Tax Evasion at the Top of the Income Distribution:

Agency decision

Ask Donna

What actually matters in this document.

Text

Theory and Evidence ∗

John Guyton

Patrick Langetieg

Daniel Reck

Internal Revenue Service

University of Maryland

Max Risch

Gabriel Zucman

Carnegie Mellon University

Paris School of Economics and UC Berkeley

December 22, 2025

Abstract

We study tax evasion at the top of the U.S. income distribution using micro-data from audits and enforcement initiatives. Leveraging enforcement that revealed noncompliance, we

fnd that recent individual random audit data fail to capture sophisticated evasion via offshore

accounts and pass-through businesses. Consequently, estimates based solely on these data

under-state evasion by the highest-income Americans. We propose a theoretical explanation

and construct new distributional estimates of US noncompliance. Circa 2006–2013, accounting

for sophisticated evasion increases the share of unreported income belonging to the top 1% of

the distribution by between 9 and 33 percentage points.

∗

Corresponding Author: Daniel Reck, dreck@umd.edu. We thank Gerald Auten, Pierre Bachas, Karen Cincotta, Brian

Galle, Bhanu Gupta, Anne Herlache, Tom Hertz, Xavier Jaravel, Drew Johns, Barry Johnson, Camille Landais, Katie Lim, Emily

Lin, Larry May, Alicia Miller, Erik Ogilvie, Annette Portz, Ahmad Qadri, Peter Rose, Emmanuel Saez, Brenda Schafer, Clifford Scherwinski, Joel Slemrod, Matt Smith, Johannes Spinnewijn, David Splinter, Alex Turk, and Alex Yuskavage for helpful

discussion, support, and comments on preliminary versions of this work. Francesco Armillei, Jeanne Bomare, Idann Gidron, and

Baptiste Roux provided excellent research assistance. All remaining errors are our own. Financial support from the Washington

Center for Equitable Growth, the Stone foundation, Arnold Ventures, and the Economic and Social Research Council is gratefully

acknowledged.

All fndings, opinions, and errors are those of the authors alone and do not necessarily represent the opinions of the Internal

Revenue Service or Treasury Department.

Introduction

How much do high-income individuals evade in taxes? What are the main forms of tax noncompliance of the top of the income distribution? Because taxable income and tax liabilities are highly

concentrated at the top of the income distribution, understanding noncompliance by high-income

taxpayers is critical for the analysis of tax evasion, for tax enforcement, and for the conduct of tax

policy.

A key diffculty in studying tax evasion by high-income individuals is the complexity of the

forms of tax evasion available to them. Non-compliance by high earners often involves legal and

fnancial intermediaries, such as shell companies and offshore banks, sometimes in countries with

a great deal of secrecy. Moreover, a large fraction of the income of top earners derives from closelyheld businesses. Misreporting in closely held businesses can be signifcant and yet diffcult to

detect due to resource constraints within the tax authority, to the complexity of business structures

with intertwined entities, and to the lack of third-party reporting. This complexity means that

one single data source is unlikely to uncover all forms of noncompliance at the top of the income

distribution.

In this paper, we attempt to quantify and overcome this limitation in the U.S. context by combining a wide array of sources of micro data, including (i) individual random audit data, (ii) random audits of S-corporations, an important form of private businesses in the U.S., (iii) the universe

of operational audits conducted by the Internal Revenue Service (IRS), and (iv) focused enforcement activities (e.g., on offshore bank accounts and on specifc evasive schemes used by private

business owners). Drawing on this unique combination of data, we show that estimates based

solely on individual random audits from 2006–2013 underestimate tax evasion at the top-end of

the income distribution. Specifcally, we fnd that under the audit procedures used during this

period, individual random audit data do not capture most sophisticated evasion occurring via offshore intermediaries and pass-through businesses. We provide a theoretical explanation for these

facts, and we propose a methodology to improve the estimation of the size and distribution of tax

noncompliance in the United States.

The starting point of our analysis is the IRS individual random audit program, known as the

National Research Program (NRP). Stratifed random audits are commonly used to study and

measure the extent of tax evasion.1 Researchers use random audit data to test theories of tax

evasion (Kleven et al., 2011), and tax authorities use them to estimate the extent of tax evasion – the

tax gap – and select audits (IRS, 2019). Analyzing what individual audits detect in a representative

We use the term tax evasion in this paper to refer to both unintentional and intentional noncompliance with tax

obligations, and do not attempt to distinguish between them. We note that this defnition of evasion may include

behaviors taxpayers would argue are legal, and where the line between legal tax avoidance and evasion is unclear, our

defnition requires taking a stance. When using audit data, we defer to auditor recommendations to delineate what

behaviors are evasion, which is conventional in the literature. Gray areas nevertheless introduce some uncertainty into

our analysis, which we discuss throughout the paper.

sample of the full population yields important insights for policymakers and researchers, but when

it comes to tax gap estimation especially, the main diffculty is that general purpose audits may

not detect all forms of tax evasion. The NRP individual random audits from 2006–2013 appear

to be well-designed to detect simple and common forms of tax evasion, such as unreported selfemployment income, overstated deductions, and the over-claiming of tax credits. But, we argue,

these audits did not detect more sophisticated types of evasion, at least during the period we study,

because doing so would require additional information, resources, and/or specialized staff.

Our main contribution is to demonstrate the limits of NRP random audit data spanning tax

years 2006 to 2013 when it comes to estimating tax evasion at the top of the income distribution.

We fnd that in NRP individual random audit data from this period, detected evasion declines

sharply at the top of the income distribution, with less than 1% of income found by auditors to

be under-reported in the top 0.01%. Our analysis uncovers two key limitations of these audits

which can account for this drop-off: tax evasion through foreign intermediaries (e.g., undeclared

foreign bank accounts) and tax evasion via private businesses (e.g., under-reported revenues or

overstated expenses in a partnership). We stress that these limitations apply to NRP individual

random audits during our sample period and the audit procedures they entailed; they are not

necessarily fundamental limitations of random audit data.

We fnd that offshore tax evasion went almost entirely undetected in the random audit data.2

To establish this result, we analyze the sample of U.S. taxpayers who disclosed hidden offshore

assets in the context of specifc enforcement initiatives conducted in 2009–2012. The very top of

the income distribution is dramatically over-represented in this sample. We fnd that offshore

evasion was rarely detected in the audits, even when it was in fact occurring.

We also fnd that tax evasion occurring in pass-through businesses (S corporations and partnerships) is highly under-detected in the individual random audit data. Pass-throughs are private businesses whose income is taxed as income to their owners; pass-through ownership and

income are highly concentrated (Cooper et al., 2016). During our sample period, examiners auditing pass-through owners during an individual random audit usually did not examine the degree

to which pass-through businesses duly reported their income, especially for complex businesses.

Thus, while the taxable income of taxpayers in the bottom 99% of the income distribution is comprehensively examined, up to half of the taxable income earned at the very top is not comprehensively examined under the current individual random audit procedures. Analyzing microdata from a small-scale random audit program for S corporations, we show that evasion in these

businesses is substantial: about 20% of net business income is under-reported, fve times more

than the total pass-through business income under-reporting detected in individual random audits. We also show that the adoption of sophisticated tax evasion schemes involving pass-through

Our data cover the period prior to the collection of third-party reported information on foreign bank accounts,

which started in 2014.

entities—micro-captive insurance and syndicated conservation easements—is prevalent among

high-income taxpayers.

We investigate the consequences of undetected offshore and pass-through under-reporting for

the measurement of the distribution of unreported income and overall income inequality. The

precise quantitative importance of undetected offshore and pass-through evasion is uncertain because we do not have enough data to defnitively correct estimates based on individual random

audit data. We propose and discuss the empirical rationale for three sets of assumptions, under

which we impute the effect of accounting for offshore and pass-through evasion. These three sets

of assumption represent our chosen benchmark scenario, a more conservative low-end set of assumptions, and a stronger high-end set of assumptions.

Across all three imputations, we fnd only slightly more evasion on aggregate than in the standard IRS methodology, but we fnd that accounting for offshore and pass-through evasion is quantitatively important at very the top of the income distribution. Our adjustments increase estimated

unreported income by a factor of 1.11 to 1.29 in the aggregate.3 Individual NRP audit data suggest

that 11.4% of under-reporting belongs to the top 1% of the true income distribution. Including imputed pass-through and offshore under-reporting increases this share by 9 to 33 percentage points.

The quantitative importance of undetected offshore and pass-through evasion is also robust to the

inclusion of other undetected evasion identifed using the Detection Controlled Estimation (DCE)

methods the IRS uses in its Tax Gap Studies; nevertheless, we show that accounting for undetected evasion identifed by these methods introduces a large amount of uncertainty about the

distribution of unreported income. Our imputation results suggest that pass-through and offshore

under-reporting are also important for the overall extent of income inequality (e.g. the top 1%

income share including under-reported income). Quantiative uncertainty about the exact extent

and concentration of offshore and pass-through under-reporting, as well as distributional uncertainty around Detection Controlled Estimation, also generate uncertainty about the top 1% share

of income including under-reported income.

We suggest a theoretical explanation for why general-purpose audits are not uniformly able to

detect noncompliance across the income distribution. If we extend the canonical model of Allingham and Sandmo (1972) to allow taxpayers to pay a fxed cost to reduce their probability of detection, high-income taxpayers will fnd it optimal to pay this fxed cost if their desired rate of evasion

is nonzero. Audit rates are signifcantly higher at the top than at the bottom of the distribution,

making evasion that is less likely to be detected and corrected on audit more attractive at the top.

Related Literature These fndings provide a new picture of the distribution and composition of

tax non-compliance, with implications for tax enforcement and for the effective progressivity of

the US tax system. To optimize tax enforcement, it is important to know where in the income

These ratios compare total under-reported income with and without including imputed offshore and pass-through

under-reporting.

distribution missing tax revenues can be recovered and their sources. For a complete understanding of how the tax system shapes incentives, and how it treats different taxpayers, it is necessary

to complement information on the distribution and composition of reported income with that on

unreported income.

Our fndings also have implications for the academic literature on tax evasion. Recent research

suggests that taxpayers seldom evade taxes on third-party-reported income (Kleven et al., 2011;

Carrillo et al., 2017; Slemrod et al., 2017; IRS, 2019), and that deterring evasion where enforcement

is not supported by third-party information requires increasing the audit rate, the penalty rate, or

tax morale (Luttmer and Singhal, 2014). This characterization works well for the middle and the

bottom of the income distribution, but it misses the importance of the concealment of evasion (even

from auditors) at the top and the adoption of aggressive interpretations of tax laws. Our results

suggest that beyond audit rates and/or penalties, compliance among high-income, high-wealth

taxpayers is infuenced by audit procedures, audit selection rules, and the subtleties of reporting

requirements. From a government revenue perspective, the top of the income distribution is the

sub-population where understanding tax evasion is the most important, due to the concentration

of income and the progressivity of the individual income tax.

Due to a lack of data, there is relatively little research about noncompliance by high-income

individuals. Recent research sheds light on one such form of noncompliance, offshore tax evasion

(Johannesen et al., 2020; Alstadsaeter et al., 2019; Londoño-Vélez and Ávila-Mahecha, 2021). Relative to this body of work our main contribution is to cover a wider range of sophisticated noncompliance, including most importantly noncompliance through private businesses, a key source of

income at the top of the income distribution. This is particularly important in the US, where a high

and rising fraction of taxable individual income at the top of the distribution fows from private

pass-through businesses (Cooper et al., 2016; Smith et al., 2019). As much as 50% of taxable income

for the top 0.01% highest earners derives from such pass-throughs, a distinguishing feature of the

US tax system. Understanding noncompliance involving pass-throughs is therefore essential to

paint an accurate picture of the size and distribution of US individual income tax noncompliance.

Our results suggest that the growth of pass-through structures since the 1980s coincided with the

development of new strategies for sophisticated tax evasion, strategies facilitated by pass-through

ownership structures. We fnd that circa 2007, evasion via pass-through businesses made a contribution to the tax gap that was perhaps twice as large as the contribution of offshore evasion. Since

2007, offshore evasion has been subjected to an extensive international crackdown (De Simone

et al., 2020; Casi et al., 2020), while, as we discuss further below, pass-through businesses have

grown in importance and audits of pass-through businesses have fallen dramatically. Compared

to earlier work, our results suggest that sophisticated tax evasion is broader and more systematic

than the concealment of wealth in offshore accounts specifcally, and that tax policy shapes the

types of sophisticated evasion strategies that high-income, high-wealth persons adopt.

The rest of this paper is organized as follows. Section 2 studies the distribution of noncompliance detected in individual random audit data. Sections 3 and 4 provide direct evidence that

offshore and pass-through evasion are highly concentrated at the top of the income distribution

and rarely uncovered in individual random audits. In Section 5, we impute the effect of accounting for undetected evasion on rates of under-reporting through the income distribution and the

distribution of unreported income. Section 6 summarizes our theoretical reasoning, and Section 7

concludes. The Appendix contains additional empirical and theoretical analysis.

The Distribution of Evasion Detected in Random Audits

2.1

Background on IRS Random Audits

The National Research Program (NRP) random audits are the main data source used to study the

extent and composition of individual tax evasion in the United States (see, e.g., Andreoni et al.,

1998; Johns and Slemrod, 2010; IRS, 2016a, 2019; DeBacker et al., 2020).4 NRP auditors assess compliance across the entire individual tax return—the Form 1040—based on information from the

schedules of this Form, third-party information reports, the taxpayer’s own records, and measures

of risk comparing all this information to information on the broader fling population.5 The procedures followed in the 2006–2013 random audits were standard audit procedures for audits of

individual taxpayers conducted by the Small Business and Self-Employed operating division of

the IRS.

Our analyses of the NRP pool data from the random audits conducted for tax years 2006–2013.

The NRP uses a stratifed random sample which over-samples top earners. Our pooled sample

includes 105,167 audited taxpayers, of which 12,003 are in the top 1% of the reported income distribution. We use the NRP weights to compute statistics that are representative of the full population of individual income tax flers. Our sample is large enough to obtain precise estimates for

groups as small as the top 0.01% (although splitting this top group by other characteristics tends

to leave us with too little statistical power for informative analysis). Our main statistic of interest

is the rate of income under-reporting—the amount of income under-reported expressed as a fraction

of true income.6 In the main text we focus on point estimates; confdence intervals for the rate of

income mis-reporting are reported in the Online Appendix (Figure A1). We use a market income

Further background on the NRP is in the Internal Revenue Manuals here: https://www.irs.gov/irm/part4/irm_

04-022-001.

We use the terms “NRP audits” and “NRP auditors” to refer to audits conducted as part of the National Research

Program. Earlier IRS random audit studies under the Taxpayer Compliance Measurement Program (TCMP) consisted

of line-by-line examinations of the individual tax return. The NRP aims to provide a similarly comprehensive measure

of compliance at a reduced administrative cost and burden on the taxpayer. See Brown and Mazur (2003) for more on

the TCMP and how the NRP uses revised procedures to achieve similar objectives.

Tax Gap studies (IRS, 2016a, 2019; Johns and Slemrod, 2010) often estimate a similar quantity called the Net Misreporting Percentage, income under-reporting divided by the total of the absolute value of true income, which can differ

from what we estimate for components of income that can be negative. We use a different term here partly because we

never use absolute values of negative components of income.

concept from Piketty and Saez (2003), defned as total income (Line 22 of the 2012 Form 1040) less

Social Security benefts (Line 20b), unemployment compensation (Line 19), alimony (Line 11), and

taxable refunds (Line 10).7 As we show below, other income defnitions yield similar results.

It has long been acknowledged that in the context of an audit, some noncompliance may go

undetected. The IRS uses a methodology, known as Detection Controlled Estimation (DCE), to

estimate undetected noncompliance (Feinstein, 1991; Erard and Feinstein, 2011). Offcial estimates

of the individual income tax gap for the period we study (e.g., IRS, 2019) use DCE methodology, as

does prior work on the distribution of noncompliance (Johns and Slemrod, 2010). In this section,

we describe what is detected in the course of individual random audits without any correction

for undetected noncompliance.8 We incorporate the additional under-reporting identifed by DCE

into our analysis in Section 5.3; until this Section, we disregard DCE.

2.2

The Distribution of Detected Evasion

We start the analysis by showing the rate of detected income under-reporting across the income

distribution. Unless otherwise noted, all our analyses in this paper rank taxpayers by their estimated true income (with different measures of “true income” depending on the method used to

estimate unreported income).9 On aggregate, 4.0% of true income is found under-reported in NRP

random audits.10 Figure 1a shows that the rate of income under-reporting hovers around 4% to 5%

through most of the income distribution and falls sharply within the top 0.1% to less than 1% in

the top 0.01%.1112 As shown by Appendix Figure A1, this profle of evasion is relatively precisely

estimated.

