑌𝑒𝑎𝑟𝑡 = 2004 + 𝑗) ∗ (𝐴𝑠𝑠𝑒𝑡𝑠 < $10𝑚)𝑖

∑

𝑗∈{−4,...−2,0,…6}

+

∑

𝛽2𝑗 ∗ (𝑌𝑒𝑎𝑟𝑡 = 2004 + 𝑗) ∗ 𝐶 − 𝑐𝑜𝑟𝑝𝑜𝑟𝑎𝑡𝑖𝑜𝑛𝑖

𝑗∈{−4,...−2,0,…6}

+

∑

𝛽3𝑗 ∗ (𝑌𝑒𝑎𝑟𝑡

𝑗∈{−4,...−2,0,…6}

= 2004 + 𝑗) ∗ (𝐴𝑠𝑠𝑒𝑡𝑠 < $10𝑚)𝑖 ∗ 𝐶 − 𝑐𝑜𝑟𝑝𝑜𝑟𝑎𝑡𝑖𝑜𝑛𝑖 + 𝛼𝑖 + 𝛿𝑡 + 𝜀𝑖𝑡

In this model, the coefficients of interest are the 𝛽3𝑗 , which show the mean difference in assets

between C and S corporations who had less than $10 million in assets in 2003 for each year 2004+j.

We plot the estimates of these coefficients in Figure 6. The event study shows statistically insignificant

26

differences between small S and C corporations before 2004, supporting the parallel trends

assumptions needed for identification of the DDD model. Beginning in 2004, C corporations report

lower asset values than similarly sized S corporations, with a difference of about $1 million by 2007

and growing to almost $1.5 million by 2010. There is a slight pause in the divergence of C and S

corporation assets in 2006, the first year that Schedule M-3 was required by S corporations with assets

of at least $10 million. With the exception of 2006, the difference between C and S corporations

grows over time, indicating persistent effects of the changes in enforcement and disclosure on larger

C corporations that result in lower growth of smaller C corporations. Appendix A.2 provides

additional analyses of bunching persistence.

Figure 6: DDD Event Study: Differences in assets between small C-corps and

small S-corps

8. Conclusion

In this study, we employ a bunching estimator and administrative data for the universe of

corporate income tax returns to determine whether and how size-based enforcement thresholds affect

27

U.S. C corporations. Our setting allows us to disentangle the effects of three size-based enforcement

policies using a relatively homogeneous population of corporate taxpayers. We find little evidence that

C corporations manipulate assets to avoid the scrutiny of the LMSB/LB&I enforcement regime.

However, we do find economically significant bunching starting tax years 2004 and 2005, the first two

years after Schedule M-3 adoption and a discrete jump in audit rates. In those years, C corporations

near the $10 million threshold reduce their reported assets by about 5 percent to avoid being subject

to policies that apply to business taxpayers with assets of $10 million or more.

When we consider the variation in audit rates from 2004 to 2010, we find that the size of the

excess bunching estimator tends to move with changes in audit rates. The elasticity of reported assets

with respect to the audit rate in 2004-2005 is about 0.37, suggesting that a one percentage point

increase in audit rate around the threshold is associated with C corporations reducing reported assets

by 0.37 percent. This elasticity of assets with respect to the audit rate is similar in magnitude to the

elasticity of tax payments with respect to the audit rate found in Bergolo et al. (2023). And our

bunching estimator, that suggests firms near the threshold with higher audit rates reduce assets by

about $0.5 million, can be compared to Almunia et al. (2018), who find that Spanish corporations

reduce revenues between 0.3 and 0.5 million euro to avoid the large taxpayer unit, which has increased

enforcement capacity. We view this as evidence that C corporations respond to audit rates rather than

the additional Schedule M-3 disclosure or the LMSB/LB&I Division’s enforcement efforts more

generally. Using the effective tax rate as a measure of tax aggressiveness, we find no evidence of

differences in reporting aggressiveness between C corporations just above and below the $10 million

threshold. Finally, we show that the effects on C corporation size from this bunching behavior have

persistent effects on corporate growth. Overall, we view our results as strongly suggestive of C

corporations bunching below the asset-based enforcement threshold to avoid IRS audit scrutiny.

28

References

Adhikari, Bibek, James Alm, and Timothy Harris. 2021. “Small Business Tax Compliance under

Third-party Reporting.” Journal of Public Economics, 203, 104514.

