«Eyfl‘wcvlfuaf . , , . ‘ ., . H ‘ ‘ . . ‘ .u . i .m...} . i

h‘er .1 .9: H . V
E... 3. .. u. v V
u L3... A
, , fl ... ._
V ‘ kit-.1”: . ‘
NFL: .

:22 a.
. if
L

. a.
.1124 $52». a .
.w fiqfiaamfiau
a .

.;.r,
x? ‘

4’1
:5

L7

a. £4. .
t... w : ”Ly

.3 W thY..

{H.- 35......

v-
.
u

a
1.3,. _ a

W...)

 

 

 

 

 

nflda

 

$

é... .. ($.11 .

«~310- ....:. A 5?.»
>5:1h€5s 11

 

 

't
a u a 't a.)
3?. lu.$.3.....

A“, .59 U. ~

 

 

‘ an». .. .

. .. 53.1. . . .

.r.._u.'.q....,=a,.. . . . 2 . 2.3....
.1‘

., , y _ . ‘ . ‘ .42}?

 

.H .t... u .74} .. . , . , a . ~33...in . . . . , . . . .A .. y , V. .:w ,..‘.
dififiwr. 5&5: . .. _ _, H . .. . . ggfififi. V . . . . .. . ., .. ”an?”

,( ..

"News

" (\D
\‘n ‘

This is to certify that the

dissertation entitled

THE MARGINAL COST OF PUBLIC FUNDS
IN THE PRESENCE OF TARIFF EVASION

presented by

YOON SANG KIM

has been accepted towards fulfillment
of the requirements for

Ph. D . degree in Economics

mxwa

Major professor

 

Date MW 200/

MS U i: an Affirmative Action/Equal Opportunity Institution 0-12771

 

 

LIBRARY

Michigan State
University

 

 

 

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

‘ifitfil'fl’ZW

 

“A? ‘2 04,9 20b}?

 

03:11:?de £2003

 

MM 3* 2805’

 

 

 

 

 

 

 

 

 

 

 

6/01 chlRC/DateDue,p65-p.15

THE MARGINAL COST OF PUBLIC FUNDS IN THE PRESENCE
OF TARIFF EVASION

By

Yoon Sang Kim

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Economics

2001

ABSTRACT

THE MARGINAL COST OF PUBLIC FUNDS IN THE PRESENCE
OF TARIFF EVASION

By

Yoon Sang Kim

In this paper, I examine the major determinants of tariff evasion and calculate the
marginal cost of public funds (MCF) associated with various policy tools, using micro
data for 1998 in Korea. I show that, if we incorporate the tariff evasion and concealing
costs incurred by firms in calculating MCFs, different values are measured than those
derived without considering evasion.

I find that tariff evasion increases with tariff rates in all commodity categories.
Estimates of the elasticity of tariff evasion with respect to a tariff rate are in the range of
0.102 and 1.026, depending on the commodity categories. This wide range of elasticities
is due to the differences in the slope of evasion in terms of the tariff rate and the ratio of
the tariff rate to evasion. However, the probability of detection has no significant effects
on evasion in any of the categories. Only for Consumer Goods does the probability of
audit have deterrent effects, with the elasticity of -0.416.

I also calculate MCFs of tariff rates without evasion, which vary from 1.1446 to
1.3064 for three commodity categories. Surprisingly, for all commodities, MCFs of tariff
rates in the presence of evasion are found to be smaller than those derived without
considering evasion, although the differences are usually small. Overall, an increase in

the probability of audit, an increase in the penalty rate, and an increase in the probability

of detection are found to be more efficient policy tools for raising additional revenues
than tariff rates, if these enforcement variables have deterrent effects on evasion (with
negative values of the partial effects of enforcement variables).

As novel results, the main contributions of this paper are an estimation of the major
determinants of tariff-evasion behavior and customs fraud and the provision of numerical

values of MCFs associated with different government policy tools.

Copyright by
Yoon Sang Kim
2001

To my family

ACKNOWLEDGMENTS

First of all, I would like to give my special thanks to my advisor, Professor
Charles Ballard, for his encouragement and insightful comments, from the initial idea to
the final draft. I would also like to thank Professor Lawrence Martin and Professor
Jeffrey Wooldridge for their guidance and helpful comments on the theoretical model and
econometric methodology.

Throughout my studies in the doctoral program, my fellow students in economics,
my fiiends in merchandising management, and my tennis partners helped me out with my
troubles and stress that otherwise would have been intolerable. In addition, my gratitude
goes out to all of the MAEBIT members in the US and in Korea. They are among the
few people I could talk to and discuss all of my problems with for more than the past ten
years.

I also owe a great debt of gratitude to Doctor Jungkee Lee, Doctor Kichul Lee,
Doctor Seongchul Kim, Kapseok Yoon, my fiiends, and all of my prayer-group members
at church, who anxiously took care of me and my wife and prayed for me to get well,
when I had my long-lasting medical problem. Without their encouragement and support,
I might not have been able to finish this dissertation.

In particular, I would like to express my special gratitude to my parents and
parents-in-law, who have made me what I am now. My father, a former director general
of Korea Customs Services, and my mother, a professor in Korean literature, have been
supporting and encouraging me to do whatever I want to do throughout my entire life.

My parents-in-law also have given me invaluable support during my doctoral studies.

vi

My greatest debt and gratitude is to my family: my wife, Ho Jung, and my
beloved son, Jung Ho. Ho Jung has dedicated all of her life to supporting and taking care
of me since we first met. Without her dedication, I would not have been able to go
through many difficulties. In addition, I am greatly indebted to my son, Jung Ho, who
has consistently asked me to play with him, but never complained even when I could not
play with him.