Figure 1a also decomposes unreported income by type of income. By far the largest type of

Piketty and Saez (2003) use this market income concept to examine the composition of income, while most of their

reported tabulations are based on a gross income concept that does not exclude these components of income but does

exclude capital gains. See further discussion in (Piketty and Saez, 2001, pp. 42-44).

DeBacker et al. (2020) perform a similar analysis and obtain similar results. However, because of our subsequent

results on offshore and business evasion, our interpretation of the patterns seen in the NRP is different. We argue that

the low detected evasion at the top is a consequence of the fact that sophisticated evasion is less likely to be detected,

not that high-income taxpayers are much more compliant.

True income is reported income plus unreported income. Ranking taxpayers by reported income would lead to

downward-biased estimates of the rate of income under-reporting at the top, because the act of under-reporting income

moves taxpayers down the distribution of reported income. Figure A3 and Table A2 illustrate how re-ranking from

reported to corrected income shapes these estimates.

This estimate is smaller than the offcial IRS (2019) estimates of noncompliance for this period, in which 11% of

aggregate true income is unreported, because the offcial estimates include undetected noncompliance estimated by

DCE methods.

Unless otherwise noted, our estimates include all taxpayers, including those found over-reporting income and those

with no change in income upon audit. Taxpayers who under-reported income under-reported 4.5% of aggregate true

income, while taxpayers who over-reported income over-reported the equivalent of 0.5% of aggregate true income. The

majority of over-reported income is in the bottom half of the exam-corrected income distribution; the implications of

over-reporting for aggregate tax liabilities and for noncompliance at the top are negligible.

We omit the P0-P10 group in Figure 1a because aggregate exam-corrected income in this group is negative, which

complicates the interpretation of the rate of under-reporting. This excludes 1.4% of total net under-reporting from

depiction in the Figure. Although about 21% of total net under-reporting is found for individuals in the bottom bin of

the reported income distribution, we show in Appendix Table A2 that the vast majority of this under-reporting is done

by taxpayers who rank further up in the exam-corrected income distribution. See also Figure A3 and Table A3.

unreported income is Sole Proprietor income, reported on Schedule C of the income tax return.

Under-reporting in this category comprises about 50% of all detected evasion. The next-largest

category involves corrections to Form 1040 Line 21 income (“Other income”), which mostly refect

disallowed business losses carried over from previous years. Adjustments to line 21 are relatively

uncommon but they can be large when they occur. The taxpayers with the largest adjustments in

this category typically have negative reported income due to past business losses, but once those

losses are disallowed, their exam-corrected income falls in the top 1% of the income distribution.

Additional information on the composition of under-reported income in the full population and

in the top 1% is reported in Appendix Table A1.

Four remarks are in order. First, our results are robust to using other defnitions of income.

Appendix Figure A2 shows that using adjusted gross income rather than market income gives

nearly identical results except at the very bottom of the distribution. Second, NRP random audits

detect little evasion on wages, interest, dividends, and capital gains. A natural explanation for this

fact is that these forms of income are subject to substantial third-party reporting (IRS, 2016a; Kleven

et al., 2011).13 However, interest, dividends, and capital gains accruing to offshore accounts only

started being subject to information reporting with the implementation of the Foreign Account

Tax Compliance Act in 2014, after our period of study. Thus, the low evasion rates on fnancial

capital income recorded in the 2006–2013 NRP may, in part, be due to the fact that some evasion

on offshore capital income went undetected.

Third, there is an asymmetry in the detected rates of evasion across different types of business

income. For sole proprietor income, which is supported by relatively little third-party information

(Slemrod et al., 2017), the under-reporting rate is 37% overall and 19% for the top 1%. For partnership and S corporation business income, also supported by relatively little third-party information,

detected noncompliance is much lower: 5% overall and 2% for the top 1%.14 As discussed below,

sole proprietor income is subject to extensive examination in the context of the NRP, while the

examination of S corporation and partnership business income faces practical limitations.

Fourth, detected evasion among those with very high incomes is extremely low. In the top

0.01% by exam-corrected income, just 0.6% of true income is under-reported. This is a direct consequence of the fact that the NRP uncovers little noncompliance on interest, dividends, pass-through

business income, and capital gains—the key sources of income in the top 0.01%. This observation

is especially surprising given the lack of comprehensive information reporting on key sources of

income at the top. Figure 1b illustrates how the extent of information reporting changes across

Capital gains were subject to some information reporting throughout our period of study; a reform added the

requirement that brokers report not just sale price but also cost basis starting in tax year 2011.

Schedule K-1 provides some information on pass-through business income, but whether this is truly “third-party”

information depends on whether a given individual owner can control what is being reported on the K-1. Moreover,

the K-1 amounts could already be understating income due to under-reporting on the business’s tax return. We discuss

both of these issues in Section 4.

F IGURE 1: D ETECTED U NDER -R EPORTING AND THE C OMPOSITION OF I NCOME A CCORDING

TO R ANDOM A UDIT D ATA

(a) Detected Under-Reporting

P0-P10 omitted because total true income < 0;

under-reporting / abs(true income) = 2.7% in P0-P10

F1040 "other" income

Partnerships and S-corporation

income

Rental income

Schedule C income

Dividends

Interests

Wages

Capital Gains

P99.99-100

P99.95-P99.99

P99.9-99.95

P99.5-P99.9

P99-99.5

P95-99

P90-95

P80-90

P50-60

P40-50

P30-40

P20-30

P10-20

P70-80

Other

P60-70

income under-reporting, % of true income

(reported + unreported)

Position in the income distribution (exam-corrected income)

(b) Composition of Income

Income Share by Information Category

100.0%

90.0%

80.0%

Substantial Information and

Withholding

70.0%

60.0%

50.0%

40.0%

30.0%

Some Information

20.0%

10.0%

Little or no Information

P99.99-100

P99.95-P99.99

P99.9-99.95

P99.5-P99.9

P99-99.5

P95-99

P90-95

P80-90

P70-80

P60-70

P50-60

P40-50

P30-40

P20-30

P10-20

0.0%

Position in the distribution (exam-corrected income)

Notes: Figure 1a shows unreported income detected in 2006–2013 NRP individual random audit data, without any

correction for undetected evasion, expressed as a fraction of exam-corrected income. Tax units are ranked by their

exam-corrected market income (defned as total income reported on form 1040 minus Social Security benefts, unemployment insurance benefts, alimony, and state refunds). Detected unreported income decreases sharply within the

top 1% of the income distribution. Misreporting of Schedule C income comprises the bulk of detected evasion. Little

evasion is detected for partnership and S corporation business income and fnancial capital income, important sources

of income at the top. We omit the P0-P10 group in Figure 1a because aggregate exam-corrected income in this group

is negative (refer to Appendix Tables A2 and A3 for more information on the bottom 10%). Figure 1b plots the share

of exam-corrected market income in each of the four categories used by IRS (2016a) to describe the comprehensiveness

of third-party information and withholding. “Little or no information” includes non-farm proprietor income, line 21

other income, rents and royalties, farm income, and form 4797 income. “Substantial information” includes interest and

dividends. “Some information” includes partnership and S corporation business income and capital gains. “Substantial

information and withholding” includes just wage and salary income. Chart 1 of IRS (2016a) reports estimated underreporting rates of 63% for “little or no information,” 19% for “some information,” 7% for “substantial information,”

and 1% for “substantial information and withholding.” We observe in the fgure a shift in the composition of income

from the bottom 99% to the top of the top 1%, away from income with substantial information reporting and toward income with only limited information reporting (e.g. Schedule K-1 reporting for partnership sand S corporations, 1099-B

reporting for some but not all capital gains).

the income distribution. In the bottom 99% of the distribution, about 90% of market income is

wage and pension income, which is subject to substantial information reporting and withholding.

The remainder is primarily self-employment (sole proprietor) income with little to no information reporting. Within the top 1%, there is a dramatic shift away from wages, pensions and sole

proprietor income, towards private business income and capital gains, for which there is only limited information reporting.15 Prior literature suggests a strong empirical relationship between the

presence of third-party information and rates of evasion (Kleven et al., 2011; IRS, 2019). IRS (2019)

estimates that the rate of under-reporting on income subject to substantial information reporting

and withholding is just 1%, while the rate of under-reporting on income subject to "some information" reporting is 17%. As such, we might expect the shift in the composition of income we observe

in Figure 1b to be accompanied by an increase in the importance of under-reporting of investment

and business income, but we observe the opposite in Figure 1a.

2.3

Direct Evidence for Undetected Evasion at the Top

We now provide simple and direct evidence that the random audit data miss signifcant top-end

evasion. We compare the amount of evasion estimated at the top of the income distribution in random audit data with the amount found in operational audits. Operational audits include all audits

other than NRP random audits: correspondence audits, conventional in-person audits, sophisticated audits of high-income/high-wealth individuals, and a variety of other specialized audit programs. The IRS prioritizes audits of taxpayers who, based on a variety of factors, it expects to be

non-compliant, so operational audits are not typically conducted at random. On average between

2007-2013, 17% of tax units in the top 0.01% were audited per year.16

Because only a fraction of the population is subject to an operational audit in a given year,

evasion uncovered in operational audit data should be much lower than the population-weighted

NRP estimates of evasion. However, as shown by Figure 2, this is not the case at the top. The

population estimate of evasion for the top 0.01% from the NRP falls substantially below the amount

detected in operational audits alone. On average over fscal years 2007-2013, in operational audits,

auditors recommended $967 million (in 2012 dollars) per year of additional taxes for the top 0.01%

(by reported income). In the extreme scenario where all non-audited taxpayers evaded zero, this

implies that the top 0.01% by reported income evaded at least 1.2% of taxes owed. This lower

bound for the amount of evasion in the top 0.01% is already far larger than the NRP point estimate,

which is less than 0.05%.17 In the next two bins—P99.9 to P99.95 and P99.95 to P99.99—the lower

In the top 0.01%, about two-thirds of income derives from private businesses, capital gains, and “other income.”

An additional 15% derives from interest and dividends, some of which (e.g., investment income accruing to offshore

accounts) were subject to limited information reporting during our sample period

Appendix Figure A5a depicts the fraction of tax units subject to an operational audit by fscal year for the top 1%,

0.1% and 0.01% of reported AGI over time. Consistent with publicly available data, audit rates increased from the

beginning of our sample period until about 2013 and then fell due to budget cuts. Public data on audit rates comes from

the annual IRS Data Book, available at https://www.irs.gov/statistics/soi-tax-stats-irs-data-book.

We note that the difference between the NRP tax gap and the operational audit tax gap in Figure 2 may not be

F IGURE 2: O PERATIONAL A UDITS VER P OPULATION -W EIGHTED R ANDOM A UDITS

Recommended additional tax

as a % of corrected tax liability

NRP, no DCE

(TY2006-2013)

Operational Audits

(FY2007-2013)

P99.99-P100

P99.95-P99.99

P99.9-99.95

P99.5-99.9

P99-99.5

P95-99

P90-95

Position in the reported income distribution

Notes: This fgure compares noncompliance detected in operational audit data to population estimated noncompliance

in the NRP, focusing on the top 10% of the distribution. Unlike in other fgures, we rank individuals by reported income,

and we focus on unpaid taxes, including e.g. over-reporting of deductions and credits, rather than the extent of underreporting of income. Corrected income and under-reporting of income specifcally are not observed in the operational

audit data. As in all our analysis of NRP audit data, we use the auditors recommended adjustments before any disputes

and appeals. We plot total evaded tax as a fraction of total tax due in each bin, for both operational audits (pooling

fscal years 2007–2013) and population-weighted NRP audits (pooling fscal years 2006–2013). Random audits uncover

a very small amount of evasion in the top 0.01% by reported income. Operational audits represent a lower bound on

total evasion because they only capture evasion of audited taxpayers. Yet they uncover substantially more evasion than

the NRP population estimate in the top 0.01%.

bound based on unpaid taxes according to operational audits is comparable to the full population

NRP estimate, despite the fact that less than 10% of taxpayers in these groups were subject to an

operational audit. These results provide direct evidence that NRP audits from our sample period

miss some noncompliance at the top.18

Random audit data from the period we study face two main limitations. First, the available

statistically signifcant, due to sampling variation in the NRP (see Appendix Figure A1). However, a statistically significant difference is not necessary for the substantive point we make here. If operational audits of 17% of the top 0.01%

each year uncovers half of all evasion in this group, which is less extreme than the lower bound scenario but still very

conservative, the NRP point estimate should be twice the operational audit total. The NRP point estimate is suffciently

precise to rule this out with high confdence.

Although Figure 2 provides an informative comparison between what is assessed in operational vs. NRP audits,

it is not directly informative about the true level of the tax gap at the top, for two reasons. First, since only a small

and selected fraction of the population is subject to an operational audit, one cannot infer the true level of tax evasion

using operational audit data alone. Second, because in operational audit data we only observe reported (not corrected)

income in the year of the audit, we rank tax units by reported income in Figure 2. When ranking by reported income,

top earners mechanically have low evasion, since they are selected on high declared income.

tools, procedures, and resources placed limits on the extent to which some types of evasion are

detected, especially at the top. Until recently, there was little evidence available on the nature and

magnitude of these sophisticated types of evasion, making it hard to quantify them. We examine

new evidence on this issue in Sections 3 (offshore evasion) and 4 (pass-through businesses). Second, even for a given set of tools and procedures, examiners varied in their propensity to detect

evasion (e.g., because of different experience). The DCE methodology we incorporate in Section 5.3

was designed to address this issue.

Offshore Evasion

In 2008, the IRS and the U.S. Justice Department began an ambitious crackdown on offshore tax

evasion, described in Johannesen et al. (2020). Key steps in this process included 1) legal actions

against specifc foreign banks, during which banks turned over records on Americans’ concealed

accounts, 2) the establishment in 2009 of Offshore Voluntary Disclosure programs, whereby taxpayers could disclose prior noncompliance and pay penalties but avoid criminal prosecution, and

3) the 2010 passage and 2014 implementation of the Foreign Accounts Tax Compliance Act. In

this Section, we leverage the retrospective information created by this crackdown to study how

accounting for offshore evasion modifes evasion detected in 2007, the year preceding the start of

the crackdown.

3.1

Background and Data on Offshore Evasion

To frst give some texture to our analysis, we gathered all public records from the Department of

Justice website listing indictments, settlements, or convictions for offshore tax evasion in 2009-2021

(DOJ, 2021). These records contain descriptions of 178 cases of prosecuted offshore tax evasion.

Naturally, these cases are selected and likely represent the high-end in terms of noncompliance.

A few results are worth noting. First, noncompliance in this sample is sophisticated. In 78% of

the cases, offshore assets were held through at least one intermediate offshore structure (e.g., shell

company or trust) and in 35% of the cases through multiple intertwined offshore structures.19

Second, the amounts of unreported income are highly skewed, ranging from hundreds of thousands of dollars to $200 million over several years. Third, in nearly all cases, individuals were

accused or convicted of failing to pay tax on the fnancial income generated by the assets in the

account. Moreover, in about 30% of cases, the records suggest that pre-tax business profts were

also diverted offshore and the corresponding income taxes under-paid. Last, 93% of prosecuted

individuals were male.

We now present the IRS micro-data we use for our analysis of offshore evasion. We construct

two lists of individuals who are likely to have been evading taxes on income from their offshore

assets prior to the crackdown that started in 2008. These lists build on Johannesen et al. (2020), who

Because descriptions of cases do not always detail how concealed offshore assets were held, these numbers are

lower bounds for the frequency of sophisticated holdings of assets.

found that the initial wave of enforcement in 2008 and 2009 caused a large increase in the reporting of offshore accounts and the associated fnancial income. The frst list comprises participants

in the Offshore Voluntary Disclosure (OVD) Program. We gathered data on all OVD Program participants from 2009 to 2015 and matched 50,020 OVD participants to their individual tax returns.20

We refer to this sample as the OVDP sample.

The second list consists of individuals reporting that they own offshore assets by fling a Foreign Bank Account Report (FBAR) for the frst time between 2009 and 2011. U.S. persons that are

the benefcial owners of more than $10,000 in offshore fnancial wealth have been required to disclose this wealth to the government since the 1970s by fling an FBAR. We use only frst-time FBAR

flers with U.S. addresses, disclosing an account in a tax haven.21 Johannesen et al. (2020) show that

the large majority of these taxpayers had been evading U.S. tax on these assets prior to disclosing

them in response to enforcement.22 We match 31,752 such taxpayers to their individual income tax

returns. We refer to this sample as the FBAR sample. Individuals in this sample disclosed $124

billion in offshore wealth between 2009 and 2011. Total reported FBAR wealth was about $275

billion in 2011, indicating that a sizable share of all wealth reported on FBARs corresponded to

wealth in tax havens newly disclosed by flers with U.S. addresses. As we discuss below, the best

estimate we know of for total wealth held in tax havens by US households is around $1 trillion

in 2007 (Alstadaeter et al., 2018). This suggests that around 12 percent of all the wealth that was

concealed in 2007 was disclosed by individuals in our FBAR sample in 2009–2011.