Almunia, Miguel and David Lopez-Rodriguez. 2018. “Under the Radar: The Effects of Monitoring

Firms on Tax Compliance.” American Economic Journal: Economic Policy, 10(1): 1-38.

Ayers, Benjamin, Jeri Seidman, and Erin Towery. 2019. Tax Reporting Behavior Under Audit

Certainty.” Contemporary Accounting Research, 36(1): 326-358.

Bachas, Pierre and Mauricio Soto. 2021. “Corporate Taxation under Weak Enforcement.” American

Economic Journal: Economic Policy, 13(4): 36-71.

Baer, Katherine. 2002. “Improving Large Taxpayers’ Compliance: A Review of Country

Experience.” International Monetary Fund, Washington DC.

Basri, M. Chatib, Mayara Felix, Rema Hanna, and Benjamin A. Olken. 2021. “Tax Administration

versus Tax Rates: Evidence from Corporate Taxation in Indonesia.” American Economic Review,

111(12): 3827-71.

Bergolo, Marcelo, Rodrigo Ceni, Guillermo Cruces, Matias Giaccobasso, and Ricardo Perez-Truglia.

2023. “Tax Audits as Scarecrows: Evidence from a Large-Scale Field Experiment.” American

Economic Journal: Economic Policy, 15(1): 110-153.

Best, Michael Carlos, Anne Brockmeyer, Henrik Jacobsen Kleven, Johannes Spinnewijn, and

Mazhar Waseem. 2015. “Production versus Revenue Efficiency with Limited Tax Capacity: Theory

and Evidence from Pakistan.” Journal of Political Economy, 123(6): 1311-1355.

Blouin, Jennifer L., Cristi A. Gleason, Lillian F. Mills, and Stephanie A. Sikes. 2010. “Pre-empting

Disclosure? Firms’ Decisions Prior to FIN No. 48.” The Accounting Review, 85(3): 791-815.

Chetty, Raj, John N. Friedman, Tore Olsen, and Luigi Pistaferri. 2011. “Adjustment Costs, Firm

Responses, and Micro vs. Macro Labor Supply Elasticities: Evidence from Danish Tax Records.”

Quarterly Journal of Economics, 126(2): 749–804.

Donohoe, Michael P. and Gary A. McGill. 2011. “The Effects of Increased Book-Tax Difference

Tax Return Disclosures on Firm Valuation and Behavior.” Journal of the American Taxation

Association, 33(2): 35-65.

Frischmann, Peter J., Terry Shevlin, and Ryan Wilson. 2008. “Economic Consequences of

Increasing the Conformity in Accounting for Uncertain Tax Benefits.” Journal of Accounting and

Economics, 46(2): 261-278.

Gardner, Matthew, Steve Wamhoff, Mary Martellotta, and Lorena Roque. 2019. “Corporate Tax

Avoidance Remains Rampant Under New Tax Law.” Institute on Taxation and Economic Policy

Report.

Guyton, John, Patrick Langetieg, Daniel Reck, Max Risch, and Gabriel Zucman. 2021. “Tax Evasion

at the Top of the Income Distribution: Theory and Evidence.” National Bureau of Economic

Research Working Paper 28542.

Hargaden, Enda Patrick. 2020. “Tax payer responses in good times and bad.” Journal of Economic

Behavior and Organization, 176(August): 653-690.

29

Henry, Erin, Norman Massel, and Erin Towery. 2016. “Increased Tax Disclosures and Corporate

Tax Avoidance.” National Tax Journal, 69(4): 809-830.

Hoopes, Jeffrey L., Devan Mescall, and Jeffrey A. Pittman. 2012. “Do IRS Audits Deter Corporate

Tax Avoidance?,” The Accounting Review, 87(5): 1603-1639.

Hope, Ole-Kristian, Mark Ma, and Wayne Thomas. 2013. “Tax Avoidance and Geographic

Earnings Disclosure.” Journal of Accounting and Economics, 56(2-3): 170-189.

Internal Revenue Service. 2000. “New IRS Begins Work; Reorganized Agency Reaches Milestone to

Improve Service.” IR-2000-67. Washington, D.C: Department of Treasury.

Internal Revenue Service 2023. “Federal Tax Compliance Research: Tax Gap Projections for Tax

Years 2020 and 2021.” IRS Publication 5869, Department of Treasury, Washington, D.C.

Internal Revenue Service. 2018. “Internal Revenue Service Data Book, 1988-2018.” Washington,

D.C: Department of Treasury.