I dedicate this dissertation to all of those whom I love and to all of those who

have loved me, throughout my life.

vii

TABLE OF CONTENTS

LIST OF TABLES .............................................................................................................. x
CHAPTER I
INTRODUCTION .............................................................................................................. 1
CHAPTER II
THE MODEL ...................................................................................................................... 7
1. Theoretical Framework ................................................................................................ 7
2. Comparative Static Analysis ..................................................................................... 13
CHAPTER III
THE MARGINAL COST OF PUBLIC FUNDS WITH TARIFF EVASION ................. 17
1. Marginal Tariff Revenues in the Absence of Evasion ............................................... 17
2. Marginal Tariff Revenues in the Presence of Evasion .............................................. 21
3. Consumer Welfare Losses ......................................................................................... 24
4. The Marginal Cost of Public Funds of Major Policy Changes .................................. 26
4-1. The MCF of a Tariff Rate Increase in the Absence of Evasion .......................... 27
4-2. The MCF of Major Policy Changes in the Presence of Evasion ........................ 28
4-2-1. The MCF of a Tariff Rate Increase in the Presence of Evasion ................... 28
4-2-2. The MCF of a Penalty Rate Increase in the Presence of Evasion ................ 30
4-2-3. The MCF of an Enforcement Expenditure Increase in the Presence of
Evasion .................................................................................................................... 31
CHAPTER IV
THE MARGINAL COST OF PUBLIC FUNDS ASSUMING EXOGENOUS AUDIT
AND DETECTION .......................................................................................................... 37
1. Theoretical Framework for Tarifi Evasion ................................................................ 37
2. The Marginal Cost of Public Funds in the Presence of Tariff Evasion ..................... 41
2-1. Marginal Tariff Revenues ................................................................................... 41
2-2. Consumer Welfare Losses .................................................................................. 44
2-3. The Marginal Cost of Public Funds of Major Policy Changes ........................... 44
CHAPTER IV
THE EMPIRICAL ANALYSIS OF TARIFF EVASION ................................................ 47
1. Empirical Framework ................................................................................................ 47
2. Data Description ........................................................................................................ 52
3. Empirical Results ....................................................................................................... 62
3-1. Total Estimation .................................................................................................. 62
3-1-1. Estimation Results ........................................................................................ 62

viii

1-2. Specification Tests ....................................................................................... 63

3-
3-1-3. Implications .................................................................................................. 66
3-2. Estimation by Commodity Categories ................................................................ 73
3-2-1. Estimation Results ........................................................................................ 73
3-2-2. Specification Tests ....................................................................................... 74
3-2-3. Implications .................................................................................................. 75
CHAPTER V
THE EMPIRICAL ANALYSIS OF CUSTOMS FRAUD ............................................. 102
1. Empirical Framework .............................................................................................. 102
2. Data Description ...................................................................................................... 104
3. Empirical Results ..................................................................................................... 105
3-1. Total Estimation ................................................................................................ 105
3-1-1. Estimation Results ...................................................................................... 105
3-1-2. Tests for Homoskedasticity ........................................................................ 106
3-1-3. Implications ................................................................................................ 108
3-2. Estimation by Commodity Categories .............................................................. 113
3-2-1. Estimation Results ...................................................................................... 113
3-2-2. Tests of Homoskedasticity ......................................................................... 114
3-2-3. Implications ................................................................................................ 1 15
CHAPTER VI
COMPUTATION OF THE MARGINAL COST OF PUBLIC FUNDS ....................... 131
1. The Empirical Studies on the MCFs ........................................................................ 131
2. The MCF in the Absence of Tariff Evasion ............................................................ 133
3. The MCF in the Presence of Tariff Evasion ............................................................ 138
3-1. The MCF of the Tariff Rate Change in the Presence of Evasion .................... . 141
3-2. The MCF of the Penalty Rate Change in the Presence of Evasion ................... 145
3-3. The MCF of the Audit Rate Change in the Presence of Evasion ...................... 147
3-4. The MCF of the Detection Rate Change in the Presence of Evasion ............... 150
CHAPTER VII
POLICY IMPLICATIONS ............................................................................................. 174
CHAPTER VIII
CONSLUSION ............................................................................................................... 179
APPENDIX
THE MCF AND THE OPTIMAL PROVISION OF PUBLIC GOODS ........................ 184
BIBLIOGRAPHY ........................................................................................................... 1 88

ix

LIST OF TABLES

Table 1. Comparison of the Models on the MCF in the Presence of Evasion ................ 34
Table 2. Different Effects on Price, Utility, and Tariff Revenue .................................. 35
Table 3. Relationships among MCFs .......................................................................... 36
Table 4. Understatements and Overstatements of Tariff Payments by Commodity ........ 80
Table 5. Tax and Tariff Revenue From Statutory, Conventional, and Elastic Tariff ........ 81
Table 6. Commodity Classification ............................................................................ 82
Table 7. Variable Descriptions ................................................................................... 83
Table 8. Summary Statistics ...................................................................................... 84
Table 9. Mean Values by Commodity Category .......................................................... 85
Table 10. Regression Results of Tariff-Evasion Behavior with Cluster Effect ............... 86
Table 11. Regression Results of Tariff-Evasion Behavior without Cluster Effect .......... 87
Table 12. Results of Specification Tests in the Tobit Model ......................................... 88
Table 13. Tobit Estimates of Tariff-Evasion Behavior ................................................. 89
Table 14. Regression Results of Tariff-Evasion Behavior for Consumer Goods ........... 90

Table 15. Regression Results of Tariff-Evasion Behavior for Raw Materials & Fuels 91

Table 16. Regression Results of Tariff-Evasion Behavior for Capital Goods ................ 92
Table 17. Tobit Estimates of Tariff Evasion by Commodity Category .......................... 93
Table 18. Observasions per Firm across Commodity ................................................... 94

Table 19. Tobit Estimates of Tariff Evasion by Commodity without Cluster Effect ....... 95
Table 20. Results of Specification Test in the Tobit Model by Commodity Categories .. 96

Table 21. Heteroskedastic Tobit Estimates of Tariff Evasion for Consumer Goods ....... 97

Table 22.

Table 23.

Table 24.

Table 25.

Table 26.

Table 27.

Table 28.

Table 29.

Table 30.

Table 31.

Table 32.

Table 33.

Table 34.

Table 35.

Table 36.

Table 37.

Table 38.

Table 39.

Table 40.

Table 41.

Heteroskedastic Tobit Estimates of Tariff Evasion for Raw Materials & Fuels...

.................................................................................................................. 98
Heteroskedastic Tobit Estimates of Tariff Evasion for Capital Goods ............ 99
Different Elasticity of Evasion across Commodities ................................... 100
Elasticity of Evasion with respect to the Foreign Exchange Rate ................ 101
Types of Customs Fraud ............................................................................ 119

OLS, Logit, and Probit Estimates of Customs Fraud with Cluster Efiect ..... 120

OLS, Logit, and Probit Estimates of Customs Fraud without Cluster Effect 121

Tests for Homoskedasticity in Probit and Logit .......................................... 122
Heteroskedastic Probit Estimates of Customs Fraud ................................... 123
Heteroskedastic Logit Estimates of Customs Fraud .......................... , .......... 124

Probit Estimates of Customs Fraud by Commodity Category with Cluster Effect

................................................................................................................ 125

Probit Estimates of Customs Fraud by Commodity without Cluster Effect .. 126
Tests for Homoskedasticity in the Probit Model by Commodity Categories. 127
Heteroskedastic Probit Estimates of Customs Fraud for Consumer Goods... 128

Heteroskedastic Probit Estimates of Customs Fraud for Raw Materials & Fuels

................................................................................................................ 129

Heteroskedastic Probit Estimates of Customs Fraud for Capital Goods ....... 130

Empirical Studies on the MCF in the Presence of Evasion .......................... 152
Effective Tariff Rates by Category ............................................................ 153
Demand Elasticities of Imported Goods by Commodity ............................. 154
The MCF in the Absence of Tariff Evasion ................................................ 155

xi

Table 42.