We rank each individual in these lists in the income distribution, using income data from the

tax year after the disclosure of the offshore account. The results in Johannesen et al. (2020) suggest

that this is the year in which individuals start complying with their tax obligations. We rank

individuals by adjusted gross income (AGI) for simplicity. We report similar results with rankings

based on other defnitions of income.

3.2

Empirical Analysis of Offshore Evasion

We frst show that NRP audits very seldom detected offshore tax evasion in the time period we

study. As the NRP stratifed random sample and the offshore lists contain disproportionately

many observations of high-income individuals, there is some overlap between them—i.e., some

This 50,020 fgure does not include approximately 6,000 program participants who we could not match to their

individual tax returns. About two-thirds of these were businesses. The rest did not fle a tax return in the year we

wished to analyze, e.g., because their participation in OVDP was too recent.

We use the same list of tax havens as Johannesen et al. (2020): Anguilla, Antigua and Barbuda, Aruba, Bahamas,

Bahrain, Belize, Bermuda, British Virgin Islands, Cayman Islands, Cook Islands, Cyprus, Dominica, Grenada, Guernsey,

Hong Kong, Isle of Man, Jersey, Liechtenstein, Luxembourg, Malta, Mauritius, Monaco, Netherland Antilles, Panama,

Singapore, St. Kitts and Nevis, St. Lucia Island, Switzerland, and Turks and Caicos Islands. Johannesen et al. (2020)

constructed this list using countries mentioned in OECD (2000), plus some additional countries with strong banking

secrecy. This is not an offcial defnition used by the IRS, which has no offcial list of countries it considers tax havens.

We note that this sample may contain some individuals who had an offshore account for legitimate reasons and

were unaware of their FBAR fling obligation prior to increased enforcement, but we fnd it less plausible that very

high-income individuals with accounts in havens were unaware of these obligations.

of the taxpayers in our lists of likely offshore evaders were also randomly audited as part of the

NRP. Before tax year 2009, FBAR compliance was not specifcally examined in NRP audits. After

auditors started examining FBAR compliance, 41 taxpayers were subject to a NRP audit before

they appeared on our list of frst-time FBAR flers with US addresses and haven accounts. Of

these, fewer than 5 were fagged by NRP examiners for non-compliance with their FBAR obligation. Among all frst-time FBAR flers, 187 taxpayers were audited in the NRP before disclosing

an offshore account in 2009-2011, and again in fewer than 5 cases was any FBAR non-compliance

detected. Likewise, of the 98 OVD participants whose FBAR compliance was examined in a random audit before entering OVD programs, fewer than 5 were fagged for FBAR non-compliance.

In brief, random audits very rarely uncovered offshore evasion among taxpayers who were in fact

likely to be evading.

Two caveats about our interpretation of these fndings are worth noting. First, these fgures

are based on small samples. Second, taxpayers with offshore evasion detected in a random audit

would no longer be non-compliant by the time we construct our lists of likely evaders. To alleviate

these concerns, in Figure 3a we plot the rates of detected noncompliance over offshore wealth in the

full NRP sample (not restricting to taxpayers who also appear in our lists of likely evaders). These

rates are extremely small. In fact, they are much smaller than the fraction of taxpayers from each

group of the income distribution that appear on our lists of likely evaders (shown in Figure 3b), a

lower bound for the true fraction of offshore evaders. As Figure 3b shows, the probability to appear

on our offshore lists reaches close to 7% in the top 0.01% of the income distribution.23 Far more

high-income individuals appear on our lists of offshore evaders than the (population-weighted)

estimates from NRP audit data would suggest, even though these lists only cover a subset of all

offshore evaders.

Figure 3c shows the distribution of offshore wealth by rank in the income distribution using

wealth reported on FBARs in the FBAR sample. Of the $124 billion in offshore wealth disclosed,

21% belongs to the top 0.01% of the income distribution and 29% to the rest of the top 0.1%. This

offshore wealth is far more concentrated at the top than non-hidden wealth, of which about 7%

belongs to the top 0.01% according to estimates from Saez and Zucman (2016), as updated in Saez

and Zucman (2020).24 The contrast confrms that fndings in the previous fgures were not simply

driven by the overall concentration of wealth at the top of the distribution, but that concealed

offshore wealth is especially concentrated at the top. Even the relatively small amount of FBAR

wealth attributed to the bottom 90% of the income distribution (17% when ranking by AGI and

Appendix Figure A4 shows that our choice of income defnition matters little for these estimates. We compare

specifcations using AGI, as in Figure 3b, with total positive income (which essentially re-ranks those with large business

losses towards the top), and the sum of interest, dividend, and capital gains income (a rough proxy for wealth). The

profle is similar across these three income defnitions, though it is steepest for fnancial capital income, followed by

positive income.

Our focus is documenting the large difference between ownership shares of offshore and domestic wealth rather

than the change in wealth shares over time. Using alternative estimates of wealth shares would lead to similar fndings.

F IGURE 3: F INDINGS FROM D ATA ON O FFSHORE E VASION

(a) Detection of offshore accounts in NRP

(b) Share Disclosing by Income Rank

5.0%

4.5%

Compliant

4.0%

Share of taxpayers

3.5%

OVDP + FBAR

3.0%

2.5%

First-time FBAR filers 2009-11

(haven accouts & US filers only,

excl. OVDP participants)

2.0%

1.5%

1.0%

Non-compliant

0.5%

OVDP

P99.9-100

P99.99-100

P99.9-P99.99

P99.5-99.9

P99-99.5

P95-99

P0-50

Position in the income distribution (exam-corrected income)

P90-95

P50-90

P99-P99.9

P91-P99

P51-P90

P0-P50

0.0%

Position in the income distribution (AGI after disclosure)

Non-hidden wealth

(Saez and Zucman, 2020)

40%

FBAR wealth, bottom 90%

reassigned to top 10%

30%

Raw FBAR wealth

% of wealth

20%

10%

Position in the income distribution (AGI after disclosure)

P99.99-100

P99.9-P99.99

P99-99.9

P90-P99

P50-90

P0-50

Note: This fgure summarizes the main fndings on offshore wealth from data on the OVDP and FBAR samples of likely

offshore evaders. Panel (a) reports the share of taxpayers estimated to own either a compliant or non-compliant offshore

account in the NRP random audit data. Auditors indicated whether each taxpayer had an FBAR fling requirement and

whether the taxpayer complied with that requirement. We observe that rates of detected non-compliance are very low,

much lower than the lower-bound rates of non-compliance shown in panel (b). Panel (b) plots the fraction of the full

population that appears on our list of likely offshore evaders, by bin of adjusted gross income (measured in the tax year

after disclosure of the offshore account), accounting for overlap between the lists (as OVD participants were required

to fle delinquent FBARs). We observe a steep profle of the probability of disclosing a previously hidden account by

income rank. Nearly 7% of taxpayers in the top 0.01 percent of the income distribution appears in one of the two

lists. In Panel (c), we plot wealth shares for non-hidden wealth from Saez and Zucman (2016) updated in Saez and

Zucman (2020) by bin of market income (defned as total income reported on form 1040 minus Social Security benefts,

unemployment insurance benefts, alimony, state refunds, and other income), versus wealth reported on FBARs by

the frst-time FBAR flers with U.S. addresses and accounts in tax havens (ranking by positive market income after

disclosure). We observe that FBAR wealth is extremely concentrated at the top of the income distribution.

11% when ranking by total positive income) may actually belong to the top. Most FBAR wealth in

the bottom 90% comes from a small number of large accounts; the median level of FBAR wealth

for those in the bottom 50% of the distribution is around $200,000 and the mean is $2.5 million.

If the true income of these large account holders could be observed, it would likely be near the

top. To illustrate how much this might matter, Figure 3c shows the impact of reassigning FBAR

wealth from the bottom 90% to the top 10% of the income distribution, in proportion with the

FBAR wealth already attributed to the top 10%.

One caveat is that our samples of likely evaders are not necessarily representative of all individuals engaging in offshore evasion. Both the FBAR and OVDP samples comprise individuals

who chose to disclose their offshore wealth via the OVDP or via a likely “quiet disclosure” (see

Johannesen et al., 2020). We do not have direct evidence on the direction of the selection, but

data from other countries suggests offshore wealth may be even more concentrated at the very top

than we estimate. Alstadsaeter et al. (2019) use leaked data from HSBC Switzerland and estimate

that 52% of offshore wealth in that bank was owned by taxpayers in the top 0.01% of the wealth

distribution in Scandinavia.

Evasion on Pass-Through Business Income

Pass-through businesses (S corporations and partnerships) are not subject to the corporate income

tax. Instead, all of the income of these businesses “passes through" to their owners’ tax returns,

where it is subject to applicable taxes. Ownership of pass-through entities is highly concentrated

among the highest-income taxpayers, and the use of pass-through business structures has been on

the rise since 1986 (Cooper et al., 2016). As such, obtaining accurate estimates of noncompliance in

such structures is increasingly important. In this section we attempt to make progress on this question by leveraging additional data, focusing on the year 2007 to facilitate aggregation (in Section 5

below) with the offshore evasion results.

4.1

Background on Pass-Through Businesses

Administratively, pass-through businesses fle returns reporting entity-level income. They allocate

this income to their owners on a form called the Schedule K-1. Individual owners should report

the income allocated to them on Schedules K-1 on their individual tax returns. When the owner

of a pass-through business can control what the business reports, the Schedule K-1 issued by that

business is not from an independent or unrelated third-party; in businesses where there is a single

owner or a tight network of owners, the potential scope for noncompliance is similar to sole proprietorship income.25 When the owner is passive, the business may under-report income and as

In 2017, according to Statistics of Income (SOI) tabulations of S corporation tax returns, 66% of all S corporations

(earning 42% of all S corporation business income) had a single shareholder. When there is more than one owner,

misreporting of pass-through income may be riskier than misreporting of sole proprietorship income (Kleven et al.,

2016).

a result allocate too little income to its owners. Confrming the fow of income, deductions, credits and other tax features of pass-through businesses takes expertise and resources. Partnerships

create a specifc additional challenge to the audit process, because partnerships can be owned by

other entities, sometimes leading to complex ownership structures involving numerous partnerships, corporations, trusts, or other intermediaries (Cooper et al., 2016).

Conceptually, there are two types of potential non-compliance by individual owners of passthrough businesses: individual-level under-reporting, e.g., failure to report income allocated to the

individual on a Schedule K-1, or mis-reporting whether that owner is active or passive (which has

tax consequences), and entity-level under-reporting, when income on the pass-through business return is under-reported. In the context of NRP random audits of individuals, resource constraints,

audit procedures, and the tools available to auditors limit the examination of entity-level underreporting. First, as detailed below, when individuals report partnership or S corporation income

on their individual tax returns, auditors rarely examine the tax returns of the corresponding passthrough businesses. Signifcant resources and expertise are necessary to audit up through passthrough returns and ensure that the correct income (and other tax features) fow through to the

relevant individual income tax returns. The IRS does conduct audits to examine tax compliance of

pass-through businesses and their owners, but these audit programs are unrelated to NRP random

audits of individuals. Second, there is no recurring random audit program of pass-through business returns themselves. The most recent random audit program for partnerships was conducted

in 1982. A small random audit study on S corporations was conducted for tax years 2003-2004.

This situation has a number of implications. First, individual random audit data likely underestimate tax evasion occurring via entity-level under-reporting. In Appendix Table A1, we see that

in these data, only 4.6% of partnership and S corporation income is found to be under-reported,

compared to 36.7% for sole proprietorship income, which is subject to direct and comprehensive

examination. Figure 4a contrasts the probabilities that noncompliance was detected in NRP individual random audits for sole proprietorship income (Schedule C) and pass-through business

income (Schedule E). About 60% of the population of taxpayers with Schedule C income are estimated to be under-reporting their Schedule C income. By contrast, only 14.5% of taxpayers with

pass-through business income are estimated to be under-reporting their pass-through business

income. These cases primarily correspond to individual-level under-reporting rather than entitylevel under-reporting. Only 3.8% of taxpayers with pass-through business income are found to be

under-reporting income via an audit of the pass-through business return.26 Detected entity-level

under-reporting represents one-quarter of all instances of detected pass-through under-reporting.

More NRP related business audits occur for S corporations than for partnerships: About 75% of the audited related

businesses are S corporations despite the fact that more NRP participants are partnership owners than are S corporation owners. Based on conversations with experts, our understanding is that when audits of pass-through businesses

occur in the context of the NRP, the audited pass-through entities are typically small, simple businesses where the individual taxpayer being audited has access to the business’s records. Large pass-through businesses—where income is

F IGURE 4: PASS -T HROUGH I NCOME : D ETECTED E VASION IN THE I NDIVIDUAL R ANDOM A U DITS AND C ONCENTRATION

(a) Probability to Detect Unreported Income

(b) Income Underreported (% True Income)

80%

70%

income under-reporting, % of corrected income,

conditional on detected under-reporting

70%

% of taxpayers found under-reporting

60%

50%

40%

30%

20%

Of which:

evasion at the

business level

10%

60%

50%

40%

30%

20%

10%

Partnerships and S-corporations income

Sole proprietorship income

Partnership or S-corp income, Partnership or S-corp income,

no business-level exam

with business-level exam

(d) Investment Income Earned via Passthroughs

50%

60%

40%

Note: Panel (a) shows the unconditional probabilities that noncompliance is detected for sole proprietorship income

(schedule C) and pass-through business income (schedule E) in 2006–2013 NRP random audits. Only 14.5% of taxpayers with pass-through business income are found to be under-reporting their pass-through business income. Among

these cases, only a quarter correspond to cases in which evasion is detected at the business level. Panel (b) shows the

fraction of true business income found underreported in the in 2006–2013 NRP random audits for sole proprietorships,

passthrough businesses when the entity is examined, and passthrough businesses when the entity is not examined.

Panel (c) shows the fraction of reported market income earned via pass-through businesses by reported market income in 2007, by bin of reported market income. Income earned via pass-through businesses includes business income

(S corporation and partnership profts), investment income (capital gains, dividends, and interest), and positive rents

(lumped with business income) fowing from pass-throughs. For rents, we assume that 50% of positive rental income

derives from pass-throughs and assume that this ratio is constant across the distribution; for other forms of investment

income, pass-through investment income is taken from Panel (d). Panel (d) shows the fraction of reported dividends,

interest, and capital gains earned via pass-throughs, by reported market income. These fractions are estimated in the

population-weighted 2006–2013 NRP sample by matching 1040 individual income tax returns to Schedules K-1.

P99.99-100

P99.95-P99.99

P99.5-99.9

P95-99

P99-99.5

P90-95

P80-90

Position in the reported income distribution

P99.9-P99.95

Position in the reported income distribution

P70-80

P99.99-100

P99.9-P99.95

P99.95-P99.99

P99.5-99.9

P95-99

P99-99.5

P90-95

P80-90

P70-80

P60-70

P50-60

P40-50

P30-40

P20-30

P10-20

Dividends

10%

P60-70

20%

P50-60

10%

P40-50

15%

Capital

gains

30%

P30-40

Pass-through capital gains

20%

40%

P20-30

Pass-through dividends & interest

25%

P0-10

Pass-through business income

30%

Interest

50%

P10-20

35%

Share of income flowing from pass-throughs

% of total income from pass-throughs

45%

Second, when entity-level tax returns are examined, detected noncompliance tends to be high.

As reported in Figure 4b, conditional on an entity being examined, more than 70% of business

income is found under-reported on average, as opposed to 50% for sole proprietorships. Consequently, the small number of cases where examiners found under-reporting at the entity level

(one quarter of cases with detected under-reporting) account for 58% of all detected pass-through

under-reporting in dollar terms. Naturally, the high under-reporting in examined businesses may

partly refect selection (auditors may be more likely to audit entities with observable traces of

under-reporting). Still, this fnding suggests that a lack of comprehensive examination of passthrough entities may bias overall estimates from individual random audit data signifcantly.

Third, because the ownership of pass-through businesses is concentrated, the bias occurs primarily at the top of the distribution. Figure 4c shows the fraction of reported income that the

various groups of the reported income distribution earn via pass-through businesses in 2007, our

benchmark year. For the bottom 90%, less than 5% of income comes from pass-throughs. However,

among the highest earners this fraction increases to 45%. In other words, comprehensive audits

of associated pass-through entities would be required to fully examine almost half the income

reported by the highest-income individuals.