Jacobs, Deborah L. 2012. “Most Foreign Banks Don’t Want U.S. Clients; How To Find One That

Does.” Forbes, July 31.

Jolly, David and Brian Knowlton. 2011. “Law to Find Tax Evaders Denounced.” New York Times,

December 26.

Journal of Accountancy. 2010. “IRS Changes Name of LMSB Division, Expands Focus.”

Association of International Certified Professional Accountants. August 4, 2010.

Kleven, Henrik J. and Mazhar Waseem. 2013. “Using Notches to Uncover Optimization Frictions

and Structural Elasticities: Theory and Evidence from Pakistan.” Quarterly Journal of Economics, (2):

669–723.

Laughlin, Felix B. and Mark D. Allison. 2001. “The IRS Reorganization: Programs and Initiatives of

the New Large Case Division.” Administrative Law Review, 53: 679.

Mills, Lillian F. and George A. Plesko. 2003. “Bridging the Reporting Gap: A Proposal for More

Informative Reconciling of Book and Tax Income.” National Tax Journal, 56(4): 865-893.

Mills, Lillian F., Leslie Robinson, and Richard C. Sansing. 2010. “FIN 48 and Tax Compliance.” The

Accounting Review, 85(5): 1721-1742.

Baum, Anja, Sanjeev Gupta, Elijah Kimani, and Sampawende Jules Tapsoba. 2017. “Corruption,

Taxes and Compliance,” eJournal of Tax Research, 15(2): 190-216.

Onji, Kazuki. 2009. “The Response of Firms to Eligibility Thresholds: Evidence from the Japanese

Value-Added Tax.” Journal of Public Economics, 93(5-6). 766-775.

Oronde, Small and Leanora Brown. 2020. “Taxpayer Service Provision and Tax Compliance:

Evidence for Taxpayers in Jamaica.” Public Finance Review, 48(2): 250-277.

Padmakumar, Kalyani. 2023, “Small by Choice: Reassessing the Aggregate Implications of SizedBased Regulations.”, Penn State University Working Paper.

Plesko, George A. 2003. “An Evaluation of Alternative Measures of Corporate Tax Rates.” Journal

of Accounting and Economics, 35(2): 201-226.

Saez, Emmanuel. 2010. “Do Taxpayers Bunch at Kink Points?” American Economic Journal: Economic

Policy, 2(3): 180-212.

30

Sandmo, Agnar. 2012. “An Evasive Topic: Theorizing about the Hidden Economy.” International

Tax and Public Finance, 19: 5-24.

Slemrod, Joel and Tejaswi Velayudhan. 2018. “Do Firms Remit at least 85% of Tax Everywhere?

New Evidence from India.” Journal of Tax Administration, 4(1): 24-37.

Tennant, Sandria N. and Marlon R. Tracey. 2019. “Corporate Profitability and Effective Tax Rate:

The Enforcement Effect of Large Taxpayer Units.” Accounting and Business Research, 49(3): 342-361.

Towery, Erin. 2017. “Unintended Consequences of Linking Tax Return Disclosures to Financial

Reporting for Income Taxes: Evidence from Schedule UTP.” The Accounting Review, 92(5): 201-226.

Yagan, Danny. 2015. “Capital Tax Reform and the Real Economy: The Effects of the 2003

Dividend Tax Cut.” American Economic Review, 105 (12): 3531-63.

Appendix

A.1 Theory

A firm makes production decisions and can choose to under-report its taxable income and

assets to a tax authority, though it faces resource costs for doing so. The probability of detecting

misreported income and misreported assets is a function of the firm’s paper trail generated by legal

and tax reporting requirements, public financial statement disclosures, and third-party information

reporting to the tax authority (“enforcement technology”) and the tax authority’s efforts to examine

the firm’s paper trail (“monitoring effort”).

In the baseline setting, we assume there is no

discontinuous increase (i.e., a notch) in the enforcement technology or the tax authority’s monitoring

effort.