Table 43.

Table 44.

Table 45.

Table 46.

Table 47.

Table 48.

Table 49.

Table 50.

Table 51.

Table 52.

Table 53.

Table 54.

Table 55.

Table 56.

Table 57.

Table 58.

Table 59.

Table 60.

The MCF: by Category for Different Import Demand Elasticities .............. 156

The MCF; for Different Parameter Values ................................................ 157
Estimated Elasticities of Tariff Evasion by Category .................................. 158
Concealing Costs per One Million Korean Won of Imports ........................ 159
Expected Tariff Rates by Category ............................................................ 160
The MCF, in the Presence of Tariff Evasion with sea = 0.05 and C2 ............. 161
Sensitivity Analysis of the MCF, with Different Parameter Values .............. 162
Expected Tariff Rates with Higher Evasion Ratio (or = 0.1) ........................ 163
The MCF, in the Presence of Tariff Evasion with or = 0.1 ........................... 164
The MCFe in the Presence of Tariff Evasion with sag = -0.01 ..................... 165
The MCFg in the Presence of Tariff Evasion with Sue = -0.05 ..................... 166
The MCFg in the Presence of Tariff Evasion with Sue = -0.1 ....................... 167
The MCFA in the Presence of Tariff Evasion with em = 0.05 and c; ............ 168
Sensitivity Analysis of the MCFA with Different Parameter Values ............. 169
The MCFA in the Presence of Tariff Evasion with or = 0.1 ......................... ' . 170
The MCFD in the Presence of Tariff Evasion with em = 0.05 and C2 ............ 171
Sensitivity Analysis of the MCFD with Different Parameter Values ............. 172
The MCFD in the Presence of Tariff Evasion with or = 0.1 .......................... 173

The MCFs in Our Central Case ................................................................. 178

xii

CHAPTER I

INTRODUCTION

There has been a substantial amount of public finance literature on the marginal cost
of public funds (MCF) per additional dollar of tax revenue, such as Campbell (1975),
Stuart (1984), Wildasin (1984), Ballard, Shoven, and Whalley (1985), and Ballard and
Fullerton (1992). The MCF, which accounts for the distortionary effects of taxes
necessary to finance additional government spending, can be calculated by taking the loss
in consumers’ welfare brought about by a tax change, and dividing it by the amount of
additional tax revenue collected. The traditional method of analyzing the MCF ignores
the existence of tax evasion, compliance costs, evading costs, administrative costs, and
enforcement expenditures. They usually focus only on an efficiency loss from a wedge
between the relative prices for the calculation of the MCF.

However, it is widely believed that tax evasion and direct resource costs, such as
concealing costs incurred by taxpayers and enforcement costs, can also affect the MCF.l
That is, all the indirect costs and direct resource costs should be considered in identifying
the MCF. Thus, the literatures on the MCF that fail to consider the existence of these
costs may overstate or understate the true value of the MCF.

Until recently, only a few economists have been concerned with this issue, and those

who have been concerned with it have used unrealistic assumptions about the probability

 

' Slemrod and Yitzhaki (1996) refer to the costs incurred by society in the process of transferring
purchasing power from the taxpayers to the government as the social costs, which include the cost
of administering the law, compliance costs, and the deadweight losses and expenditures caused by
taxpayers' activities to reduce the tax burden, such as evasion, avoidance, and switching to more
lightly taxed, but otherwise less attractive, consumption. They argue that conclusions from
models of the costs of taxation that ignore these social costs can be misleading.

of detection, enforcement expenditures, and the penalty rates, and have failed to show the
corresponding empirical evidence.

Usher (1986) develops the formulas for the MCFs of an income tax and an excise tax,
incorporating the distortionary effect of tax evasion. His study is the first theoretical
work on the MCF of taxes in the presence of evasion. He argues that the MCF should
include allowances for marginal deadweight loss and marginal cost of tax evasion per
dollar of public revenue acquired.

Yitzhaki (1987) argues that, if we ignore the income effect, then the excess burden of
tax evasion and the excess burden of the tax rate can be added to estimate the total excess
burden of a tax. Basically, he uses a differential analysis, which should be distinguished
from the balanced-budget analysis adopted to calculate the MCF.2 Although Yitzhaki
recognizes that tax evasion leads to welfare losses and tries to formulate the excess
burden of tax evasion, he does not consider the MCF caused by an increase in tax rates to
finance additional public spending.

Kaplow (1990) examines the relationship between optimal taxation and optimal tax
enforcement, and argues that some expenditures on enforcement may be optimal despite
their resource costs, their distortionary effects, and the availability of other revenue
sources having no enforcement costs. Although he does not focus on the MCF, he

indicates that measures of the efficiency cost of raising government revenue may differ in

 

2 Ballard (1990) points out that if the question is whether to undertake an additional tax-financed
government project, and if that project cannot be considered as equivalent to a lump-sum transfer,
then it is appropriate to use a balanced-budget analysis for the calculation of the MCF, in which
income effects and uncompensated elasticities must be considered. However, Browning (1976,
1987) uses a differential analysis to calculate the MCF, in which distortionary taxes are replaced
with lump-sum taxes, holding government spending constant. As indicated by Ballard (1990),
methodologically, it is incorrect to calculate the MCF using a differential experiment, unless one
believes that the relevant marginal government expenditures are very close substitutes for cash.

the presence of tax evasion and enforcement costs.

Falkinger (1991) shows that tax evasion may lead to less public expenditure, but may
also imply a higher optimal level of public-good provision.3 He also proves that if public
goods have zero income effects, which means that the marginal rate of substitution,
UG/ U X, is independent of X, tax evasion has no impact on the optimal level of public good
provision, where UG (=6U/aG) and UX (=6U/6X) denote the derivatives of the taxpayer’s
utility with respect to the level of public good provision, G, and private consumption, X,
respectively.