Fourth, the bias affects several income categories, not only partnership and S corporation business income. As shown by Figure 4c, the income that fows to individuals from pass-throughs does

not only consist of business income, but also includes dividends, interest, rents, and capital gains.

Determining compliance for such pass-through investment income (reported on Schedule K-1) is

more resource-intensive than for investment income reported (on 1099 Forms) by third parties directly to the individual and the IRS.27 On aggregate, about 15% of dividends, 20% of interest, and

30% of capital gains fow from pass-through businesses. The size of these fows (which largely go

unexamined in the context of individual random audits) suggests potential for a signifcant underestimation of aggregate noncompliance on investment income. Moreover, as shown by Figure 4d,

the fraction of investment income that derives from pass-throughs is highly skewed, so that the

potential bias is again concentrated at the very top. In the top 0.1% of the income distribution,

about 45% of interest and capital gains and 30% of dividends fow from pass-throughs.

Fifth, the magnitude of the tax gap due to pass-through evasion, and thus the magnitude of the

bias from failing to account for entity-level pass-through evasion, is likely to have grown over time.

Appendix Figure A6 illustrates how partnership and S corporation income has grown as a share

of overall income at the top of the distribution since the 1980s. The frst panel shows that top 1%

income shares have risen substantially over time (by 1.83 times), and the majority of that growth

(66%) is attributable to pass-through income, which rises from 11% to 36% of top 1% incomes

concentrated—are very rarely examined.

A further complication is that income that should be taxed as e.g. business income may be reported as capital

gains due to the lower tax rate on long-term capital gains. There is potential for legal avoidance along these lines,

noncompliance and, possibly, some gray area.

over this period. The second panel isolates the trend in partnership and S corporation income

going to the top 1% as a share of total income, which rises steadily over this period (from 1% to

7%), including through the period of heightened offshore enforcement. Additionally, Figure A5b

depicts audit rates of individuals, pass-throughs and other corporations over time using data from

IRS (2022). We observe that pass-throughs were already subject to low audit rates prior to the

precipitous decline in the last decade. These already low audit rates declined by over 80% from

2011 to 2019; the current audit rate for partnerships and S-corporations is about 0.05%, which is

lower than the audit rate for even very low-income individuals and very small C corporations.28

We do not have data with which to credibly quantify the growth in evasion via pass-throughs over

time, but these two facts suggest that, perhaps unlike offshore evasion, pass-through evasion may

have grown substantially in recent years.

4.2

Evidence from Random Audits of S corporations

To provide quantitative insights on entity-level under-reporting, we analyze data on the S corporation random audit program from tax years 2003-2004. The data come from audits of 4,515 S

corporation tax returns. For line items on the S corporation tax return (Form 1120-S), we observe

originally reported and exam-corrected amounts.29 The sample is stratifed based on risk and assets; large S corporations are over-sampled (for more details, see Johns, 2009). We use sampling

weights to construct representative population estimates, pooling tax years 2003 and 2004.

We estimate entity-level under-reporting as under-reported receipts plus over-reported deductions. Following Johns (2009), we do not include corrections to offcer compensation deductions in

our measure of overall income under-reporting because those corrections are offset one-for-one at

the individual owner level by changes in wage income.30 To examine entity-level under-reporting

through the individual income distribution, we also link the audited S corporations’ returns to

their owners’ tax returns via Schedules K-1. To obtain the best available estimate for owners’ rank

in the true income distribution, we use “quasi-corrected” income, defned as reported market income from the owner’s Form 1040 plus the individual owner’s share of detected entity-level S

corporation under-reporting (excluding mis-reported offcer compensation).

In 2019, individuals with between $1 and $25,000 of total positive income were subject to an audit rate of 0.37%,

while C corporations with between $1 and $250,000 of assets were subject to an audit rate of 0.16%. For further details

refer to Table 17 of IRS (2022). The audit rate for partnerships and/or S corporations in 2019 is the lowest of any group

of tax returns for which an audit rate is reported in Table 17. In earlier years, the audit rate for partnerships and S

corporations is among the lowest of the group-specifc audit rates reported in the table.

The focus of the audit program was on the accuracy of business income reporting, i.e., the front page of the Form

1120-S. We do not have data on examinations of, e.g., tax credits claimed by S corporations (i.e., the information on

Schedule K of the 1120-S).

For example, an owner who chooses to under-report offcer compensation relative to the “reasonable compensation"

standard will report too much ordinary business income relative to that standard. Mis-classifcation of offcer compensation as ordinary business income allows owners to avoid Medicare and potentially other employment taxes. This type

of non-compliance is frequently detected in these audits (Johns, 2009). It matters for the tax gap but not for the extent of

income under-reporting. Our only methodological departure from (Johns, 2009) is that we estimate misreporting rates

as a share of total net income, rather than the total of the absolute value of net income (see footnote 6).

These data are subject to similar potential limitations as individual random audit data. First,

we make no corrections for undetected evasion, as in Section 2.2. Second, resource constraints and

audit procedures can limit the comprehensiveness of the examination of noncompliance, especially

in complex businesses. Like individual random audits, these S corporation audits were conducted

by examiners from the Small Business and Self-Employed (SBSE) division of IRS, and the audit

procedures were the same as routine SBSE audits of S corporations. Given these limitations, we

interpret the data as likely providing a lower bound for the extent of entity-level under-reporting

in pass-through businesses. In distributing overall non-compliance through the individual income

distribution, we face a third limitation: we do not observe non-compliance on the individual tax

returns (Forms 1040) of the owners of the audited S corporations. Insofar as owners of underreporting S corporations tend also to under-report elsewhere on their individual tax return (see

e.g. Joulfaian, 2000), the estimated location of entity-level under-reporting will be biased toward

the bottom of the distribution due to insuffcient re-ranking (see Figure A3 and Table A2 for an

illustration of how accounting for re-ranking shifts under-reporting upward in the distribution).

We estimate that S corporations under-report 19.5% of true net business income. Appendix

Table A4 presents a line-by-line decomposition. Of total under-reporting, 26% represents underreported gross receipts or sales,31 69% over-reported deductions,32 and 5% mis-reporting of net

gains from the sale of business property (Form 4797) or “Other Income” (Form 1120-S line 5).

In Figure 5, we examine pass-through under-reporting through the individual income distribution. Figure 5a reports the distribution of 1) reported, 2) exam-corrected, and 3) under-reported

S corporation income by owner income bin.33 Both reported S corporation income and S corporation income corrected for entity-level under-reporting are concentrated at the top of the individual

income distribution, with 49% (45%) of reported (corrected) S corporation income belonging to individuals in the top 0.1%. In total, 79% (75%) of reported (corrected) S corporation income belongs

to owners in the the top 1%. The reported income shares are larger than the corrected income

shares because we estimate that under-reported income is less concentrated than reported income,

with 51% of under-reporting in the top 1%. Most of this difference is attributable to a drop-off in

estimated mis-reporting rates at the very top of the owner income distribution, which we further

examine in Figure 5b.

Figure 5b reports income under-reporting rates in different parts of the owner income distribution, separately for large S corporations—defned as those in the top decile of the net assets

The under-reporting rate for receipts on their own is lower than most other line items: 0.3% of gross receipts is

under-reported, but 0.3% under-reporting rate of gross receipts translates to 5.1% of net S corporation income.

Over-reported deductions here excludes offcer compensation but including the cost of goods sold. The largest

component of over-reported deductions is Form 1120-S Line 19: “Other Deductions.” However, some corrections to

line 19 pertain to line-switching, i.e. when an examiner determines that deductions reported on line 19 should have

been entered elsewhere. Such line switching will net out for our primary purpose here, but it necessitates caution when

interpreting Table A4 in terms of which types of deductions are most commonly under-reported.

We are unable to use the fner income bins we use in previous fgures due to the smaller sample size of the S

corporation study.

distribution—and others. For S corporations in the bottom 90% of the asset distribution, the decline in under-reporting rates as one moves up the income distribution is not entirely surprising

given the characteristics of many of these frms (see, e.g., Smith et al., 2023). In P90-P99.5, S corporations often comprise self-employed, skilled professionals operating as S corporations. These

corporations are not very different from sole proprietorships, and indeed we estimate comparably high under-reporting rates of about 30%. In the top 0.1%, small-asset S corporations are more

likely to represent supplemental income to the individual’s main sources of income, e.g., speaking

fees or consulting work, where there may be more information reporting and less evasion. This

decline in the under-reporting rate may be over-stated due to the insuffcient re-ranking discussed

above, however.34

S corporations in the top 10% of the assets distribution are more similar to C corporations. For

these large S corporations, we estimate under-reporting rates of about 15% through most of the

income distribution, similar to the C corporations under-reporting rate suggested by IRS tax gap

studies of (12-18% depending on the year). For top 0.1% owners, however, the under-reporting rate

for large S corporations falls to about 5%. There are at least two possible interpretations for this

fnding. First, there might truly be little noncompliance in these large, top-owned private frms.

Second, there may be signifcant noncompliance in these large private frms that the random audit

program failed to uncover. This may be a consequence of audit procedures: thoroughly auditing

multi-establishment S corporations (e.g., retail chains or car dealerships), which are concentrated

in the top 10% of assets and top 0.1% of income, requires far more resources than auditing a small

S corporation. Given this, we approach with caution the low estimated rate of entity-level S corporation under-reporting at the very top of the individual income distribution, which echoes the

drop in detected individual noncompliance in Figure 1a.35

To quantitatively investigate this issue, we analyze data on the participation in two specifc

evasive schemes involving pass-through entities: microcaptive insurance schemes and syndicated

conservation easements. In evasive micro-captive insurance schemes, business owners pay a fee

to an offshore company that they also own and supposedly provides insurance services. This payment is deducted as a business expense in the US, reducing US tax liabilities, while the fee does not

increase tax liabilities because the offshore company that receives it is incorporated in a territory

Appendix Table A5 shows estimated under-reporting rates for single owner, two-owner and multiple-owner S

corporations. Under-reporting rates are similarly high among single- and two-owner frms, around 25% on average.

Under-reporting rates are lower, though not trivial, among multiple owner frms at about 12.5%. Single-owner S-corps

are the predominant form across the owner quasi-corrected income distribution, including at the very top. Among each

type, under-reporting rates fall to about 5% at the very top of the distribution.

In Appendix Figure A7, we decompose large S corporations further into those in the 90-95th percentiles and those

in the top 5% of assets. For those in the 90-95th percentiles, under-reporting is fairly stable across the distribution,

including at the top, and the average under-reporting rate among these frms is about 22%. Under-reporting is lower

among top 5% frms, particularly at the bottom and very top of the distribution, though relatively high in the bottom

part of the top 1% of the distribution. Lower under-reporting among top 5% frms could refect higher compliance

among the largest frms or NRP audit procedures, as just discussed.

where there is no income tax (e.g., Bermuda). If the micro-captives do not meet the legal criteria for

providing insurance, then the payments are evasive (for more details, see IRS, 2016b; GAO, 2020).

The IRS identifed microcaptive schemes as abusive and has taken steps to curb them, including

a letter campaign and the issuance of Notice 2016-66, which identifed a class of microcaptive insurance schemes as potential tax evasion and required taxpayers who had participated in these

schemes to disclose their participation on Form 8886, which is the form for disclosing all such

“Listed Transactions” (see IRM 4.32.2.8.2.2.3.1). To identify participants in microcaptive insurance

schemes, we use the initial Form 8886 flings in response to IRS Notice 2016-66. We observe 12,874

S corporation owners disclosing participation in microcaptive insurance schemes during tax years

2017 to 2020 and we assign them to income ranks using their AGI in the year after Form 8886 fling. An internal IRS review of these flings found that the “majority” of them were indeed abusive

GAO (2020).

Syndicated conservation easement schemes involve the donation of the development rights

on property to a non-proft entity for conservation, which generates a tax deduction (see IRM

4.38.1.11.5). By over-valuing the development rights, taxpayers in evasive syndicated conservation easement schemes claim deductible donations larger than the deductions they are entitled

to. The word “syndicated” refers to the version of the conservation easement where a group of

individuals jointly own a pass-through business (typically a partnership) that owns the property

on which the conservation easement is taken (for more details, see United States Senate, 2020; Investigations, 2025).36 To identify participants in likely evasive syndicated conservation easements,

we use the initial Form 8886 flings in response to IRS Notice 2017-10, which identifed syndicated

conservation easements as “transactions of Interest.” We observe 26,117 such Form 8886 flers during tax years 2017 to 2020 and we assign them to income ranks using their AGI in the year after

8886 fling.37 US Department of Treasury (2020) reports statistics on these 8886 flings: in 2018, the

value of the charitable deduction exceeded investors’ investment in the syndicate in all but 1.8%

of disclosures; the ratio of the claimed deduction to investment was 4.9 on average. Total claimed

deductions were about $6 billion in 2016 and about $9 billion in 2018.

Figure 5c presents the probability that an individual appears on our list of S corporation own36

There may also be evasive non-syndicated conservation easements, but we do not have a clean way to isolate them

and they are not directly related to pass-through evasion.

Both frms and individuals respond to these notices. For microcaptive insurance schemes, we limit our analysis to S

Corporation owners to maximize comparability with the S Corporation random audit data. For syndicated conservation

easements, S corporation owners are less well-represented, so we include individuals fling Form 8886 and also the

individual owners of the passthrough businesses fling Form 8886. In the case of passthrough ownership, there is some

subjectivity about which owners should be included. First, some owners may not even claim the relevant charitable

deduction, so we restrict our sample to owners who do claim the Schedule A line item in the year of interest or a recent

prior year. Second, some owners, e.g., minority investors in a complex fund of funds, may not know they are invested

in the position. Together with the Schedule A line item restriction, we use the materiality of the owner’s stake in frms

engaged in this position to proxy for the likelihood that the owner knows about the position. We require the owner owns

at least 0.01% of the pass-through; we obtain a very similar profle with a 0.1% ownership or 1% ownership restriction,

or with no such restriction.

ers participating in a micro-captive insurance scheme by individual income bin. To help us understand how these fndings relate to the S corporation random audit results, we report separate

probabilities of micro-captive participation in large and small S corporations, splitting out frms in

the top decile vs. bottom 90% of the assets distribution as before. Figure 5d plots the probability

for participation in a syndicated conservation easement transaction. In both of these fgures, as for

offshore evasion in Figure 3b, we observe that participation in sophisticated schemes rises sharply

with income even within the top 1%. For micro-captive insurance, we further fnd that the relatively high participation rate among very high-income owners is almost entirely driven by large

S corporations, which contrasts sharply with the random audit fndings in Figure 5b. Figures 5c

and 5d suggest that some of the top-end drop in detected entity-level under-reporting in S corporations from Figures 5a and 5b refects differential detection of evasion through the distribution

of S corporation size × owner income. The available data, however, limit our ability to defnitively correct for the under-detection of sophisticated, undetected, entity-level under-reporting in

passthrough businesses.

Implications for the Distribution of Evasion

In this section, we explore the implications of the fndings of the previous two sections for the

distribution of under-reported income. An implication of our results so far is that evasion at the

top of the distribution is not only under-estimated in random audit data, it is also more uncertain.

We do not have a representative random sample of tax returns in which offshore and pass-through

evasion are comprehensively captured. Instead, we have the evidence above, some aggregate

statistics, and suggestive evidence from other studies. Given this uncertainty, we present a range

of imputations under some transparent assumptions. We present a benchmark set of assumptions

that represents our best guess, and low-end and high-end sets of assumptions in which we make

signifcantly more conservative or aggressive assumptions about the importance of offshore and

pass-through evasion at the top of the income distribution. Each of these entails multiple specifc assumptions, whose importance we further unpack in the Appendix. We begin by adding

imputed undetected offshore and pass-through evasion to NRP estimates without DCE. Then we

additionally impute the distribution of undetected under-reporting identifed by DCE.

5.1

Assumptions for Imputations

Table 1 lists all the assumptions we make in our benchmark, low-end, and high-end imputations.

Assumptions about Offshore Evasion. Our method for imputing offshore evasion follows Alstadsaeter et al. (2019). We designate 2007 as the base year in which we seek to impute offshore

evasion; this is prior to the crackdown on offshore tax evasion that began in 2008.