We then extend our baseline framework to model how a firm responds to an arbitrary

discontinuous increase in tax monitoring and an arbitrary discontinuous increase in enforcement

technology. To do so, we adapt the Almunia and Lopez-Rodriguez (2018) model of a firm’s response

to a discontinuous increase in monitoring effort at an arbitrary revenue threshold. Our setting is

unique in that firms face discontinuous increases in both enforcement technology and monitoring

effort, and an arbitrary asset threshold rather than an arbitrary revenue threshold triggers these

increases. The under-reporting of assets, unlike the under-reporting of revenues, does not directly

31

affect a firm’s taxable income or a firm’s tax liability, but it can indirectly affect the tax liability for a

subset of firms around the asset threshold by discontinuously increasing the enforcement technology

and the tax authority’s monitoring efforts.16 The use of an asset threshold rather than a revenue

threshold generates a set of theoretical predictions, some of which are similar to Almunia and LopezRodriguez’s (2018) predictions and others are substantially different. We discuss these predictions in

detail in Sections 3.2 through 3.4.

A.1.1. A Model of Firm Tax Compliance

Consider a firm with a net income of y = ψf(a) − r ∗ a, where a represents assets and r is

the interest rate on assets, which is assumed to reflect the marginal cost of holding assets. The

function ψf(a) represents the production function, where ψ is an exogenously determined parameter

of firm productivity and f(⋅) is strictly continuous, increasing and concave in assets. The firm must

pay taxes on its reported net income. The firm’s expected after-tax profit is given by:

Eπ = (1 − τ)(ψf(a) − r ∗ a) − ca (ua ) − cy (uy) + τuy [1 − ϕh(ua , uy )(1 + θ)] − 𝜙𝛿,

(1)

where τ is the tax rate on net income, which we assume is linear. The cost of under-reporting assets

is represented by the function ca (⋅), which is convex and increasing in under-reported assets (ua).

The cost of under-reporting income is represented by the function cy (⋅), which is convex and

increasing in under-reported income (uy). The parameter ϕ represents tax monitoring intensity, h(⋅,⋅)

represents the enforcement technology, which is assumed to be increasing and convex in both underreported assets and under-reported income, and θ represents penalties paid on under-reported taxes.

16 Almost all prior studies that examine taxpayer bunching responses analyze reactions to revenue or income thresholds.

Because under-reporting of revenue or income directly affects taxable income and thus tax liability, these studies must rely

on the changes in these thresholds to disentangle the bunching response motivated by a taxpayer’s desire to avoid extra

monitoring from the taxpayer’s attempt to evade taxes by under-reporting true revenue or income. The use of an asset

threshold provides a unique opportunity to estimate the bunching response to monitoring effort because under-reporting

of assets itself does not directly affect taxable income.

32

The final term represents the expected direct costs of audit, which are given by the financial, time, and

hassle costs of complying with the audit, 𝛿, times the monitoring intensity, 𝜙.

Under-reported assets do not directly affect penalties levied by the tax authority because assets

do not affect taxable income and penalties are a linear function of the firm’s tax deficiency. However,

firms incur costs to under-report assets. For example, to the extent lenders utilize tax return

information when deciding whether to extend a loan or set interest rates, under-reporting assets could

make the firm appear less creditworthy and thus affect the firm’s ability to borrow and/or their cost

of borrowing (r).17 The probability of detection is also increasing in under-reported assets because

larger deviations from true assets could raise red flags to the IRS.

We assume that under-reported assets and under-reported income must be non-negative.

With the non-negativity constraints, the Lagrangian function for the firm’s expected profit

maximization problem is given as:

ℒ = (1 − τ)(ψf(a) − r ∗ a) − ca (ua ) − cy (uy ) + τuy [1 − ϕh(ua , uy )(1 + θ)] − 𝜙𝛿 + λa ua + λy uy (2)

The first order conditions are:

∂ℒ

∂f(a)

= ψ ∂a − r = 0

(3)

∂h(ua, uy )

∂ℒ

∂ca (ua )

(1 + θ) + λa = 0

=−

− τuy ϕ

∂ua

∂ua

∂ua

(4)

∂cy (uy)

∂h(ua , uy )

∂ℒ

(1 + θ) + λy = 0

=−

+ τ[1 − ϕh(ua , uy )(1 + θ)] − τuy ϕ

∂uy

∂uy

∂u𝑦

(5)

∂a

17 For simplicity, we assume the cost of borrowing does not vary with the size of reported assets.

33

The first condition can be rewritten as ψf ′ (a) = r, which implies that the firm chooses its

level of assets based on its productivity and the marginal cost of assets, but not tax rates. The second

condition can be rewritten as:

λ𝑎 =

∂ℎ(𝑢𝑎 , 𝑢𝑦 )

∂ca (𝑢𝑎 )

(1 + θ)

+ τ𝑢𝑦 ϕ

∂𝑢𝑎

∂𝑢𝑎

(6)

Both right-hand side terms in Equation 6 are unambiguously positive, which means λa > 0.