F ortin and Lacroix (1994) compute not only the MCF associated with tax rates, but
also the MCF with respect to enforcement instruments, such as penalty rates and the
probability of detection, using a simultaneous model of labor supply in the regular and
irregular sectors. In this respect, despite some limitations", their work is a pioneer study
in this area, since it is the first study that makes empirical measurements of the MCF of
tax rates in the presence of tax evasion and the MCFs of revenue sources other than tax
rate increases.

Cremer and Gahvari (1999) show that ignoring tax evasion may lead to

 

3 Falkinger (1991) argues that the optimal level of public goods depends on the derivative of

evasion with respect to government spending EG (= aElaG) and the derivative of evasion with

respect to the tax rate E, (=6E/6t) respectively. For instance, E, >0 means that a higher tax rate

leads to a less-than-proportionate increase in expected tax yields, since tax evasion increases with

t. Thus, the marginal rate of transformation is higher than that in the no-tax evasion case, which

indicates less public expenditure. On the contrary, E,<O implies a higher optimal level of public
oods.

For example, they assume that enforcement expenditures by the government are an increasing
function of the exogenously determined probability of detection. As I explain in detail later, it is
more appropriate and intuitively appealing to assume that the probability of detection is a
function of the fraction of undeclared income or import (evasion), concealing costs, and
enforcement expenditures. In addition, their empirical results are based on a survey of taxpayers’
evasion behavior. However, the results from surveys, especially on evasion, may not be reliable,
since individuals have an incentive to hide information on their illegal behavior.

underestimating as well as overestimating the MCF of commodity taxes. In the presence
of evasion, an increase in the legislated tax rate may increase, as well as decrease, the
size of the marginal tax revenue (MT R) and the marginal utility loss (MUL) of taxation.
The intuition for the ambiguity in the MT R is that, although the new equilibrium value of
the effective tax rate will be less with tax evasion than without, the equilibrium value of
quantity demanded will be more (because price will increase by less). For the MUL,
increasing the legislated tax increases the consumer price by a lesser amount with
evasion, resulting in a smaller increase in the utility loss, while the additional loss due to
concealment increases the value of the MUL. Therefore, tax evasion may decrease, as
well as increase, the MCF = MUL/MT R. Only if the taxed good is perfectly inelastic in
demand can one be certain that tax evasion causes the MCF to increase and exceed its
value in the absence of evasion, since desired effects of tax evasion on the MCF (smaller
changes in price) wash out.

Recently, Poapongsakom et al. (2000) compare the MCF of additional tax
enforcement with the MCF of tax rate changes in the presence of income-tax evasion,
using data for the 1993 taxpayer survey in Thailand. Their empirical analysis indicates
that, in their base case, an increase in income-tax rates on wage earners is a more
attractive policy than devoting more resources to tax enforcement.

However, most studies of the MCF with evasion have been focused on the MCF of the
income tax. No work has been done to measure empirically the MCF of indirect taxes,
such as commodity taxes and tariffs, in the presence of evasion. Furthermore, I am not
aware of any theoretical or empirical works that consider the MCFs of other revenue

sources, such as the penalty and the enforcement activities, when they attempt to analyze

the MCF of indirect taxes.

The purpose of this paper is to provide numerical measures of the MCFs of tariffs
(including internal taxes on imported goods) and other enforcement instruments, in the
presence of evasion, for the case of Korea. For the calculation of the MCFs, I use
detailed data on import declarations in 1998, randomly selected for inspection by the
Korea Customs Service (KCS) at the time of import.

It is meaningful to quantify the MCF of tariffs in the presence of evasion in Korea.
First, the tariff and tax revenues collected by the KCS are a large share of total Korean
government revenues. The pure tariff revenue is 3,836 billion Korean won (W) in 1998,
which is only 5.66% of total tax revenues (including tariffs) collected by the central
government agencies in Korea. However, Article 26-2 of Korea Customs Act prescribes
that internal taxes on import goods, such as value-added tax, special consumption tax,
liquor tax, education tax, etc., shall be collected by the customs house along with the
customs duty. If we include all internal taxes collected by the KCS, the total tariff
revenues (including internal taxes) are 15,9563 billion W, 23.54% of the total Korean tax
and tarifl revenues in 1998.

Second, most countries (including Korea) have adopted simplified clearance
procedures to keep up with rapidly changing international trade environments and the
growing need for a prompt customs clearance, to reduce customs-clearance-related costs.
However, the simplified clearance procedures, such as the self-declaration system, are
frequently abused in efforts to claim lower duty rates, falsify country-of-origin markings,
infringe trademarks, etc. According to the KCS (1999), in the year 1998, the total

number of customs ofi’ences had reached 1,550 cases (about 198 billion W in value). This

may seem like a relatively small amount of fraud. However, even if the total value of
customs fraud is only a little part of the total trade volume, without considering the other
costs associated with these illegal activities, our results on the measure of the MCF may
be misleading. It is shown that the MCF can be changed in the presence of tariff evasion,
private costs of evasion, and administrative costs on enforcement.

Third, we can determine whether a government project should be financed through
higher tariff rates (higher commodity tax rates) or greater enforcement activities to detect
evasion, by comparing the MCF of each of the government policy tools to raise revenues.
This comparison has important policy implications, which are neglected in most of the
theoretical and empirical literature on the MCF.

In chapter 11, I use a simple model of tariff evasion to predict how tariff evasion can be
affected by changes in tariff rates, penalty rates, and enforcement expenditures by the
government. In chapter 111, I derive the formulas for the MCF associated with each of the
government policy tools, and compare them with each other as well as with the MCF
without evasion. In chapter IV, I estimate the major determinants of tariff evasion by a
Tobit model, using the most detailed data sets from the KCS. In chapter V, I examine the
major determinants of customs fraud by a Probit model. In chapter VI, I measure the
MCFs, using the formulas derived from chapter 111 and the parameter values from chapter
IV. In addition, it is shown how the values of the MCF are sensitive to changes in
parameter values. In chapter VII, I discuss policy implications for cost-benefit analysis,
choice of policy tools, and the optimal tax design or tax reform. In the last chapter, I

point out some limitations of this study, and suggest areas for future work.