In each imputation, we begin with an estimate of aggregate offshore wealth in tax havens

owned by US households in 2007. In the benchmark, the fgure is $1,058 billion, the equivalent

F IGURE 5: PASSTHROUGH EVASION : S- CORPORATION RANDOM AUDITS & SPECIFIC SCHEMES

(a) Owner-level concentration of reported,

corrected, & under-reported business income

(b) Rates of entity-level under-reporting

Under-reported net business income (excl off. comp.), % of

corrected net business income

50%

45%

Share of population total, %

40%

35%

30%

25%

20%

15%

10%

40%

35%

30%

25%

20%

15%

10%

P90-95

-5%

Bottom 90%

P90-95

P95-99

P99-99.5

P99.5-99.9

Position in the income distribution (quasi-corrected income)

Reported income

Corrected income

P99.99-100

Under reporting (excl off compen)

P99-99.5

Bottom 90% of S-corp assets

P99.5-99.9

P99.99-100

Top 10% of S-corp assets

(d) Syndicated conservation easement participation rates

2.5%

3.5%

Top 10% of S-Corp assets

Probability of scheme participation, %

3.0%

2.0%

1.5%

1.0%

0.5%

2.5%

2.0%

1.5%

1.0%

0.5%

Bottom 90% of S-Corp assets

Position in the income distribution (AGI after disclosure)

Top 0.01

P99.9-99.99

P99.5-99.9

P99-99.5

P95-99

P90-95

P50-90

P99.99-100

P99.9-99.99

P99.5-99.9

P99-99.5

P95-99

P90-95

P50-90

P0-50

0.0%

P0-50

probability of scheme participation, %

P95-99

Position in the income distribution (quasi-corrected income)

Note: This fgure presents data on entity-level under-reporting in pass-through businesses. Panels (a) and (b) are drawn

from the S corporation random audit data from tax years 2003–2004. We rank individuals by their reported income

plus the S corporation owners’ share of S corporation under-reporting, which we call “quasi-corrected” income. Panel

(a) presents the distribution of reported, exam-corrected, and under-reported S corporation net business income by

owner income bins. We exclude mis-classifed offcer compensation from under-reporting for the reasons discussed

in the text. Panel (b) depicts estimated rates of entity-level under-reporting in S corporations, by owner income bin,

splitting large and small S corporations (i.e. top 10% versus bottom 90% by assets). We observe that under-reporting

is highly concentrated but not quite as concentrated as reported business income, because of a drop in detected underreporting for large S corporations with top-income owners. The next two panels show, however, that sophisticated

evasion schemes are relatively frequently taken up by top-income owners of pass-through businesses. In panel (c) we

depict the rate of participation by owners of S corporations in micro-captive insurance schemes. We split these owners

into two groups based on whether they own small or large S corporations. We observe that sophisticated evasion via

micro-captives is concentrated among top-income owners of large S corporations (the same group Panel (b) suggests

are relatively compliant). In panel (d), we show that rates of participation in syndicated conservation easements are also

high among top-income owners of partnerships and S corporations.

TABLE 1: I MPUTATION A SSUMPTIONS

Aggregate offshore wealth

Share of hidden offshore wealth

Taxable rate of return

on hidden wealth

Distribution of offshore wealth

Rate of under-reported

business income

Rate of under-reported

capital gains

Rate of under-reported

dividends and interest income

Distribution of undetected

pass-through income

The distribution of DCE

-identifed under-reporting

Benchmark

High-End

Low-End

Offshore

$1,058 billion

95%

6.00%

$1.5 trillion

100%

11.00%

$750 billion

85%

4.65%

100% Nordic dist.

100% FBAR dist.

Pass-Through

20%

26%

19.5%

10%

As reported pass

-through income.

As reported pass

-through income.

As Scorporations

DCE

As exam-corrected

incomes by type

of income

Non-compliance

same as sole

proprietorships

As exam-detected

underreporting in a

third macro allocation

60% to top 0.1%

35% to top 0.01%.

Note: This table lists the assumptions we make about undetected under-reporting in our imputations. Refer to the main

text for detail and the rationale on each assumption.

of 1.7% of total U.S. household wealth. This number is taken from Alstadaeter et al. (2018), Appendix Table A.3; Alstadaeter et al. (2018) imputed it by allocating the 2007 global amount of

offshore wealth estimated in Zucman (2013) to each country, using retrospective statistics on the

ownership of offshore bank deposits released by the Bank for International Settlements in 2016.

In the low-end scenario, we assume total offshore wealth is $750 billion. This fgure is based on

the Boston Consulting Group’s (BCG) Wealth Report of 2007, which estimated that wealthy North

American residents held about $37.7 trillion of wealth, 2% of which was held offshore.38 In the

high-end imputation, we assume total wealth is $1.5 trillion, following Guttentag and Avi-Yonah

(2005). These authors built on the BCG Wealth Report of 2003, according to which the total holdings of high-net-worth individuals in the world were $38 trillion, including $16.2 trillion for North

America residents. “Less than 10%” of this wealth was held offshore according to BCG; adopting

Alstadsaeter et al. (2019, footnote 28 p. 2090) list all the available estimates of the global amount of offshore around

2007; the BCG estimate is the second lowest one, immediately after (and close to) an OECD estimate which is not broken

down by country.

this purported upper bound, as in Guttentag and Avi-Yonah (2005), yields a total of $1.5 trillion.

Second, we make an assumption about the share of offshore wealth that is hidden from the

IRS. Our assumptions here are based on United States Senate (2008, 2014) reports investigating

the practices of several Swiss banks in the U.S. (prior to the 2008 crackdown). The investigation

committee found that about 90% of the wealth held by U.S. taxpayers at UBS Switzerland was

undeclared, and that between 85% and 95% of the accounts held by U.S. taxpayers at Credit Suisse

were undeclared. We assume 95% of haven wealth was hidden in the benchmark imputation, 85%

in the low-end imputation, and 100% in the high-end imputation.

Third, we assume a nominal, taxable rate of return on offshore wealth to convert the stock of

concealed offshore wealth to a taxable income fow. In the benchmark imputation, we assume a

6% return. This is an educated guess based on what is known about the portfolio composition of

global offshore wealth and the rate of return on these assets in 2007.39 In the low-end imputation,

we use the average daily 10-year Treasury rate for the year 2007, 4.65%; this is similar to the interest

rate on Swiss bank deposits. In the high-end imputation, we use the return on average equity for

all U.S. banks, averaged over the year 2007: 11%.

Under these assumptions, total under-reported income from offshore accounts is $60.3 billion

in the preferred scenario (0.7% of true total taxable income), $28.7 billion in the low-end scenario

(0.3% of total income), and $165 billion in the high-end scenario (1.9% of total income). Our fourth

assumption concerns how these totals are distributed through the income distribution. We plotted

the share of wealth belonging to different parts of the income distribution for U.S. flers who disclose a haven account by fling an FBAR for the frst time in 2009-2011 in Figure 3c. These shares are

based on partial and potentially selected samples of offshore evaders. Some results in Johannesen

et al. (2020) suggest the new FBAR flers from this period are negatively selected on income relative to all owners of offshore wealth. As such, we assume these share are equal to the distribution

of all concealed wealth our low-end imputation. In the benchmark, we take a weighted combination of the distribution of offshore wealth observed among new FBAR flers and the distribution of

hidden wealth estimated by Alstadsaeter et al. (2019) in Scandinavia, which is more concentrated

at the top. We put equal weight on the two distributions, which implies that 60% of hidden wealth

belongs to the top 0.1% and 35% to the top 0.01% (vs. more than 50% in Scandinavia). In the

high-end imputation, we put 100% of the weight on the Scandinavian distribution.

In the frst draft of our paper, our rationale for this fgure was as follows: Zucman (2013) estimates that in 2007,

around 75% of global offshore wealth was invested in securities (mostly equities and mutual fund shares) and 25% in

bank deposits. The average interest rate paid by Swiss banks on their term deposits was 4.3% in December 2006. Similarly, in 2006 the U.S. Federal fund rate was in range of 4.3% to 5.25%; the total nominal return (dividends reinvested)

was 13.4% for the the S&P 500 (and 20.65% for the MSCI world). With 25% of assets earning a 4.3% return in bank deposits and 75% earning half of a 13.4% return in securities (with the other half in unrealized capital gains), we arrived at

our taxable 6% return. In newer data from third-party reporting under FATCA, discussed in more detail below, interest

and dividend income in offshore haven accounts are somewhat higher than our initial guess; realized capital gains are

probably lower. Overall, a taxable return of 6% continues to appear to be reasonable based on these newer data.

Assessing Offshore Assumptions with Newer FATCA data. We developed the assumptions laid

out above using information that was available when we released the original working draft of

this paper. A recent analysis of far more comprehensive data on offshore wealth in havens, using

the universe of third-party information reports under the Foreign Accounts Tax Compliance Act

(FATCA), allows us to gain some insight about the accuracy of our initial assumptions (Johannesen

et al., 2024). This validation exercise is subject to some caveats. Johannesen et al. study data

from tax years 2015-2018, after FATCA attempted to curb offshore tax evasion. Offshore wealth

and income could be different between 2007 and this period for a variety of reasons, including

repatriation in response to FATCA (De Simone et al., 2020), and the possibility that taxpayers may

avoid realizing capital gains more often in tax-compliant accounts.

That being said, these authors fnd that Americans’ total offshore wealth in havens was around

$2 trillion, which is similar to what we would expect under our frst assumption about total

wealth.40 While income reporting is incomplete in FATCA data, as discussed in detail in Johannesen et al. (2024), their analysis of reported income in offshore accounts suggests that our benchmark 6% return is also reasonable.41 The results in this study are consistent with our assumption

that new FBAR disclosures are negatively selected on income relative to all offshore wealth. The

new FBAR disclosures we analyzed here, and drawing from Johannesen et al. (2020), are restricted

to offshore accounts owned directly by individuals, while Johannesen et al. (2024) fnd that accounts owned by entities, mainly partnerships, are an important component of haven wealth and

much more concentrated in their ownership. These authors estimate that about 33% of wealth in

havens owned by individuals and partnerships belongs to the top 0.01% of the income distribution, compared to 36% in our benchmark imputations and 23% in the new FBAR sample and the

low-end imputation.

Assumptions about Pass-through Evasion. To impute pass-through evasion we require assumptions about 1) what share of various types of pass-through income are under-reported at the entity

level, and 2) how this under-reporting is distributed by owner income. One way to discipline these

quantities would be to rely on data from the S corporation audit study data from Section 4.2. Under

the assumptions that the S corporation random audits detect all entity-level under-reporting and

that the rate and distribution of under-reporting are similar between S corporation business income

and other forms of pass-through income (and similar between S corp study years, tax years 2003-4,

and our period of interest), we would assume 19.5% of pass-through income is under-reported and

If our benchmark for total offshore wealth, $1.058 trillion was correct and offshore wealth grew by 8% year over

year on average from 2007 to 2016, total wealth would equal about $2 trillion in 2016.

The authors estimate a nominal taxable rate of return for all offshore wealth of about 5%. The estimated return is

higher in tax havens: the rate of return in havens is 5.0% based on interest and dividends alone, and it is 6.5% when

including capital gains (imputed as 14% of reported gross proceeds from sales of assets). The estimated return is also

somewhat higher for accounts owned by individuals and US partnerships (7.4% and 6.4% respectively). The correct

number for our imputations is the nominal taxable rate of return for wealth in havens with owners subject to individual

income taxes.

distribute this according to the shares of under-reported income plotted in Figure 5a.

We argue that in light of the available evidence, the S corp study should be regarded as a lower

bound for the amount and concentration of all under-reported income in partnerships and S corporations; we therefore make these assumptions in our low-end imputation. How we discipline

a more middle-of-the-road set of assumptions depends on how we think about the limitations of

these data and legal ambiguities. As discussed above, in our analysis of syndicated conservation easements and microcaptive insurance schemes, the S corporation NRP data are subject to

limitations in their capacity to capture sophisticated and potentially legally ambiguous forms of

noncompliance.

The most important form of pass-through income besides S corporation business income is

partnership income, where we have much less comprehensive data on under-reporting: the last

random audit study of partnerships happened in 1982, when TCMP random audits produced an

estimated income under-reporting rate of 26% for partnerships, though obviously in a different tax

policy regime (GAO, 1995). Partnerships are smaller than S corporations in terms of assets, which

could suggest a higher non-compliance rate (smaller S corps have more non-compliance in the S

corp audit data). Simple partnerships have similar margins for evasion as small S corporations,

while complex partnerships are very different from large S corporations. Empirically, the complexity of partnership structures (defned as tiered ownership) rises sharply with owner income

within the top 1% of the income distribution (see Appendix Figure A8).

There are some additional margins for evasion in complex partnerships; such margins are myriad and usually involve shifting income or assets between partners. One simple example would

be for a US individual to incorporate a shell company in an offshore tax haven, make this company a partner in a US partnership, and conceal the ownership of the company from the IRS.

New reporting rules under FATCA make this more diffcult,42 but this is one potential mechanism

for offshore tax evasion. Partnerships can also allocate income between partners fexibly using

“special allocations” to take advantage of differential tax rates across partners or types of income.

Love (2024) estimates that these allocations generate about $20 billion in tax reductions per year

over 2011-2020, 66% of the which accrues to the top 1% of the distribution (29% to the top 0.01%).

Partnerships can also shift assets within ownership networks to reduce taxes on capital gains and

exploit depreciation rules. The IRS (2024) suggests that this “shell game” generates an estimated

$10 billion in tax reductions per year; they do not report distributional statistics but given the

complexity of the partnerships playing the shell game it is reasonable to expect it is highly concentrated.43 Cooper et al. (2016) fnd that 15% of entity-level partnership income fows in circles

One limitation to FATCA is that such “shell companies” could be converted into “shell banks” and dodge FATCA

reporting, which does not apply to fnancial institutions themselves. The extent to which this is occurring is unclear, but

it is the source of “the largest tax evasion case in U.S. history” (in nominal dollars) (United States Senate, 2022).

While the concentration of S corporation and partnership business income at the top of the owner income distribution is similar, the industry mix is somewhat different. The S corporation industry distribution better mirrors that of C

within ownership networks.44

Although their effect is to avoid tax on a substantial amount of income, the legality of many

of these arrangements is unclear. Some of the tax benefts of fexible partnership arrangements

are legally obtained, and they are subject to anti-abuse provisions in the law. As discussed above,

partnership audits are low and falling (Appendix Figure A5b), so complex positions are seldom

challenged in practice. When they are challenged, the resulting disputes can take years and strain

IRS resources (GAO, 2023). However, a recent study of operational audits of partnerships estimates

that audits of complex (tiered) partnerships yield $20 for every $1 invested in auditing them, compared to $10 to $1 for simple partnerships (Hess et al., 2024), so there is some signal that these

arrangements are not always compliant.

Rather than try and integrate all of the piecemeal evidence above into a complicated structural

imputation, which would require taking an explicit stance on the extent and legality of all these

positions, in our benchmark we elect for a transparent and simple approach, supposing that rates

of under-reporting are equal in the top and bottom of the distribution but total under-reporting

is similar in magnitude to what we assume in the lower bound. In the benchmark imputation for

pass-through evasion, we assume 20% of pass-through business income is under-reported. We

also include some under-reporting of investment income received by pass-throughs (which was

unexamined in the S corporation audit study), assuming 5% of pass-through capital gains, and 3%

of pass-through dividends and interest are under-reported. These assumed rates are equal to the

under-reporting rates for investment income directly earned by individuals in the individual NRP

data (Appendix Table A1). We then assume under-reported pass-through income is distributed

similarly to reported pass-through income. For example, if the top 1% (by reported income) earns

60% of reported pass-through business income, we assume that the top 1% (by true income) also

earns 60% of unreported pass-through business income, which is equivalent to assuming rates of

evasion are constant in different parts of the distribution.45 The resulting distribution of underreporting is more concentrated at the top of the income distribution than the shares of unreported

S corp income (see Figure 5a), which we attribute to limited detection of sophisticated evasion

corporations whereas partnership income is highly concentrated in fnance and holding companies, and these industries account for much of the concentration of partnership income at the top of the distribution (Cooper et al. (2016)).

Therefore, the degree to which the estimates from the S Corp NRP extrapolate to partnership non-compliance across

the distribution also depends on whether non-compliance rates in fnance, holding companies and professional service

partnerships match those in other industries, which may be related to differential use of the practices discussed here.

Love (2024) suggests that these circular structures likely arise due to the organic “daisy-chaining” of partnerships,

implying these structures are unlikely to refect intentional tax fraud. If the end result of this is that income in circular

structures goes untaxed, they could nevertheless be noncompliant.

In the bottom of the income distribution, net pass-through income is negative. Rather than assigning negative

amounts of under-reporting, we assume zero under-reporting belongs to the part of the distribution where total income

is negative. We observe that this does not require assuming that those with reported losses are perfectly compliant, but it

does require that after including evaded income, evaders with reported losses have positive corrected net income. How

we handle negative incomes in implementing the assumption that rates of misreporting are constant over the income

distribution has a negligible quantitative effect on our imputations: we obtain very similar results using e.g. absolute

values to deal with negative total incomes.

in the S corp random audit data and/or the importance of complex partnerships at the very top

of the distribution. In the high-end scenario, we increase the rate of under-reporting of business

income to 26%, and we double the rates of under-reporting of pass-through investment income,

supposing that the higher sophistication of pass-through investment income is associated with

elevated misreporting. The 26% fgure is similar to the estimate from the 1982 TCMP study and

the estimate of sole proprietor misreporting in the top 1% (25% in our NRP data).46 We use the

same distributional assumption as the benchmark imputations in the high-end imputation.