Thus, the non-negativity constraint on ua is binding and the firm does not under-report its assets.

∂λ

Comparative statics reveal two insights. First, we observe that ∂𝑢𝑎 ≥ 0, meaning that the firm’s

𝑦

incentive to accurately report assets is increasing in the magnitude of its under-reported income. The

intuition is that the firm wants to minimize the probability of under-reported income being detected,

and accurately reporting its assets to the tax authority lowers the probability of detection via the

∂λ

∂λ

enforcement technology term. Second, we observe that ∂ϕa ≥ 0 and ∂θa ≥ 0, meaning that the firm’s

incentive to accurately report assets is increasing in monitoring effort and penalty rates. In other

words, higher enforcement rates reduce the firm’s incentive to under-report assets because the tax

authority is more likely to detect income misreporting.

The third condition can be rewritten as:

τ[1 − ϕℎ(𝑢𝑎 , 𝑢𝑦 )(1 + θ)] + λ𝑦 =

∂𝑐𝑦 (𝑢𝑦 )

∂ℎ(𝑢𝑎 , 𝑢𝑦 )

(1 + θ)

+ τ𝑢𝑦 ϕ

∂𝑢𝑦

∂𝑢𝑎

(7)

The first term on the left-hand side of Equation 7 represents the benefits to under-reporting

income: lower expected taxes paid. The right-hand side terms in Equation 7 represent the resource

costs of under-reporting income, cy (⋅), and the additional expected taxes and penalties from underreporting income. The first term on the right-hand side is positive and the second term is non-

34

negative. Thus, the firm may find itself at a corner solution, accurately reporting income (uy = 0).

However, an interior solution with positive amounts of under-reported income is also possible.18

In considering Equations 6 and 7, it is worth noting that ua and uy are independent of

productivity and the marginal cost of assets (i.e., neither ψ nor r are in Equations 6 or 7). Per Equation

7, income under-reporting will fall when monitoring effort, ϕ, rises or when penalties, θ, increase. As

Equation 3 illustrates, if productivity, ψ, is continuously distributed across firms and all firms have

the same cost functions and face the same tax/penalty rates, monitoring effort, and enforcement

technologies, then we will observe a continuous distribution of firms’ assets. Because reported assets

equal true assets (as shown in Equation 6), there is a continuous distribution of reported assets that

matches the true asset distribution. Reported income will also be a smooth distribution; Equation 7

shows that reported income will be a level shift down from true income because true income (𝑦),

productivity (ψ), and true assets (a) do not enter equation 7.19

A.1.2. Bunching in the Presence of a Monitoring Notch

Now assume that there is a discontinuous jump in monitoring effort, ϕ, at asset threshold a̅

+ ϵ, where ϵ is some small number. In this case, the firm faces a discontinuous choice of reported

assets around this threshold. As a result, there is no first order condition, and we have to instead look

at the decision to report assets below or above the threshold and determine which reporting decision

yields higher expected profits for the firm. The firm will choose to report assets below the threshold

if:

Eπ(a, ua , uy|ϕ0 , ψ) > Eπ(a′, ua ′, uy ′|ϕ1 , ψ),

(8)

18 Whether or not there is a corner solution depends on the slopes of the cost function and the enforcement function. If

either function is steeply increasing even near zero, then a corner solution would obtain.

19 One exception to this would be the special case of a corner solution noted earlier (i.e., (u = 0)). In this case, the

y

distribution of reported income will equal the distribution of true income.

35

where ϕ0 and ϕ1 represent the lower and higher enforcement regimes below and above the threshold,

respectively. Because firms will not over-report assets, all firms with a ψ below a certain level will

have true assets a ≤ a̅. We refer to this level of ψ as ψL . No firm with ψ < ψL will misreport assets

because doing so imposes a cost but does not decrease monitoring effort as they are already choosing

true assets below the threshold.

Now consider a firm that is indifferent between reporting assets below or above the threshold.

Let the level of productivity that defines this threshold condition be ψM, where ψM satisfies:

(9)

Eπ(a, ua , uy|ϕ0 , ψM) = Eπ(a′, ua ′, uy ′|ϕ1 , ψM )

A firm with productivity between ψL and ψM will want to report assets a̅ to avoid the high

enforcement regime. This firm will choose true assets a > a̅, but benefit from under-reporting assets.