CHAPTER II

THE MODEL

1. Theoretical Framework5

Consider a small open economy with a large number of identical importers. Imports
are subject to an ad valorem tariff at a rate t, and are sold in a competitive market at a
given price of P. We assume that imports are the only source of a finn’s profit, and that
domestic production is zero. Now suppose that a firm may attempt to evade tariff
payment by declaring only a portion, (l-a), of its true per-unit import price, P“, or its
true quantity, X, to the customs authority, in the presence of the probability of getting
caught, [3, and the penalty rate, 0, when caught. Interpreting this assumption in another
way, such that the importer may declare a fraction of its true tariff payment, (l-a)tP‘X,

we can also include another type of smuggling, false declaration of items so that a lower

tariff rate would apply.7

In this model, we mainly focus on legally disguised smuggling, such as the

 

5 This model is an extension and generalization of the excise-tax framework of Cremer and
Gahvari (1993, 1999). Their model is amended to a tariff-evasion framework. 1 also generalize
their model, making the probability of detection endogenous, explicitly including enforcement
expenditures, etc. For these generalized assumptions and comparative static analysis, I am
greatly influenced by Martin and Panagariya (1984), which is the pioneer work in the analysis of
the uncertain nature of smuggling, introducing the risk and uncertainty into the firm’s smuggling
decision framework for the first time.

6 The world price, P‘, is fixed, since we assume a small Open economy.

7 There are two methods of determining the duty amount: Declaration & Payment and Notice of
Assessment. For the general case in Korea, Declaration & Payment System is practiced, in which
a firm which desires to import goods makes a declaration on payment of the customs duties, upon
making a declaration of import, to the customs authority. In this context, declaration of tariff
payment may be a more appropriate interpretation than declaration of unit price and quantity. It
may be the case that the importers can intentionally claim lower duty rates through false
invoicing of items, even if they report the true unit price and quantity.

undervaluation, the underdeclaration of a quantity (weight), the declaration of false items,
etc., so that we rule out the possibility of smuggling through illegal points of entry
without declaring anything to the customs.8

The probability of getting caught, [3, consists of a two-part process by which the
authorities detect evaders. One part is the probability that the taxpayer is audited, A. The
other part is the probability that evasion is detected, conditional on the occurrence of an
audit, D. Suppose that there are 1,000 import declarations, from which 100 declarations
are selected for inspections. From the 100 declarations audited by customs, 30
declarations are detected for fraud. Then, the probability of audit, A, is 10% (=
100/ 1,000), and the probability of detection given the audit, D, is 30% (= 30/ 100). In this
case, the probability of getting caught, B, equals the probability of audit multiplied by the
probability of detection, conditional on the occurrence of the audit, A x D, which is 3%
(= 0.1 x 0.3). Both the probability of audit and the probability of detection given the
audit are functions of the proportion of imports unreported, or, concealing costs per dollar
of imports by the firm to escape detection, c, and the government expenditures per dollar
of imports to detect tariff evasion, d. If follows that A = A(0t, c. d) and D = D(0t, c, d).
Therefore, [3 = B(or, c, d), where [30, > 0, [3c < O, and [3,; > O. In general, inspection of
imported goods, which leads to detection of evasion or illegal trade, is conducted to
determine: (i) The value of the goods for customs purposes and their dutiable status; (ii)

Whether the goods must be marked with the country of their origin or are in need of

 

8 According to the Korea Customs Service (1999), legally disguised importation accounts for
about 80% of total customs offences in Korea. The other 20% includes direct smuggling by air
and by sea, drug trafficking, money laundering, etc.

special marking or labeling. If so, whether they are marked in the manner as prescribed;
(iii) Whether the shipment contains prohibited articles; (iv) Whether the goods are
correctly invoiced; (v) Whether the goods are in excess of the invoiced quantities.

The concealing cost incurred by the finn is an increasing and convex function of the
undeclared portion of imports, so that cat > 0 and ccm > 0. These costs can be real resource
costs for legally disguised smuggling goods, such as costs of special packaging, costs of
misinvoicing (undervaluing, underweighing, false items, etc.) paid to foreign exporters or
professional counterfeiters, extra costs (premium) of purchasing the foreign exchange in
the black market, etc.

Government expenditures, d, can be thought of administrative costs or efforts to detect
tariff evasion. These may include the development of an information-management
system, the integration of computer systems among law-enforcement agencies, the
introduction of advanced techniques (e.g., financial transaction tracing techniques), the
modernization of inspection equipment (e.g., X-ray inspection machines, wiretapping
devices), and, after inspection equipment is modernized, it has to be operated, training of
investigators, and, after investigators are trained, they have to be paid, etc. The penalty
for tariff evasion is assumed to be proportional to the amount of evaded tariffs.

Compared with Cremer and Gahvari (1999), first of all, I generalize the structure of
the probability of getting caught, and account explicitly for enforcement expenditures by
the government. They assume that the probability of getting caught is exogenously
determined, and they do not consider enforcement expenditures. In this paper, the
probability of getting caught is an increasing function of the proportion of imports

unreported and the enforcement expenditures per dollar of imports, and a decreasing

function of concealing costs per dollar of imports by the firm to escape detection.
Furthermore, I explicitly recognize that the probability of getting caught consists of a
two-part process, the probability of audit and the probability of detection, given an audit.
Each part of the two-part process will be considered as a separate policy variable in the
next chapter. In addition, their model is in a unit-excise-tax evasion framework, while
this model is in an ad-valorem-tariff evasion framework.

The firm chooses or to maximize its expected profit, given t, 9, and d, which are

determined by the government. Thus, the firm maximizes

En = {(1-A)[P-P’-(1-oc)tP’-cP‘]

+ A[(1-D)(P-P‘-(1-a)tP’-cP‘) + D(P-(1+t)P.-90ttP'-cP‘)] }X.

The first term in {} is unit profit if the firm is not audited, and the second term is unit
profit if the firm is audited. The second term is divided into two parts. One part is the
unit profit if the firm is not detected, and the other is the unit profit if the firm is detected,
both of which are conditional on the occurrence of an audit. The firm pays concealing
costs (c) regardless of whether it is audited or detected. The penalty (0) is only assessed
on tariff payments which the firm tried to evade.

Simplifying the firm’s expected profit, we get

En = {(P-P’-(1-or)tP'-CP.) - AD(1+9)ortP‘}X,

= {(P-P‘-(l-a)tP'-cP') — B(1+e)atP‘}X. (1)

IO

The first- and second-order conditions are as follows.

{t - ca- B,(1+9)az - B(1+9)1}P‘X= 0 (2)

{-c.m - Bl'(1+0)0tt — zs.(1+e)z}P‘X< o, (3)

where B, = Ba + Bcca and B.’ = [3m + lime,m denote the first and second derivatives of [3
with respect to or, respectively. It should be noted that both [3. and B" are positive for an
interior solution. Without this condition, nothing will be declared to customs, such that or
= 1. These assumptions are also intuitively appealing. As the deviation from the true
imports or tariff gets greater, it is more likely that the customs authority will be able to
detect the legally disguised smuggling. This implies that the slope of B in terms of or, B1,

is positive, and increases with the fraction of undeclared imports or tariffs.