Double Counting. There is potential for overlap between the three components of under-reporting

we include in these imputations: audit-detected evasion in individual random audit data, imputed

offshore evasion, and imputed pass-through evasion. Double counting between imputed offshore

evasion and the individual random audit data is negligible, because detection of offshore evasion

is extremely rare in the random audit data, and detected FBAR noncompliance is associated with

small corrections to income.47 We focused our pass-through imputations on business income, and

our offshore imputations on investment income, in order to minimize double counting between

our imputations for offshore and pass-through income. In reality, pass-through owners may evade

signifcant tax on their business income by diverting profts to offshore accounts; we do not explicitly account for this in our imputations.48 Relatedly, our pass-through imputations presume small

amounts of entity-level misreporting of investment income in pass-throughs, to minimize double

counting of unreported offshore investment income that fows through pass-through businesses.

The main form of double counting we do need to address is entity-level pass-through income

that is included in both our imputations and individual random audit data (see Figure 4a). When

we add imputed pass-through under-reporting per the assumptions laid out above, we also remove entity-level pass-through evasion uncovered in the context of individual random audits.

Specifcally, 57.6% of partnership and S corporation income evasion detected in individual random audits is associated with an entity pick up (i.e., an audit of the corresponding business). We

therefore remove 57.6% of audit-detected partnership and S corporation evasion.49

The 26% fgure from TCMP includes all forms of partnership income, not just business income. We observe that in

all three imputations, our under-reporting rate falls between the random audit fgure for all sole proprietorships (37%

under-reporting rate) and the IRS Tax Gap estimate for C corporations (12-18% depending on the year), which seems

reasonable given the nature of most pass-through business activity. Another source of uncertainty owes to the fact that

taxpayers may attempt to classify income as capital gains to beneft from the relatively low rate on long-term capital

gains (see footnote 27). Because we are focused on overall income-under-reporting, we do not include any of this in our

imputations.

Using the under-reported fnancial income of those found to be FBAR non-compliant during NRP audits in Figure 3a, we estimate that detected offshore under-reporting amounts to at most 0.01% of exam-corrected income in each

income group.

For example, if offshore wealth grew by 5% per year due to evasion on business income, then accounting for this

would be equivalent to increasing our assumed rate of return by 5 percentage points.

An alternative would be to keep the entity-level under-reporting observed in the audit data as is and make a doublecounting adjustment on the imputation side. However, entity pick-ups occur for less than 5% of pass-through owners,

and they are highly selected on the nature of the business (see also footnote 26). We would need to account for the

resulting selection bias in audit-detected entity-level evasion to make the double counting adjustment on the imputation

5.2

Imputation Results without DCE

Figure 6 illustrates how imputing offshore and pass-through evasion reshapes the profle of rates

of under-reporting from Figure 1a. The frst panel illustrates the benchmark imputation, and the

second and third panel illustrate the low- and high-end imputations. The fnal panel compares the

profle of under-reporting across different imputations. Table 2 reports the share of under-reported

income belonging to different parts of the income distribution under our various assumptions.

TABLE 2: I MPUTATION R ESULTS (W ITHOUT DCE): S HARES OF U NREPORTED I NCOME , 20062013, IN % OF T OTAL U NREPORTED I NCOME

P0–10

P10–20

P20–30

P30–40

P40–50

P50–60

P60–70

P70–80

P80–90

P90–95

P95–99

P99–99.5

P99.5–99.9

P99.9–99.95

P99.95–99.99

P99.99–100

Top 1%

NRP Only

1.4%

0.7%

2.8%

4.9%

6.2%

7.6%

9.7%

11.9%

15.6%

12.2%

15.7%

3.9%

5.1%

1.1%

0.7%

0.6%

11.4%

Benchmark

1.1%

0.4%

1.9%

3.3%

4.2%

5.1%

6.6%

8.3%

11.2%

9.3%

15.7%

5.6%

11.1%

3.7%

5.0%

7.5%

33.0%

Low-End

0.9%

0.5%

2.1%

3.7%

4.7%

5.9%

7.6%

10.0%

14.3%

10.6%

18.9%

5.9%

9.0%

1.7%

2.0%

2.1%

20.7%

High-End

1.0%

0.3%

1.5%

2.6%

3.3%

4.1%

5.3%

6.7%

9.2%

7.7%

14.2%

5.7%

12.3%

4.8%

6.9%

14.5%

44.2%

Note: This table reports the results of our imputations for the share of total unreported income belonging to each

part of the income distribution. The frst column reports the distribution of unreported income in NRP data without

DCE or any other adjustments. In the next three columns, we include imputed offshore and imputed pass-through

under-reporting under the imputation assumptions laid out in Table 1. We observe that doing so increases the share of

unreported income belonging to the top 1% by between 9 and 33 percentage points.

Despite ample quantitative uncertainty about how much offshore and pass-through evasion

matter, we extract three robust insights from these imputations. First, these forms of undetected

evasion are concentrated at the very top of the income distribution across all specifcations, so imputed offshore and pass-through evasion mainly loads onto the top 1%. This is unsurprising given

the results in the previous two sections and the corresponding assumptions in the imputations.

Second, across all our imputations, incorporating offshore and pass-through evasion meaningfully alters the profle of under-reporting. Total under-reporting of true income in the top 1% is

2.3% of true income based on random audit data alone, compared to 9.2% in the benchmark, 5.6%

side.

F IGURE 6: I MPUTATION R ESULTS (W ITHOUT DCE): U NDER -R EPORTED I NCOME AS A S HARE

OF T RUE I NCOME

(a) Benchmark

(b) Low-End

20.0%

Under-reported income, as % of true income

(reported + unreported)

Under-reported income, as % of true income

(reported + unreported)

20.0%

18.0%

16.0%

14.0%

12.0%

+ Passthroughs

10.0%

8.0%

+ Offshore

6.0%

4.0%

2.0%

NRP (No DCE)

18.0%

16.0%

14.0%

12.0%

10.0%

8.0%

6.0%

2.0%

Position in the income distribution (exam-corrected income)

(d) Comparisons

20.0%

Under-reported income, as % of true income

(reported + unreported)

Under-reported income, as % of true income

(reported + unreported)

20.0%

18.0%

+ Passthroughs

14.0%

12.0%

+ Offshore

8.0%

6.0%

4.0%

+ Offshore

Position in the income distribution (exam-corrected income)

10.0%

NRP (No DCE)

0.0%

16.0%

+ Passthroughs

4.0%

NRP (No DCE)

2.0%

0.0%

18.0%

16.0%

14.0%

12.0%

10.0%

8.0%

+ low-end sophisticated

6.0%

4.0%

2.0%

0.0%

Position in the income distribution (exam-corrected income)

+ high-end sophisticated

+ benchmark sophisticated

NRP (No DCE)

Position in the income distribution (exam-corrected income)

This fgure reports the results of our imputations. The solid black line in each panel is the same as Figure 1a: unreported

income as a share of true income according to NRP data without DCE or any other adjustments. We then add imputed

offshore and imputed pass-through under-reporting under the imputation assumptions laid out in Table 1. The frst

three panels describt the benchmark, low-end, and high-end imputations respectively; the fnal panel compares how

each set of assumptions shapes the profle of rates of under-reporting including offshore and pass-through evasion. We

observe that accounting for undetected offshore and pass-through under-reporting meaningfully alters the profle of

under-reporting in all imputations, but the extent to which it does so is uncertain.

in the low-end imputation, and 14.9% in the high-end imputation. Rates of under-reporting are

similar in the top 1% and bottom 99% in the low-end imputation and higher in the top 1% in the

other imputations.

Third, pass-through under-reporting is likely more important than offshore under-reporting,

quantitatively. Under the benchmark assumptions, total pass-through under-reporting is more

than twice the size of offshore under-reporting: $60 billion versus $130 in under-reported income

for tax year 2007 (in $2012). Relatedly, if we disregard offshore evasion – e.g. to impute evasion

in more recent years assuming that offshore evasion was nullifed by the crackdown – we would

still fnd signifcant amounts of undetected under-reporting in the top 1%. Under our low-end

assumptions with zero offshore evasion (i.e. if pass-through evasion based on the S corporation

random audit data is added to NRP data), 4.3% of true income in the top 1% is evaded, compared

to 2.3% based on random audit data alone.

Despite these robust patterns, differences in assumptions across imputations generate signifcant variation in the results. We unpack these differences further in Appendix Figures A9 and A10,

in which we begin with the benchmark and implement alternative assumptions one-by-one. For

offshore evasion, most of the difference across specifcations is attributable to our widely varying

assumptions about the total amount of concealed offshore income, due to both the total amount

concealed and the rate of return. Our assumption about how concealed income is distributed

mainly matters for what portion of this under-reporting belongs to the top 0.01% versus the rest

of the top 1%. For pass-through evasion, in contrast, the main difference between the low-end

and benchmark imputations is due to varying distributional assumptions. We are reasonably confdent that the overall amount of undetected pass-through misreporting is relatively large based

on the S corporation random audit data, but we are more uncertain about the distribution of misreporting because we are unsure how much the limitations of these data matter at the very top

of the distribution. In the benchmark, the rate of under-reporting falls within the top 0.1% of the

distribution, because most of the under-reporting we impute is business income and pass-through

income in the top 0.1% of the income distribution is mainly investment income (Figure 4c). This

dropoff is more pronounced in the low-end scenario than in the benchmark, and it occurs within

the top 0.5% rather than the top 0.1%. Under-reporting in the p95-p99 part of the distribution is

correspondingly higher in the low-end scenario. This occurs because in S corporation audit data,

under-reported business income is less concentrated than reported business income in the top 1%,

and more concentrated in p95-99 (see Figure 5a).50 The main difference between the benchmark

and high-end scenario is mechanically driven by the higher rate of under-reporting assumed in

the high-end scenario (we assume an identical distribution in both imputations).

The inclusion of some under-reported pass-through investment income in the benchmark and not the low-end

imputation also contributes to this difference in the profle of under-reporting rates but this is much less important

quantitatively than the assumption about the distribution of under-reported business income.

5.3

Incorporating Evasion Identifed by Detection-Controlled Estimation

It has long been recognized that audit data do not cover all forms of evasion, and not only because

of undetected evasion in the top 1%. Offcial tax gap statistics account for many types of undetected evasion using detection-controlled estimation (DCE) methods (IRS, 2019). We next consider

how introducing evasion identifed by DCE modifes the distribution of under-reported income

relative to our previous results. Doing so allows us to account for undetected under-reporting besides the high-end types of undetected under-reporting we focused on above, to reconcile our estimates based on audit data with offcial Tax Gap statistics, and to confront the question of whether

the offshore and pass-through evasion we imputed above is plausibly already included in DCEinclusive tax gap estimates.

Overview of DCE. The basic idea of DCE is to identify undetected evasion by modeling the

probability of detection of under-reporting, and then to construct a counterfactual scenario where

this probability equals one everywhere. Methods currently used in IRS Tax Gap studies implement

this idea using a model of auditor effects. The key underlying assumptions behind the method are

1) that auditor assignment rules generate exogenous variation in the probability of detection of

under-reporting, i.e., auditors are assigned quasi-randomly,51 and 2) there is a subset of auditors

with 100% detection rates. Building on these assumptions, researchers estimate a model of the

detection process to compute total evasion in a counterfactual where all auditors are replaced by

those auditors with perfect detection rates (for details, see Feinstein, 1991; Erard and Feinstein,

2011).

The evidence in Sections 3 and 4 suggests that the assumption that the top auditors have 100%

detection rates is too strong, at least for the top of the distribution. Figures 3a/3b and 4a show

that the detection of offshore evasion and entity-level evasion in pass-through businesses is so rare

that these forms of evasion are unlikely to be detected even by thorough auditors. We attribute

this fnding to limitations of auditors’ information and audit procedures. Specifcally, detection

of offshore noncompliance is so vanishingly rare that it cannot possibly be captured by auditor

effects, and this is plausibly driven by the auditors’ limited information. Relatedly, we note that

one of the most effective tools employed by the IRS during the recent crackdown on offshore

evasion was the “John Doe Summons” (see IRM 25.5.7); issuing such a summons requires the IRS

to attest to the Department of Justice that the noncompliance at issue is widespread and cannot be

detected by the IRS without a court order. Detection of entity-level pass-through under-reporting is

slightly less rare than offshore evasion – we see detected entity-level under-reporting for about 4%

of pass-through owners in Figure 4a. Our understanding, from talking to experts, is that variation

in whether an entity-level audit occurs is primarily driven by audit procedures and the nature

A concern with this assumption is that auditors may be assigned to cases in a manner that selects on suspected

evasion. We discuss this in the fnal section of Guyton et al. (2023).

of the business owned by the audited individual, not by auditor effects.52 In this case, the most

effective auditors do not always comprehensively examine pass-through entities, so entity-level

under-reporting cannot be fully captured by DCE.

Nevertheless, if we relax the 100% detection assumption to account for the possibility that

some under-reporting is undetectable by any auditor, DCE methods would still identify some

undetected under-reporting. We can think of the forms of evasion identifed by DCE methods as

detectable but undetected evasion, which could include many forms of evasion besides the two forms

of evasion we considered above, which are mostly undetectable.53

The primary challenge in incorporating evasion implied by DCE into our analysis is that existing methods are primarily used to identify a total amount of undetected evasion, rather than

evasion throughout the income distribution, especially the very top. Distributing the undetected

evasion identifed by DCE requires making additional conceptual and statistical assumptions on

the relationship between detection probabilities and true income. Because the estimates of undetected evasion are large, this issue creates signifcant uncertainty. In Guyton et al. (2023), which

is included in the Online Appendix, we discuss this methodological challenge in depth, and we

present results employing an array of different methods for distributional analysis building on

DCE. These methods include the methods employed by Johns and Slemrod (2010) and the different methods used in IRS (2007) and IRS (2019) to map total under-reporting to the tax gap (which

requires some distributional assumptions due to the non-linearity of the tax schedule).

Methodology to Incorporate DCE. We begin by estimating aggregate income under-reporting

according to the DCE model, using the methods from the most recent tax gap study, IRS (2019).

These estimates pertain to tax years 2008–2013, because IRS (2019) used random audit data for

this period to implement DCE. We adapt these estimates to our full sample period of 2006–2013,

assuming that rates of under-reporting for each type of income are similar for tax years 2006–2013

and 2008–2013.

We then make a distributional assumption about the location of DCE-identifed evasion in the

income distribution. For transparency, our preferred approach is to distribute DCE adjustments

similarly to how we distributed offshore and pass-through evasion above: to assume some share of

under-reporting belongs to different parts of the distribution and then use empirical data to discipline these shares. We label this approach Macro-Allocated DCE (MA-DCE) to distinguish it from

microsimulation methods in which the distribution of DCE adjustments emerges from allocating

DCE evasion to specifc taxpayers and then simulating re-ranking by DCE-adjusted income, as we

discuss further in Guyton et al. (2023).

See footnote 26 for more details.

“Detectable” here refers to whether detecting some type of non-compliance is feasible given audit procedures and

information and resource constraints. The boundary between “detectable” and “undetectable” depends on the design

of the audit program and related features of the overall tax system.

As above, we present a benchmark, low-end, and high-end imputation based on distributional

assumptions. In the benchmark, we allocate undetected under-reporting proportionally to examcorrected income, income component by income component. For example, if X% of interest income

belongs to the top 1% of the exam-corrected income distribution according to NRP random audit

estimates from Section 2.2, we assume X% of DCE-identifed undetected under-reported interest

income belongs to the top 1% of the true income distribution. With this allocation, X% of interest

income belongs to the top 1% both before and after we add DCE adjustments to exam-corrected

estimates, so this method is approximately distributionally neutral for each type of income by

design. Adopting this distributionally-neutral-by-component assumption is essentially equivalent

to assuming that for each component of income, the share of true income that is DCE-identifed

under-reporting is constant throughout the income distribution.54 For example, our approach is

approximately equivalent to assuming that if Y% of total true interest income is DCE-identifed

under-reporting in the full population, then in each part of the income distribution, Y% of interest

income is DCE-identifed under-reporting.