By under-reporting assets, they can avoid higher monitoring by the tax authority.

Note that the firm has no incentive to report assets any less than a̅ because doing so does not

further decrease the monitoring effort it faces, but it increases misreporting costs, ca (⋅) and the

likelihood the tax authority detects its noncompliance (through h(⋅,⋅)). For a high-productivity firm

with ψ > ψM, the amount of under-reporting that would be required to report at or below a̅ is not

worth the resulting increases in h(⋅,⋅) and ca (⋅) that would be incurred. Thus, in the presence of an

enforcement notch that is a function of assets, we expect to see a mass of firms at a̅ with no firms just

above that. This mass represents the firms with productivity ψL < ψ < ψM who report assets a̅. All

other firms will report their true assets. Figure 2 illustrates the theoretical distribution of reported

assets in the presence of an enforcement notch. The distribution of reported assets without a notch

is also shown for comparison.

36

Figure A.1.1: Theoretical Distribution of Reported Assets in the Presence of a Notch

(Homogenous Firms)

Notes: This figure depicts the theoretical distribution of reported assets with and without the asset-based enforcement

threshold for the baseline model, where monitoring effort and resource costs of evasion are the same for all firms.

Interestingly, the degree of under-reported income among the bunching firms is theoretically

ambiguous. On the one hand, ϕ is lower, which would increase misreporting. On the other hand,

the misreporting of assets increases h(⋅,⋅) and

∂h(ua ,uy )

∂ua

(because of its convexity), both of which

reduce income misreporting. In our empirical work, we will test to see which effect dominates.

A.1.3. Heterogeneity

More realistically, we expect that there are differences across firms in how well the

enforcement technology works and with respect to the costs of under-reporting income or assets. For

instance, firms with foreign operations may find it easier to under-report income. In contrast, public

firms that must release financial statement information could be more effectively monitored by the

IRS. Modifying the model above so that firms are heterogeneous in cy (⋅) and ca (⋅), and h(⋅,⋅) has

37

implications for the distribution of firms across reported assets. Specifically, some firms in the

productivity range ψL < ψ < ψM may no longer report assets below the threshold because the cost

of doing so is too high. As a result, there will not be a zero mass of firms in the distribution of

reported assets just above a̅. Rather, there will be a missing mass, but not a zero support, in the region

to the right of a̅. Figure 3 illustrates the theoretical distribution of reported assets under the

assumption of heterogeneity in the cost of underreporting and enforcement technology functions. As

before, we show the counterfactual distribution (i.e., the distribution without the notch) for

comparison.

Figure A.1.2: Theoretical Distribution of Reported Assets in the Presence of a Notch

(Heterogeneous Firms)

Notes: This figure depicts the theoretical asset distribution with and without the asset-based enforcement threshold for

the model with heterogeneity, where a fraction of firms does not respond to the incentives due to different changes in

monitoring effort at the threshold or due to other adjustment costs.

A.1.4. Notches in Enforcement Technology

In our institutional setting, we observe a notch not only in monitoring effort, but also in

enforcement technology because large firms are subject to more detailed reporting requirements on

their tax returns than smaller firms. A notch in the enforcement technology at asset value a̅ will have

38

the same effects we outlined above for the notch in monitoring effort. Such a notch will lead to a

mass of firms at and to the left of a̅ and a missing mass of firms to the right of a̅. In our empirical

analyses below, we will look at changes in the mass of firms around the asset thresholds related to

notches in both monitoring effort and the enforcement technology.

A.2 Additional estimates of persistent effects

A.2.1 Non-parametric evidence of persistence

We illustrate how the post-2004 notch in enforcement has affected the persistence of C

corporations remaining below $10 million in asset by considering the rates of transition between asset

bin around the $10m threshold. Figure 15 splits our sample into pre-2004 and post-2004 periods and

reports the fraction of C corporations that report assets in a given range at time t who then remain in

that same asset range 1, 2, 3, and 4 years later. The black bars represent C corporations in a $150,000

range just below $10 million, and the gray bars represent C corporations in a $150,000 at or over $10

million. Before the notch in audit rates at $10 million (i.e., the pre-2004 period), there is little difference

in the persistence of remaining in the asset bin just below and just above the threshold. In contrast,

after 2004, when there is a significant notch in audit rates, C corporations who report assets just below

the $10 million threshold are much more likely to remain the same asset bin in the following years

than are C corporations who report just above the threshold. In the post-2004 period, one year after

first reporting assets in their respective ranges, those just under $10 million are about 60 percent more

(a rate of 0.058 versus a rate of 0.038) likely to remain in the same asset bin than those just over the

threshold. These are small bins with most corporations transitioning out of the bins in a given year,

so these results should be interpreted with caution. But the pattern is suggestive of the persistence in

staying below the threshold increasing after 2004.