If we further assume P' > 0 and X > 0, then equation (2) can be rewritten as follows.

I - B;(1+0)ort - B(l+0)t = ca. (2)'

From (2)', it is clear that the firm chooses the optimal value of a so that the marginal
benefits or profits from an underdeclaration of imports are equal to the marginal cost of
tarifi evasion. The left-hand side of (2)' shows the net increase in the firm’s profit from
an underdeclaration. One dollar of undeclared imports increases the firm’s profits by the
tariff rate, t. However, at the same time, it both increases the probability of detection and

the expected penalty, which have negative impacts on profits. The right-hand side of (2)'

ll

represents the additional resource costs incurred by the firm to evade tariff payments.
From equations (1) and (2)', we know that the fraction of imports undeclared, or, is
independent of the firm’s imports, X. Thus, the amount of imports, X, is separable from
the evasion decision, and will be solely determined by the demand side. This separability
arises because total evasion costs, cP‘X, are assumed to be proportional to the amount of
imports, PIX.

It should be noted that tariff evasion takes place if the derivative of the importer’s
expected profit with respect to or, aEn/aa, is positive when evaluated at or = 0. At a = 0,
an increase in or would increase the importer’s expected profit. When evaluating 6E1t/60t
= {t - c..- p.(1+9)az - B(1+e)z}P’X> 0 at a = 0, assuming that P‘ > 0 and X > 0, we

derive the following condition:

t- B(1+9)t > 0 or B(1+0)<1.

This condition has important implications for choosing policy variables to detect evasion

and to increase government revenues. This will be discussed in more detail in later

sections.

12

2. Comparative Static Analysis

Now we examine the effects of a change in each of the policy variables (I, 0, and d) on
the fraction of imports undeclared, or. From the firm’s profit-maximization problem, we
can implicitly derive the optimal value of or = or(t, 0, d).

First, substituting the optimal choice of or into the first-order condition and

differentiating it with respect to t, we obtain

60/61 = [-1 + B1(I+B)0t + [3(l+0)] / [-caa- B1'(1+0)at - 2Bl(l+0)t]. (4)

It can be shown that ant/6t is positive, so that an increase in tariff rates will increase
(decrease) tariff evasion (tariff compliance). This effect is intuitively appealing, in
particular, at t = 0. In the neighborhood of t = 0, we can expect full compliance, since
there is no incentive at all to evade tariffs incurring concealing costs. Then, it is obvious
that an increase in the tariff rate from zero leads to evasion. The interesting thing about
this comparative-static derivative is that the result holds for any value of t. From the
first-order condition, the numerator of the right-hand side of equation (4), [-1 + B.(1+9)or
+ B(1+0)] = -ca/t is negative, since cu is positive. The denominator of equation (4), [-c.m
- B1'(l+0)at - 2B1(1+0)t] is negative because it is just the second-order condition. Thus,
the sign of Bur/at is positive. This implies that, as tariff rates increase, the proportion of
imports undeclared increases, since the additional benefit, 1, from an underdeclaration
outweighs an increase in the expected penalty, [[3(1+0) + B1(1+0)0t]. The first term,

B( 1+0), represents a direct increase in the expected penalty, and the second term,

13

Bi(1+9)(1, is an indirect increase via an increase in the probability of getting caught. It
should be noted that an increase in concealing costs, ca, is already included in (3., since [31
= Ba + Beca-

Second, substituting the optimal choice of or into the first-order condition and

differentiating it with respect to 9, we obtain

50/59 = [1310“ + Bf] / [’caa" Bi'(1+9)0tt - 291(1+9)t]- (5)

The sign of aa/ae can be shown to be negative, so that an increase in the penalty rate
decreases the fraction of true imports undeclared. Since [3. is assumed to be positive, the
numerator of the right-hand side of (5) is positive. It follows that the sign of oer/60 is
negative, since the denominator is negative, while the numerator is positive. This implies
that an increase in the penalty rate has a deterrent effect on tariff evasion, since the
expected net profit from an additional dollar of tariff evasion, {Blow + [St], is negative,
other things being equal.

Third, substituting the optimal choice of or into the first-order condition and

differentiating it with respect to d, we obtain

aa/6d= [Bld(l+9)ou + [3,)(1+0)t]/[-c(m - Bl'(1+0)0tt - 2B1(1+0)t], (6)

where [3 w = Bad + Becca“), which is assumed to be non-negative. The sign of arr/ad is
negative, so that an increase in government expenditures on the detection of tariff evasion

will decrease the fraction of imports unreported. This result is based on the reasonable

14

assumption that Bid is non-negative. We view government spending, d, as administrative
costs or efforts to enhance the effectiveness of an existing investigation or a surveillance
system, such as the introduction of new techniques or equipment to increase the
efficiency, [31 (the slope of B in terms of or) of a detecting system. Therefore, [31 can be
assumed to be an increasing function of d, or at least be constant, so that Bid is non-
negative. It thus follows that the numerator of (6) is positive, since [id is assumed to be
positive. Thus, the sign of aa/ad is negative.

Both an increase in the penalty rate and an increase in the probability of getting caught
by raising government expenditures have deterrent effects on evasion. If they have
positive effects on raising government revenues (i.e., if the marginal tarifi revenues of
these instruments are positive), they can be used as substitutes for each other, other things
being the same. The government would choose policy instruments to eliminate the

incentive of tariff evasion so that they ensure the following condition,
z- B(l+9)t < 0 or p(1+9)>1.9

There can be many combinations of the penalty rate, 0, and the probability of getting
caught, [3, (actually government expenditures, d), that would satisfy this condition. The
optimal value of the penalty rate is infinity, while the optimal level of the probability of
getting caught is zero by setting the government expenditures to zero. This is because an

increase in the probability of getting caught requires government resources, while the

 

9 This condition comes from the previous section.

15

penalty rate can be increased without limit incurring no costs to the government (from
our assumptions given in early this section), and an increased penalty rate would result in
additional government revenues transferred from the evaders, other things being the
same. It seems that an increase in the penalty rate is a more attractive policy tool for the
government.

However, as noted by many economists, such as Cullis and Jones (1998), Myles
(1995), Tanzi and Shome (1993), etc., the penalty rate cannot be raised without limit.
Cullis and Jones (1998, p. 205) point out that “justice requires that ‘the punishment
should fit the crime’ and fines must increase with extent of the crime to preserve marginal
deterrence”. In addition, Tanzi and Shome (1993, pp. 812-3) note that “many societies
feel uncomfortable about singling out and punishing particular individuals, almost by
lottery, when many others have committed the same offenses”, and “if the individual who
gets caught can bribe some tax officials, and if the bribe is less than the penalty, then the
theory becomes ambiguous”.