For a low-end assumption on the distribution of DCE-identifed under-reporting, we assume

that the share of detected and undetected evasion belonging to different parts of the distribution

is the same. In other words, if X% of detected under-reported interest income belongs to the top

1% by audit-corrected income in the NRP data, we assume X% of DCE-identifed interest income

belongs to the top 1% by DCE-corrected income. This is much more conservative with respect to

the concentration of DCE at the top of the distribution, because audit-detected under-reporting is

much less concentrated at the top than reported income. This specifcation yields a lower bound

for the distribution of DCE adjustments under the assumption that the probability of detection

of misreporting is weakly decreasing with true income. For our high-end assumption around

DCE, we assume distributional neutrality component-by component as in the benchmark, but we

combine partnership business income, S corporation business income, and sole proprietorship

income into a single component of income.

Aggregate Estimates. We replicated the analysis of individual under-reporting that generates

the estimate of the individual fler tax gap in the most recent IRS Tax Gap study (IRS, 2019).55

With DCE (and without imputed offshore or pass-through evasion), we estimate that 10.7% of true

income is under-reported, of which 3.8% was detected in exams and 6.9% is undetected evasion

identifed using DCE methods.56

As with pass-through under-reporting, when a given income component is negative in the aggregate in some part

of the income distribution, we allocate zero DCE-identifed under-reporting to this part of the distribution. That is why

we describe this approach as “approximately” distributionally neutral by type of income. Refer to Iselin and Reck (2025)

for formal analysis and more discussion of the negative incomes issue.

We thank Drew Johns for sharing the relevant simulations and for helping us to understand the underlying methods.

In 2012 dollars, $975 billion of income was under-reported and $300 billion dollars of individual income tax (including self-employment taxes and refundable credits) was unpaid on average over 2008–2013. The latter fgure closely

matches the offcial tax gap fgure for 2011-2013, $290 billions (IRS, 2019, Table 2 p. 11). The individual income underre-

Appendix Table A7 and Figure A11 describe the aggregate DCE adjustments by type of income.

About 80% of the aggregate DCE adjustment comes from forms of income that are moderately

or highly concentrated at the top of the distribution: Schedule C income, capital gains, rental

income, pass-through business income, and line 21 other income (e.g., net operating loss carryforwards).57 Less than 10% of the aggregate adjustment comes from wage and pension income,

which are relatively more equally distributed.58 As a result, we observe, comparing Table A1 to

Table A7, that DCE adjustments tilt the composition of under-reported and true income toward

forms of income that are concentrated at the top of the distribution.

Double Counting. As above, there is potential overlap between DCE-identifed evasion and imputed entity-level passthrough under-reporting. We assume the share of DCE-identifed passthrough under-reporting that is due to entity-level (as opposed to individual-level) corrections is

the same as the share of audit-detected pass-through under-reporting that is due to entity-level corrections (57.6%).59 When we include both DCE-identifed evasion and imputed pass-through evasion, we therefore remove this amount from DCE-identifed evasion, to mitigate double-counting.

Distributional Estimates. Going forward, we use the shorthand sophisticated evasion to refer to

offshore and pass-through evasion together and distinguish these types of undetected evasion

from DCE-identifed evasion. In panel a of Figure 7, we compare rates of under-reporting for the

three sets of assumptions about sophisticated evasion from the previous subsection and our benchmark assumption about DCE-identifed under-reporting. In panel b, we compare the benchmark,

high-end, and low-end imputations of DCE adjustments, under our benchmark assumptions about

sophisticated evasion. Appendix Table A8 reports the imputed distribution of unreported income

including DCE under the same sets of assumptions about DCE-identifed and sophisticated evasion.

DCE does not plausibly account for most undetected offshore and pass-through evasion, so

our imputations continue to suggest that incorporating offshore and pass-through evasion makes

a substantial difference to the profle of under-reporting, across all specifcations. With DCE but

without offshore and pass-through evasion, the share of true income that is under-reported by

those the top 1% is 12.4% in the top 1% under the benchmark DCE assumption. If we add offshore

and pass-through evasion, the top 1% misreporting rate increase to 17.1%, 14.1%, and 21.8% under

the benchmark, low- and high-end assumptions for sophisticated evasion.

porting tax gap is $245 billion and the self-employment under-reporting tax gap an additional $45 billion.

In 2012, the top 1% of tax flers earned 22.7% of reported market income, 19% of schedule C income, 83% of capital

gains, 78% of pass-through business income, and more than 100% of rental income (which on aggregate was negative).

In 2012, the top 1% of tax flers earned 12% of wage income and 6% of taxable pension income excluding Social

Security.

Some assumption along these lines is necessary because the portion of DCE-identifed evasion that is due to entitylevel evasion is not identifed in the DCE model. Our assumption is equivalent to assuming that auditor effectiveness is

orthogonal to whether an entity-level pickup occurs, which is consistent with what we know about the data-generating

process (see also footnotes 26 and 49).

However, we do observe large differences in the profle of under-reporting across the income

distribution depending on what distributional assumptions we make about DCE. DCE-inclusive

estimates of the share of income under-reported in the top 1% are especially sensitive to distributional assumptions over DCE-identifed evasion. Imputed rates of under-reporting differ by

between 2 and 5 percentage points in the bottom 99% of the distribution depending on what distributional assumption we make about DCE. At the top of the distribution these differences grow

large: the rate of under-reporting varies by about 10 percentage points in the top 0.1% in Figure 7b. Mechanically, this sensitivity of misreporting rates at the top to assumptions about DCE is

driven by the fact that reported nonwage income is much more concentrated at the top than auditdetected under-reported income, so which of these is used to discipline the substantial non-wage

under-reporting identifed by DCE matters a lot. We hope future research will develop principled

methods for refning these distributional assumptions.60 Relatedly, we observe that the effect of

accounting for offshore and pass-through evasion on the share of unreported income that belongs

to the top 1% depends on the extent to which these shares also increase when we allocate DCEidentifed under-reporting.61

5.4

Evasion and the Distribution of Overall Income

Our fnding that important forms of evasion at the top are unobserved in random audit data suggests that prior estimates of income and its distribution based on random audit data may be biased.

For the total income, the bias from missing offshore and pass-through evasion is relatively small:

in our benchmark imputation, adding offshore and pass-through evasion increases total income

by a factor of 1.016,62 Because the missing income is highly concentrated in the top 1%, however,

the bias, and the uncertainty around DCE, are much more important when we focus on the income

of the top 1%.

Appendix Tables A9 through A11 present imputations of the concentration of fscal (taxable)

market income including under-reporting under our assumptions about under-reported income,

for the full range of assumptions we considered in the previous subsection. Without underreporting, 20.3% of income belongs to the top 1% of the reported income distribution on average

over 2006–2013. Detected under-reporting in the random audit data is signifcantly less concen60

The distributions of DCE-identifed evasion that result from prior microsimulation methods – the multiplier method

from IRS (2007) and Johns and Slemrod (2010), the slight modifcation of this used by Auten and Splinter (2021), and

a newer method used in IRS (2019) – rely on ancillary assumptions that are not empirically disciplined, so they do not

fll the gap in our understanding we identify here. We discuss this issue further and illustrate rates of under-reporting

using these microsimulation methods in Guyton et al. (2023).

This is mechanical: when DCE-identifed evasion is relatively concentrated, adding additional concentrated underreporting makes relatively little difference to the overall concentration. Consequently, in every imputation in which we

include both DCE and offshore and pass-through under-reporting, unreported income shares are signifcantly higher

than the 11.4% share suggested by NRP exam corrections alone.

This fgure is the ratio of 1) fscal market income income including audit-detected and DCE-identifed underreporting and under-reported offshore and pass-through income (with the double counting adjustment), and 2) fscal

market income including audit-detected and DCE-identifed under-reporting.

F IGURE 7: I MPUTATION R ESULTS WITH DCE: U NDER -R EPORTED I NCOME AS A S HARE OF

T RUE I NCOME

(a) Benchmark DCE + Various Sophisticated (b) Benchmark Sophisticated + Various DCE

20.0%

15.0%

10.0%

25.0%

Under-reported income, as % of true income

(reported + unreported)

Under-reported income, as % of true income

(reported + unreported)

25.0%

+ Sophisticated (high)

+ Sophisticated (benchmark)

+ Sophisticated (low)

NRP

+ DCE (benchmark)

5.0%

NRP (No DCE)

0.0%

20.0%

+ DCE (high)

15.0%

+ DCE (benchmark)

+ DCE (low)

10.0%

5.0%

0.0%

Position in the income distribution (exam-corrected income)

NRP

+ sophisticated (benchmark)

NRP (No DCE)

Position in the income distribution (exam-corrected income)

Note: This fgure reports the imputed profle of rates of under-reporting in different parts of the income distribution,

including evasion identifed by DCE. As in the previous fgure, the black line is the unadjusted NRP estimate from

Figure 1a. In panel a, we add the imputed distribution of DCE-identifed evasion according the benchmark assumptions

and then we add sophisticated evasion under benchmark, high- and low-end assumptions. In panel b, we do the reverse

exercise, making the benchmark assumptions about sophisticated evasion and then adding DCE-identifed evasion

under various distributional assumptions. Refer to Guyton et al. (2023) for more detail on DCE allocation methods

and additional illustrations of the distribution of DCE-identifed evasion using microsimulation methods. We observe

that incorporating sophisticated evasion alters the profle of under-reporting regardless of how we treat DCE, but our

assumptions imply substantial quantitative uncertainty about how DCE-identifed evasion contributes to the profle of

under-reporting.

trated than reported income: Table 2 reports that 11.4% of unreported income belongs to the top

1% ranking by corrected income. As such, if only audit-detected under-reporting is included in

under-reported income, the top 1% income share falls, to 19.8%. Given what we found in Table

2, it is mechanical that adding offshore and pass-through evasion will increase the top 1% income

share. Adding these types of evasion increases the top 1% income share by 1.1 percentage points

under the benchmark assumptions without DCE, 0.3 pp under the low-end assumptions and 2.1

pp under the high-end assumptions. However, the level of the top 1% income share inclusive of

under-reporting is uncertain, especially when we include DCE, because of the uncertainty around

the distribution of DCE-identifed evasion discussed above. The imputed top 1% income share

varies from 20.1% to 21.9% excluding DCE-identifed under-reporting. Including DCE-identifed

under-reporting and imputed offshore and pass-through under-reporting, across all the specifcations illustrated in Figure 7, the top 1% income share ranges from 20.3% to 22.4% (21.3% under the

benchmark assumptions for every component of undetected evasion).

These fndings have important implications for prior work seeking to measure the distribution

of National Income over time. Untaxed business income, which includes under-reported passthrough and sole proprietor income, is a key source of uncertainty in prior work. Piketty et al.

(2018a) assume that sharp increases in the concentration of reported business income were accompanied by similar increases in the concentration of unreported business income, while Auten and

Splinter (2024a) assume (via a microsimulation) that the share of unreported income that belongs

to the top 1% is constant over time at about 18%.63 Auten and Splinter (2024a) argue that this

feature of their estimates is empirically validated by their use of audit data to calibrate their microsimulation. Our imputations of unreported income shares (including DCE) in Table A8 suggest

that between 21.9% and 39.4% of unreported income belongs to the top 1% in the period we study,

which is larger than the 18% fgure in Auten and Splitner’s microsimulations, plausibly because

the audit data and DCE methods they use do not account for offshore and pass-through evasion.

Iselin and Reck (2025) discuss the implications of this issue for the dynamics of inequality in National Income.

6 Theoretical Implications

Appendix C examines some theoretical implications of our empirical results. In the frst part of this

appendix, we propose an extension of the Allingham and Sandmo (1972) to explain why forms of

evasion that are less likely to be detected by auditors are concentrated at the top of the income

distribution. In this model, the taxpayer can choose to pay a fxed cost to adopt a concealment

This fgure is based on calculations Auten and Splinter added to their online appendix in response to Iselin and

Reck (2025). Besides evaded income, the other main source of untaxed business income driving the divergence between

these studies is excess depreciation allowances, see Piketty et al. (2024), which also reviews some more minor sources

of disagreement, and Auten and Splinter (2024b). Section V of Auten and Splinter (2024a) contains a brief discussion

of the effect of accounting for offshore tax evasion on the distribution of income, which is broadly consistent with our

benchmark imputations.

strategy that reduces their probability of detection. The fxed costs becomes a trivial share of

income as income grows large, so, assuming the taxpayer’s desired amount of evasion does not

become a trivial share of income as income grows – a suffcient condition for which is that the

taxpayers preferences exhibit constant or increasing relative risk aversion – the benefts of adopting

concealment strategies outweigh the costs at suffciently high incomes. Higher audit rates at the

top also incentivize concealment strategies, provided the audits are the type that do not detect

more sophisticated evasion.

The second part of Appendix C draws out some straightforward implications of our results

for optimal tax administration. The model considers the allocation of administrative resources

toward audits of two types of individuals, high- and low-income. One key difference between

audit allocation decisions in reality and in models of optimal tax administration (e.g. Slemrod

and Yitzhaki, 2002; Keen and Slemrod, 2017) is that in reality, IRS must allocate a fxed set of

resources are determined by Congress among various activities. Devoting more resources to highincome individuals requires devoting fewer resources to other activities. In optimal tax models, no

such tradeoff exists. Turning to sophistication, we argue that the adoption of sophisticated evasion

strategies increases the cost of recovering revenue from high-income taxpayers. With an exogenous

resource constraint, this decreases the optimal number of audits of high-income individuals, and

if this effect is suffciently strong then this can increase audits of low-income individuals. No such

spillover effect occurs in optimal tax models (or equivalently when the resource constraint is set

optimally), which we view as interesting given recent debates about the allocation of resources to

various types of audits. Naturally, the preceding discussion holds the detection technology fxed;

the presence of sophisticated evasion could increase or decrease the return to investing in specifc

detection technologies.

We observe that these theoretical arguments apply equally well to strategies that make explicitly illegal positions harder to detect and gray area avoidance strategies that make it less likely

that the IRS will contest and penalize a position. Figure A12, based on operational data, shows

that very high-income taxpayers are much more likely to disagree with initial auditor recommendations, especially large recommendations. The share of dollars of recommended additional tax

that are disputed increases sharply through the income distribution, from 10-20% in the bottom

90% to more than 50% in the top 0.01%. Such gray areas have two important implications, which

are important questions for future work. First, while we sought to only include non-compliant

positions in our estimates and imputations, gray areas generate quantitative uncertainty about the

magnitude of the tax gap at the top, and about the distribution of economic income (which includes evasion and some legal avoidance). Second, it is limiting to equate tax enforcement with

increasing audit rates in the presence of frequent tax disputes: enforcement can also entail investing in litigating tax disputes and reforming rules to make gray area positions more explicitly illegal

(for more discussion, see Hemel et al., 2021; Reck and Bomare, 2022).

Conclusion

We fnd that substantial evasion at the top of the income distribution was undetected in individual

random audits from 2006–2013. Investigating taxpayers who voluntarily declared hidden wealth

or started reporting foreign bank accounts in 2009–2012, we fnd that audits failed to uncover offshore tax evasion. Focusing on taxpayers who earn business income through partnerships and S

corporations, we fnd that due to the resource constraints inherent to the conduct of random audits, a large fraction of this pass-through income was not examined in the context of these audits,

biasing detected evasion downward at the top. Meanwhile, random audit data on S corporations

suggests that under-reporting of business income on pass-through business tax returns is substantial, fve times larger than what is suggested by individual random audit data.

Theoretically, modeling the choice to conceal tax evasion from auditors can explain why random audits that employ the same procedures for all taxpayers do not detect all evasion especially

at the top. Empirically, we provide corrected estimates of the size and distribution of tax evasion

in the United States. In our benchmark scenario, we fnd that under-reported income rises from

about 10% of true income in the bottom 90% of the income distribution to 16% at the 99th percentile

of income, where it remains constant or falls. Out of the overall top 1% under-reporting rate, at

least 6 percentage points corresponds to sophisticated evasion that is seldom detected in random

audits. Accounting for tax evasion increases the top 1% income share in the United States.

We stress that the limitations that we describe here pertain to NRP random audits circa 20062013; whether changes to audit procedures might overcome these limitations is an important question for future research. Unlike IRS audits conducted by specialists, as in the Global High Wealth

program (see IRM 4.52.1), these random audit programs were not designed to comprehensively

audit extremely high-income, high-wealth individuals. To experts who are familiar with these audits, our results may be unsurprising. However, we nevertheless view these results as important

in light of an increased academic and policy interest in top income shares and tax evasion at the

top. Countries around the world use random audits to estimate the tax gap (see, e.g., OECD, 2017,

chapter 14). Our fndings suggest that incorporating data from random audits into estimates of

top-end evasion and inequality requires carefully accounting for the sophistication of top-end evasion. In this regard, statistical tools (ex. modeling the distribution of DCE-identifed undetected

evasion) and additional data (ex. data on offshore and pass-through level non-compliance or from

more comprehensive audits of high-income taxpayers) may generate quantitatively important insights.