39

Figure A.2.1: Persistence in Corporation Size Around the $10 Million Threshold

Pre-2004

Post-2004

Notes: This figure shows the fraction of C corporations that remain in the same $150,000 bin k years after they first show

up in that asset size bin. Black bars represent this fraction for C corporations in the [$9,850,000, $10,000,000) bin up to 5

years after they are first in that bin, and the gray bars denote the fraction of C corporations remaining in the [$10,000,000,

$10,150,000) bin up to 5 years after they are first in that bin.

A.2.2 Parametric evidence of persistence

Second, we follow Padmakumar (2023) and use a difference-in-differences research design to

test whether there is an effect of the enforcement notch on asset accumulation by C corporations.

Specifically, we estimate the regression model:

𝑌𝑖,𝑗,𝑡 = ∑𝑘>1 𝛽𝑘𝑝𝑟𝑒 1[𝐵𝑖𝑛𝑖,𝑗,𝑡−1 = 𝑘] + 𝛿𝑃𝑜𝑠𝑡𝑡 + ∑𝑘>1 𝛽𝑘𝑝𝑜𝑠𝑡 𝑃𝑜𝑠𝑡𝑡 ∗ 1[𝐵𝑖𝑛𝑖,𝑗,𝑡−1 = 𝑘] +

𝜇𝑡 + 𝛾𝑗 + 𝜀𝑖,𝑗,𝑡 ,

Where I represents the corporation, j industry, and t year. The dependent variable, 𝑌𝑖,𝑗,𝑡 , is an indicator

variable for assets being reported in a higher ‘bin’ in the following year. Specifically, the dependent

variable is defined as:

1 𝑖𝑓 𝐵𝑖𝑛𝑖,𝑗,𝑡 > 𝐵𝑖𝑛𝑖,𝑗,𝑡−1

𝑌𝑖,𝑗,𝑡 = {

0 𝑖𝑓 𝐵𝑖𝑛𝑖,𝑗,𝑡 ≤ 𝐵𝑖𝑛𝑖,𝑗,𝑡−1

40

These asset bins are defined in $1 million dollar ranges. The indicator variable for moving to a bin

with higher asset values, 𝑌𝑖,𝑗,𝑡 , is regressed on a vector of indicator variables for the million-dollar bin

ranges in the prior period, an indicator variable for the post-2004 period, interactions of the post

indicator variable and vector of indicator variables for the asset bins, as well as time and industry fixed

effects. Parameter estimates from this regression are presented in Table 2. Column 1 of this table uses

our full sample of C corporations, while Column 2 uses a sample of C corporations with assets

between $4 million and $16 million.

Table A.2.1: Probability of Moving up the Asset Distribution, Differences-in-Differences

Estimates

Assets [$5m, $6m)

Assets [$6m, $7m)

Assets [$7m, $8m)

Assets [$8m, $9m)

Assets [$9m, $10m)

Assets [$10m, $11m)

Assets [$11m, $12m)

Assets [$12m, $13m)

Assets [$13m, $14m)

Assets [$14m, $15m)

Assets [$15m, $16m)

Post-2003

Corporations with

Assets [$2m, $20m]

Corporations with

Assets [$4m, $16m]

0.335

(0.001)

0.368

(0.001)

0.385

(0.001)

0.402

(0.001)

0.417

(0.001)

0.433

(0.002)

0.439

(0.002)

0.457

(0.002)

0.471

(0.002)

0.467

(0.002)

-0.009

(0.000)

-0.008

(0.001)

0.035

(0.004)

0.068

(0.005)

0.085

(0.005)

0.102

(0.006)

0.117

(0.006)

0.133

(0.007)

0.140

(0.007)

0.160

(0.008)

0.174

(0.008)

0.171

(0.009)

-0.306

(0.009)

-0.003

(0.005)

41

Assets [$5m, $6m)*Post

-0.011

(0.001)