When we consider the optimal policy in the presence of evasion, a tariff rate should
also be considered as one of the policy variables. In addition, all the policy tools should
be evaluated in the context of maximizing the social welfare or minimizing the welfare
loss of society from raising additional revenues for public spending. They should not be

confined solely to deterrence of evasion or maximization of government revenues.

l6

CHAPTER III

THE MARGINAL COST OF PUBLIC FUNDS WITH TARIFF EVASION

Now consider the policy tools available to the government to finance an additional
dollar for a certain public project. The government can raise revenues for public
spending by increasing tariff rates, by increasing the probability of getting caught by
allocating more resources to tracking down customs fraud, or by raising penalty rates.
The marginal cost of public funds (MCF) may have a different value in each case. This is
neglected in Usher (1986) and Cremer and Gahvari (1999). They only consider an
increase in tax rates as a government revenue source. As a result, they do not propose the
formulas for the MCFs of revenue sources other than tax rate changes, such as the MCFs
of the penalty rate and the enforcement activities. This is summarized in Table 1.

In the following sections, we examine the additional government revenue and the
consumer’s utility losses from each of these government policy tools. Then, we derive

the MCF), where i = t, 0, and d, using the well-known formula,
MCF,- = - (change in consumer welfare) / (additional government revenue).
0

l. Marginal Tariff Revenues in the Absence of Evasionl

\Vrthout evasion, the total tariff revenue, T R‘, is given by,

 

'0 We assume that the economy is on the left-hand side of the Laffer curve, so that the marginal
tariff revenue is positive.

17

TR‘ = zP‘X.“ (7)

Difi’erentiating (7) with respect to t, we get the marginal tariff revenue (MT R),

MTR,‘ = P‘X+ tP‘(aX/aP“)(aP“/az), (8)

where P = (1 + t)P‘ is the market equilibrium price without evasion. We can rewrite

(cf/at) as P‘. Thus, equation (8) can be simplified as follows.

MTR,‘ = P‘X + tP‘P’(aX/aP“)(P“/X)(X/P“)
= P‘X + tP’(eXpA/PA)P'X

= [1 + tP‘(eXp«/PA)]P'X, (8)'

where 8pr is the uncompensated elasticity of demand with respect to the price, PA.
In the presence of evasion, the market equilibrium price differs from the price that
would obtain in the absence of evasion. In addition, as the tariff rate increases, the

consumer price with evasion increases by a smaller amount. These can easily be shown

as follows. Let

’8 5 [(1-00 + [5(1‘r9)0t]t (9)

 

” I am describing a situation in which the government is certain, ex ante, that no evasion will be
attempted at all, so that no enforcement expenditures are needed.

18

denote the firm’s expected tariff payment per dollar of imports. The first term in
equation (9), (l-or)t, shows payment on declared imports, regardless of whether caught.
The second term, B(1+0)ort, represents penalty payment on undeclared imports, which is

only incurred if caught. The market equilibrium occurs at

P=P‘+cP‘+fP‘={1+c+f}P’, (10)

where both c and 1" are evaluated at the firm’s optimal value of or.12

At this price, the
firm’s expected profits are equal to zero. At any other price, the firm would import either
X = 0 or X = 00, and no other price is consistent with the conditions of equilibrium.

From (1), it is clear that c + f < t. Otherwise, the firm would be better off declaring
honestly. This condition has two important implications. First, I is greater than 1", such
that [(l-a) + B(1+0)0t] < 1. This is intuitively obvious in that if t _<_ t‘, then, there is no
incentive at all to evade tariff by incurring concealing costs. Second, the consumer price
without evasion is greater than in its presence, such that P = { 1 + c + 1"}Pt < (l + t)P. =

P .

Differentiating (10) with respect to t, 0, and d, and using equation (2)', we get,

aP/a: = c,(aa/at)P’ + {[(l-a)+13(1+9)a] + [-1+B1(1+0)0t+B(1+0)] (ac/anap‘
= camel/anti + {[(1-a)+B(1+0)0t] - ca(aa/az)}P‘

=[<1-a)+ B(1+9)a] P‘ (11)

 

'2 In the absence of tariff evasion, the market equilibrium occurs at the tariff-inclusive world
price, P = (l + t)P.

19

aP/ae = ca(aa/ae)P' + [Bat - ca(aa/ae)]P‘
= aazP‘ (12)
aP/ad = ca(aa/ad)P‘ + [541mm - ca(aa/ad)]P‘

= Bd(1+e)azp’. (13)

From (1 l), we can see that aP/at = [(l-a) + B(l+0)0t]Pt is less than EPA/6t = P', since
[(l-a) + B( 1+0)a] is less than one. Comparing aP/at and 6P/60, the latter is smaller than
the former. From (11) and (12), it is clear that Bat is less than [(l-a) + [3(l+0)or], so that
[(l-a) + B(l+0)or] - Bat = (l-a) + Bor(l+0-t) is positive. However, without specifying
parameter values, we cannot determine whether aP/ad is greater or smaller than the
others. So far, we see that EPA/6t > aP/at > 6P/60 > 0 and aP/ad > 0. These results have

very important implications for comparisons of the changes in consumer welfare from

each of the policy variables as well as of the changes in the marginal tariff revenue.

20

2. Marginal Tariff Revenues in the Presence of Evasion

In the presence of tariff evasion, total tariff revenue net of enforcement expenditures,

TR, is given by

TR = tep‘X— dP‘X

= {[(1-01) + [5(1+0)a]t—d}P’X. (14)

Comparing (7) and (14), total tariff revenues in the presence of evasion are less than
those in its absence. This is what we expect from the condition, t > 1". Differentiating
(14) with respect to t, 0, and d, and using derivatives of the equilibrium price (equations
(11), (12), and (13)), we can derive formulas for the marginal tariff revenue from each of

the policy variables.