From public reporting and evidence reviewed above, it seems likely that there are other specifc

forms of tax evasion that have the same properties as those we examine directly in this paper– sophistication and concentration among high income or high wealth individuals. Such forms of evasion could include, for instance, the abuse of partnership fexibility (see also Love, 2024), private

inurement in tax-exempt organizations, and the use of offshore trusts to evade tax. The potential

existence of more sophisticated strategies than we studied above underscores the main point of

our theoretical results, that we should expect sophisticated evasion to be concentrated at the top

of the income and wealth distributions. More research is needed to improve estimates of noncompliance at the very top in the United States. Further research is also needed to better understand

potentially less concentrated types of diffcult-to-detect evasion, such as payments under the table

and cryptocurrency transactions, and their importance for measuring the income distribution and

policymaking.

We identify several potentially fruitful avenues for future work. First, it would be valuable to

consider the importance of sophisticated evasion and gray area avoidance strategies for optimal

tax administration policies involving high-income, high-wealth taxpayers. Second, more research

is needed to fully understand the economics of the gray area between avoidance and evasion,

which is often important for high-income individuals and for large corporations. Third, further research could consider strategic interactions between the tax authority and high-income individuals

involving sophisticated evasion. Finally, future work could consider the implications of our work

for white-collar, fnancial crime more broadly.

References

Allingham, M. G. and Sandmo, A. (1972). Income tax evasion: A theoretical analysis. Journal of

Public Economics, 1(3-4):323–338.

Alstadaeter, A., Johannesen, N., and Zucman, G. (2018). Who owns the wealth in tax havens?

Macro evidence and implications for global inequality. Journal of Public Economics, 162:89 – 100.

Alstadsaeter, A., Johannesen, N., and Zucman, G. (2019). Tax Evasion and Inequality. American

Economic Review, 109(6):2073–2103.

Andreoni, J., Erard, B., and Feinstein, J. (1998). Tax Compliance. Journal of Economic Literature,

36(2):818–860.

Auten, G. and Splinter, D. (2021). Comment: Tax evasion at the top of the income distribution: Theory and evidence. Comment, available online at http://www.davidsplinter.com/

AutenSplinter-TaxEvasion.pdf.

Auten, G. and Splinter, D. (2024a). Income inequality in the united states: Using tax data to measure long-term trends. Technical Report 7.

Auten, G. and Splinter, D. (2024b). Reply to piketty, saez, and zucman (2024): Income inequality

in the united states. Unpublished comment, https://www.davidsplinter.com/.

Blumenthal, M., Christian, C., and Slemrod, J. (1998). The determinants of income tax compliance:

Evidence from a controlled experiment in minnesota. NBER working paper no. 6575.

Brown, R. E. and Mazur, M. J. (2003). IRS’s comprehensive approach to compliance measurement.

National Tax Journal, 56(3).

Carrillo, P., Pomeranz, D., and Singhal, M. (2017). Dodging the taxman: Firm misreporting and

limits to tax enforcement. American Economic Journal: Applied Economics, 9(2):144–64.

Casi, E., Spengel, C., and Stage, B. M. (2020). Cross-border tax evasion after the common reporting

standard: Game over? Journal of Public Economics, 190:104240.

Cooper, M., McClelland, J., Pearce, J., Prisinzano, R., Sullivan, J., Yagan, D., Zidar, O., and Zwick,

E. (2016). Business in the United States: Who Owns it, and How Much Tax Do They Pay? Tax

Policy and the Economy, 30(1):91–128.

De Simone, L., Lester, R., and Markle, K. (2020). Transparency and tax evasion: Evidence from the

foreign account tax compliance act (fatca). Journal of Accounting Research, 58(1):105–153.

DeBacker, J., Heim, B., Tran, A., and Yuskavage, A. (2020). Tax noncompliance and measures of

income inequality. Tax Notes, February 17, 2020.

DOJ (2021). Offshore compliance initiative news indictments, pleas, sentencings, and other developments. Archive, accessed june 2021, available online at https://www.justice.gov/tax/

offshore-compliance-initiative.

Erard, B. and Feinstein, J. (2011). The individual income reporting gap: What we see and what we

don’t. 2011 irs-tpc research conference, publication 1500 (rev 4-2012), Washington, DC.

Feinstein, J. S. (1991). An Econometric Analysis of Income Tax Evasion and Its Detection. The

RAND Journal of Economics, 22(1):14–35.

GAO (1995). Tax administration: IRS’ partnership compliance activities could be improved. Letter

report, 06/16/95, gao/ggd-95-151.

GAO (2020). Abusie tax schemes: Offshore insurance products and associated compliance risks.

Letter report, july 2020, gao-20-589.

GAO (2023). Irs audit processes can be strengthened to address a growing number of large, complex partnerships. Report to the chairman, committee on fnance, u.s. senate, gao-23-106020.

Guttentag, J. and Avi-Yonah, R. S. (2005). Closing the international tax gap. In Sawicky, M. B.,

editor, Bridging the Tax Gap: Addressing the Crisis in Federal Tax Administration, pages 99–110.

Economic Policy Institute.

Guyton, J., Langetieg, P., Reck, D., Risch, M., and Zucman, G. (2023). The distribution of undetected under-reported income identifed by detection-controlled estimation methods. Unpublished working paper, included in the online appendix to this paper.

Hemel, D. J., Holtzblatt, J., and Rosenthal, S. (2021). The tax gap’s many shades of gray. Working

paper, SSRN.

Hess, R., Black, E., Javed, Z., Hennessy, J., Lester, R., Goldin, J., Ho, D. E., and Portz, A. (2024). The

spiderweb of partnership tax structures. Working paper, ssrn.

Investigations, I. C. (2025). Annual report 2025. Publication 3583.

IRS (2007).

Tax year 2001 federal tax gap overview.

https://www.irs.gov/pub/irs-

soi/01rastg07map.pdf.

IRS (2016a). Federal Tax Compliance Research: Tax Gap Estimates for Tax Years 2008-2010. Publication 1415 (rev. 5-2016), Washington, DC.

IRS (2016b). Transaction of interest – section 831(b) micro-captive transactions. Internal Revenue

Bulletin 2016-46, notice 2016-66.

IRS (2019). Federal Tax Compliance Research: Tax Gap Estimates for Tax Years 2011-2013. Publication 1415 (rev. 9-2019), Washington, DC.

IRS (2020). Internal revenue service data book 2019. Publication 55-b (rev. 06-2020), Washington,

DC.

IRS (2022). Internal revenue service data book 2021. Publication 55-b (rev. 2021), Washington, DC.

IRS (2024). New irs teams being established; new guidance designed to stop partnerships from

using sophisticated tax-free transactions that lack economic substance. Notice ir-2024-166.

Iselin, J. and Reck, D. (2025). Misreported income and the dynamics of income inequality. Working

paper, https://www.danreck.com/papers.

Johannesen, N., Langetieg, P., Reck, D., Risch, M., and Slemrod, J. (2020). Taxing hidden wealth:

The consequences of us enforcement initiatives on evasive foreign accounts. American Economic

Journal: Economic Policy, 12(3):312–46.

Johannesen, N., Reck, D., Risch, M., Slemrod, J., Guyton, J., and Langetieg, P. (2024). The offshore

world according to fatca: New evidence on the foreign wealth of us households. Tax Policy and

the Economy, 38(1):61–99.

Johns, A. and Slemrod, J. (2010). The Distribution of Income Tax Noncompliance. National Tax

Journal, 63(3):397–418.

Johns, D. (2009). Preliminary results of the 2003/2004 national research program s corporation

underreporting study. Presentation, 2009 IRS-tpc research conference.

Joulfaian, D. (2000). Corporate income tax evasion and managerial preferences. Review of Economics

and Statistics, 82(4):698–701.

Keen, M. and Slemrod, J. (2017). Optimal tax administration. Journal of Public Economics, 152:133–

142.

Kleven, H. J., Knudsen, M. B., Kreiner, C. T., Pedersen, S., and Saez, E. (2011). Unwilling or Unable

to Cheat? Evidence From a Tax Audit Experiment in Denmark. Econometrica, 79(3):651–692.

Kleven, H. J., Kreiner, C. T., and Saez, E. (2016). Why can modern governments tax so much? an

agency model of frms as fscal intermediaries. Economica, 83(330):219–246.

Londoño-Vélez, J. and Ávila-Mahecha, J. (2021). Enforcing wealth taxes in the developing world:

Quasi-experimental evidence from colombia. American Economic Review: Insights, 3(2):131–48.

Love, M. (2024). Who benefts from partnership fexibility?

Technical report, Working Paper,

SSRN.

Luttmer, E. F. and Singhal, M. (2014). Tax morale. Journal of Economic Perspectives, 28(4):149–68.

OECD (2000). Towards global tax co-operation. Report to the 2000 ministerial council meeting and

recommendations by the committee on fscal affairs.

OECD (2017). Tax Administration 2017.

Piketty, T. and Saez, E. (2001). Income inequality in the united states, 1913-1998 (series updated to

2000 available). National bureau of economic research working paper no. 8467.

Piketty, T. and Saez, E. (2003). Income inequality in the united states, 1913–1998. Quarterly Journal

of Economics, 118(1):1–41.

Piketty, T., Saez, E., and Zucman, G. (2018a). Distributional national accounts: methods and estimates for the united states. Quarterly Journal of Economics, 133(2):553–609.

Piketty, T., Saez, E., and Zucman, G. (2018b). Distributional national accounts: methods and estimates for the united states. Quarterly Journal of Economics, 133(2):553–609.

Piketty, T., Saez, E., and Zucman, G. (2024). Income inequality in the united states: A comment.

Unpublished comment, https://gabriel-zucman.eu/.

Reck, D. and Bomare, J. (2022). Different from you and me: tax enforcement and sophisticated tax

evasion by the wealthy. LSE Public Policy Review, 2(4).

Saez, E. and Zucman, G. (2016). Wealth Inequality in the United States since 1913: Evidence from

Capitalized Income Tax Data. Quarterly Journal of Economics, 131(2):519–578.

Saez, E. and Zucman, G. (2020). Trends in US Income and Wealth Inequality: Revising after the

Revisionists. NBER working paper no. 27921.

Slemrod, J., Collins, B., Hoopes, J. L., Reck, D., and Sebastiani, M. (2017). Does credit-card information reporting improve small-business tax compliance? Journal of Public Economics, 149:1–19.

Slemrod, J. and Yitzhaki, S. (2002). Tax avoidance, evasion, and administration. In Handbook of

public economics, volume 3, pages 1423–1470. Elsevier.

Smith, M., Yagan, D., Zidar, O., and Zwick, E. (2019). Capitalists in the twenty-frst century. Quarterly Journal of Economics, 134(4):1675–1745.

Smith, M., Zidar, O., and Zwick, E. (2023). Top wealth in america: New estimates under heterogeneous returns. Quarterly Journal of Economics, 138(1):515–573.

United States Senate (2008). Tax Haven Banks and U.S. Tax compliance. Staff report of the permanent subcommittee on investigations, Washington, DC.

United States Senate (2014). Offshore Tax Evasion: The Effort to Collect Unpaid Taxes on Billions

in Hidden Offshore Accounts. Technical report, Washington, DC.

United States Senate (2020). Syndicated conservation-easement transactions. Investigative report

of the senate fnance committee, Washington, DC.

United States Senate (2022). The shell bank loophole. Investigative report of the senate fnance

committee, Washington, DC.

US Department of Treasury (2020). Letter from commissioner charles p. rettig to senate fnance

committee chair charles grassley, september 17, 2020. Open letter from the irs.

Yitzhaki, S. (1974). A note on ‘income tax evasion: A theoretical analysis’. Journal of Public Economics, 3(2):201–202.

Zucman, G. (2013). The Missing Wealth of Nations: Are Europe and the U.S. net Debtors or net

Creditors? Quarterly Journal of Economics, 128(3):1321–1364.

“Tax Evasion at the Top of the Income Distribution: Theory and Evidence”

Online Appendix

John Guyton, Patrick Langetieg, Daniel Reck,

Max Risch, and Gabriel Zucman

This appendix contains the following supplementary material:

• Appendix A contains additional empirical results referenced in the main text.

• Appendix B proves the results in Section 6 of the main text.

• Appendix C lays out a model in which we explore the implications of some of our results for

tax administration.

• The last part of this Appendix is a standalone paper on DCE methodology, which we cited

as Guyton et al. (2023) in the main text.

Additional Results

5.9

3.9

1.9

0.2

5.6

0.7

5.3

72.3

4.1

100.0

7.1

2.8

0.7

11.9

6.5

8.9

49.4

3.5

9.2

100.0

Total underreported

income of

this type/

Total

under-reported

income (%)

0.28

0.11

0.03

0.47

0.26

0.35

1.95

0.14

0.37

3.95

Total underreported

income of

this type/

Total

income (%)

4.8

2.8

1.5

242.3

4.6

48.1

36.7

0.2

8.8

Total underreported

income of

this type/

Total

income of

this type (%)

21.4

8.6

3.0

2.6

21.7

1.6

4.2

38.1

-1.1

100.0

Total

income of

this type/

Total

income (%)

18.8

3.3

2.0

8.5

18.8

5.4

34.9

2.9

5.3

100.0

Total underreported

income of

this type/

Total

under-reported

income (%)

0.43

0.09

0.05

0.19

0.43

0.12

0.79

0.07

0.10

2.27

Total underreported

income of

this type/

Total

income (%)

Top 1%

2.0

0.9

1.6

7.5

2.0

7.9

18.7

0.2

-11.4

Total underreported

income of

this type/

Total

income of

this type (%)

Note: This table describes the composition of detected under-reported income in the 2006–2013 NRP data, before any correction for undetected noncompliance (in

particular before any DCE correction). The NRP shares of each type of income in total income are similar to income shares we observe in Statistics of Income (SOI)

data, but the NRP shares are built using corrected income here. Consequently the largest differences with SOI income shares are observed for types of income with

signifcant detected evasion. Note that “Form 1040 Other Income” in Figure 1a is referred to as “Line 21 Other Income” here, as this item appears on Line 21 of the

Form 1040, while the residual “Other” category in the penultimate row refers to all other components of income. We note that the estimated rates of under-reporting

by type of income in the fourth column are well in excess of 100% for Line 21 income. This can occur because Line 21 income can be negative; large negative values

are common at the bottom of the income distribution because of net operating loss carryforwards or carrybacks from pass-through businesses. Large corrections to

line 21 are typically disallowed loss carryforwards or carrybacks.

Capital Gains

Dividends

Interest

Line 21 Other Income

Partnerships and S Corp

Rental

Schedule C

Wages

Other

Total

income of

this type/

Total

income (%)

Full Population

TABLE A1: I NCOME U NDER -R EPORTING D ETECTED IN NRP R ANDOM A UDITS : D ECOMPOSITION BY I NCOME T YPE , TAX Y EARS

2006–2013

TABLE A2: R ERANKING : D ETECTED U NDER -R EPORTING BY R EPORTED AND C ORRECTED

M ARKET I NCOME , AS A % OF T OTAL D ETECTED U NDER -R EPORTING , TY2006–2013

P0-10

P10-20

P20-30

P30-40

P40-50

P50-60

P60-70

P70-80

P80-90

P90-95

P95-99

P99-100

Col. Total

P0-10

P10-20

P20-30

P30-40

P40-50

P50-60

P60-70

P70-80

P80-90

P90-95

P95-99

P99-100

Row Total

4.9

-0.7

-1.0

-0.7

-0.2

1.4

0.6

-0.4

0.7

1.3

1.1

0.8

-0.2

-0.1

2.8

1.4

1.1

1.7

1.0

-0.2

-0.1

4.9

1.2

0.8

1.7

1.9

0.9

-0.1

0.0

6.2

1.3

0.7

1.0

1.4

2.1

1.3

-0.1

7.6

1.4

0.7

1.1

1.5

2.5

1.8

-0.2

-0.1

9.7

1.8

0.5

0.7

1.1

1.3

1.5

3.0

2.4

-0.3

11.8

2.0

0.4

0.8

1.1

1.4

2.1

3.5

3.6

-0.2

15.6

1.4

0.4

0.7

0.9

1.0

1.6

3.1

2.4

-0.1

12.2

1.5

0.3

0.8

0.9

1.4

2.0

2.8

5.0

-0.2

15.5

1.2

0.5

2.3

4.5

11.5

21.0

5.4

7.3

7.7

7.4

8.6

9.2

8.5

8.8

5.4

6.8

4.1

100.0

Note: This table reports the share of under-reporting in cells of both reported and correcte

This text is long and has been trimmed here. Open the source document for the complete record.

This is a copy of a public record, reproduced as it was published. It is not legal advice, and it may not be the version a court would rely on. Check the official source before you cite it.