Assets [$6m, $7m)*Post -0.010

(0.001)

Assets [$7m, $8m)*Post 0.001

(0.001)

Assets [$8m, $9m)*Post -0.014

(0.002)

Assets [$9m, $10m)*Post -0.001

(0.002)

Assets [$10m,

$11m)*Post

0.006

(0.002)

Assets [$11m,

$12m)*Post

-0.001

(0.002)

Assets [$12m,

$13m)*Post

0.002

(0.002)

Assets [$13m,

$14m)*Post

0.008

(0.002)

Assets [$14m,

$15m)*Post

0.000

(0.000)

-0.004

(0.005)

-0.004

(0.006)

0.005

(0.006)

-0.009

(0.007)

0.004

(0.008)

Year Effects

Industry FEs

Yes

Yes

0.009

(0.008)

0.002

(0.009)

0.005

(0.009)

0.010

(0.010)

0.003

(0.010)

Yes

Yes

Observations

R-Squared

13,359,512

0.271

554,527

0.391

Notes: This table presents the results of a panel regression testing for an effect of the enforcement notch on C corporation

asset accumulation.

Looking first at the coefficient estimates on the asset bin indicators, we see that they are all

positive. This indicates that C corporations who survive to file a tax return in the future, tend to grow

over time (e.g., on average increasing their asset size). Next, we consider the coefficients on the asset

bin indicator variables interacted with the post-2004 indicator. These coefficients are often statistically

indistinguishable from zero. However, one exception is the indictor variable on the bin just below

42

$10 million. The interaction of the indicator variable for being in this asset range one period ago and

the post indicator variable is significantly negative. That is, after the change in audit regime in 2004,

the probability that a C corporation just below $10 million moves to a higher asset bin declines. The

coefficient estimate implies that the probability of moving upward declines from 0.45 to 0.44, a decline

of about 2 percent in the rate at which these corporations grow.

A.3 Bunching Estimates Across Years

Table A.3.1 C Corporation Bunching Around $10 Million in Assets by Year

Year

2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

b

Std. Err.

Audit Rate

Differential (pp)

0.122

0.151

0.156

0.140

0.151

0.136

0.139

0.130

0.155

0.156

0.155

3.57

5.96

5.35

9.69

17.00

10.59

6.89

13.13

14.39

3.96

5.06

-0.007

-0.029

-0.099

-0.090

0.461

0.559

0.263

0.348

0.401

0.168

0.068

43

Table A.3.2 C Corporation Bunching Around $10 Million in Assets by Year Ranges

Years

20002001

20022003

20042005

20062007

20082009

20102011

b

Std.

Err.

Audit Rate

Differential (pp)

-0.018

0.103

4.68

-0.094

0.113

7.53

0.511

0.115

13.68

0.303

0.097

9.97

0.278

0.110

9.22

0.277

0.128

4.38

Table A.3.3 S Corporation Bunching Around $10 Million in Assets by Year Ranges

Years

2000-2001

2002-2003

2004-2005

2006-2007

2008-2009

2010-2011

b

0.038

-0.100

0.106

0.103

0.276

0.043

Std. Err.

Audit Rate

Differential (pp)

0.125

0.124

0.096

0.079

0.111

0.112

1.13

-0.60

0.91

-0.18

1.37

0.14

A.4 Robustness to alternative bunching parameters

Table A.4.1 shows the bunching estimate for C corporations in 2004-2005 under alternative

parameterizations of and the polynomial used for the counterfactual density.

44

Table A.4.1 C Corporation Bunching Around $10 Million in Assets in 2004-2005 Under

Alternative Parameterizations

b

0.346

0.346

0.337

0.337

0.359

0.523

0.524

0.511

0.511

0.548

0.523

0.524

0.511

0.511

0.548

Lower Bound

of Excluded

Area

Std. Err.

0.091

0.088

0.091

0.091

0.090

0.112

0.108

0.115

0.115

0.116

0.112

0.108

0.115

0.115

0.116

$

$

$

$

$

$

$

$

$

$

$

$

$

$

$

45

9,850,000

9,850,000

9,850,000

9,850,000

9,850,000

9,700,000

9,700,000

9,700,000

9,700,000

9,700,000

9,550,000

9,550,000

9,550,000

9,550,000

9,550,000

Polynomial

Order

2

3

4

5

6

2

3

4

5

6

2

3

4

5

6