First, the MT R from an increase in the tariff rate is given by

MTR, = {[(l-a) + p(1+e)a] - ca(aa/at)}P‘X+ (re-d)P‘(aX/aP)(aP/az)
= {[(l-ot) + (3(1+9)a] - ca(aa/az)}P‘X+(f-d)P‘(aX/ap)(P/X)[(1-a)+ (1(1+e)a]
x P’(X/P)
= {[(l-a) + (3(1+9)a] - ca(aa/at)}P‘X+ (f-d)P‘(aXp/P)[(1—a) + p(1+e)a] P‘X

= {1 + (f-d)P‘(ex,./P) - cu(6a/at)/[(l-oi) + B(l+0)or]}[(1-or) + B(1+9)a]P‘X, (15)

where exp is the uncompensated elasticity of demand with respect to the price with

evasion, P. Comparing (8)' and (15), we see that, whereas total tariff revenues are

21

unambiguously lower if there is evasion, it is ambiguous whether the same is true for
marginal revenues: MT R: may be greater than or less than MTR,. If 8pr = exp 2 0, then
the marginal tariff revenues without evasion, MT RX, are unambiguously greater than
those with evasion. However, in case of an» at exp < 0, we cannot get clear-cut results.
That is, it is possible that MR: < MT R, . This ambiguity may be explained if we
account for two opposite effects of raising the tariff rate on the MT R, in the presence of
evasion. An increase in the tariff rate will increase the firm’s expected tariff payment per
dollar of imports, te, by a smaller amount. This effect works toward a lower MT R,.
However, the amount of imports will decrease by a smaller amount, since the market
price with evasion will increase by less than 1. Thus, this effect works toward a higher
MT R,. The net effect depends on the relative strength of these two effects.

Second, the MT R from an increase in the penalty rate can be shown as

MTRe = [mama/60) + Bat]P'X+ (re-d)P*(aX/ap)(aP/ae)
= [exact/ac) + Bat]P‘X+ (f—d)P’(6X/6P)(P/X)(X/P)[BatP‘]
= {mama/ac) + Bat + (te-d)P‘(eXp/P)[Bat]}P‘X

= {1 + (f-d)P‘(eXP/P) - ca(601/60)/[Bat]} [Bat]P'X. (16)

We cannot explicitly compare the MT R9 with the MT R: or the MT 12,, even in the case
when 8pr = exp = 0, without specifying the magnitudes of [Bat] and [(l-or) + B(1+0)0t].
For example, when 8pr = exp = 0, the value of { - } in MT R9 is unambiguously greater
than those in MTR; and MTR,. However, since [But] is less than 1 and [(l-or) +

B(l+0)or], we cannot determine whether MT R9 is greater than MT R: or MT R,.

22

Third, the MT Rd is given as

MTR.) = [-Ca(5a/5d) + [3d(1+9)0tt-1]P‘X + (It-d)P‘(aX/aP)(aP/ad)
= [-ca(6or/6d) + pd(1+e)az-1]P‘X +(t"-d)P’(6X/6P)(P/X)(X/P)BA1+0)octP’
= [mafia/ad) + Bd(1+e)oa+ (f-d)P‘(sX,./P)pd(1+e)az - 1]P’X.

= {[1 + (f-d)P‘(eXp/P) — [ca(aor/ad)+1] / [34%)014} [Bd(l+0)at]P'X. (17)

For reasons similar to those given for MT R9, we cannot explicitly compare MT Rd with

MTR}, MTR,, or MTRe.

23

3. Consumer Welfare Losses

Consider an economy with identical individuals, whose utility depends on their
consumption of imported goods, X, supply of labor, L, which is assumed to be
exogenously fixed and treated as the numeraire, and public goods, G, maximizing the

utility function,

U00 + V(CD,

subject to his budget constraint,

PX

u
3“

where P equals P in the presence of evasion and PA in its absence. By assuming identical
consumers, we can ignore distributional issues. Since the utility function is separable
between imported goods, X, and public goods, G, public goods have no effects on
demands for imported goods.

Without tariff evasion, the marginal utility loss (MUL) from an increase in a tariff rate

is given by

MUL,’ = (elf/amt = P‘X.” (18)

 

’3 Mayshar (1990) shows that, in the absence of evasion, the MUL of raising unit tax on any
commodity equals the consumption level of that commodity such that MUL = X, assuming
exogenous income.

24

In the presence of tariff evasion, the marginal utility losses from each of the

government policy tools are as follows:

MUL, = (aP/amr

= [(1-or)+ (3(1+9)a]P‘X . (19)
MULe = (aP/ae)X

= BatP‘X (20)
MUL., = (aP/ad)X

= Bd(1+e)azP‘X. (21)

Comparing (18) and (19), the marginal utility loss from an increase in a tariff rate in the
presence of evasion is smaller than in its absence, since the consumer price with evasion
will increase by a smaller amount. That is, (aP/at) = [(l-a) + B(1+0)01]P. is less than
(EPA/6t) = P'. The marginal utility loss from an increase in the penalty rate is less than
the loss from an increase in the tariff rate, since (6P/60) is smaller than (dP/at).
However, it is not clear whether the marginal utility loss from an increase in the

government expenditures to detect evasion is greater or smaller.
We examine different effects on consumer price, government revenue, and the

consumer’s utility losses from each of the government policy tools. These results are

summarized in Table 2.

25

4. The Marginal Cost of Public Funds of Major Policy Changes

The MCFs can be obtained as the absolute value of the ratio of the marginal utility

losses of the consumer to the additional government revenue, so that the MCF,- =
MUL/MTR), where i = t, 6, and d. We can also get the same formula of the MCF) from
the government maximization problem. The government chooses t, 0, d, and G to

maximize the indirect utility function, which is derived from the individual problem,

U(X(t, 0, a’)) + V(G),

subject to the per-capita government budget constraint,

(f - d)P’X = PGG/H,

where P0 is the price of public goods, and H is the number of identical consumers. After

some algebra”, we can derive the well-known condition of the optimal provision of

public goods,

EMRS= MCF; x MRT,

where ZMRS = H VG/ UX is the sum of the marginal rates of substitution,'5 MCF,- = Mu is

 

’4 The algebra is given in Appendix A.

’5 UX (=6U/6X) and V6 (=6V/6G) denote the derivative of the individual’s utility with respect to
private consumption of imported goods, X and the derivative of the individual’s utility with

26

the marginal cost of public funds associated with each policy variable,16 and MRT = PG/P
is the marginal rate of transformation. The MCF,- = Mu can easily be shown to be equal

to MULi/MTRi, where i = z, 9, and d.”

4-1. The MCF of a Tariff Rate Increase in the Absence of Evasion

\Vrthout evasion, from (8)' and (18), we can derive

MCF,‘ = [P’X] / {[1 + 11013pr / P“)]P‘X}
= 1 / [1 + mien» / PA)]

= 1 / {1 + II /(1+t)18xw}- (22)

In this case, the MCF: depends on the uncompensated elasticity of demand. From the
above formula, it is clear that when an increase in t leads to an increase in P, the demand
for X is reduced, and the MCF is greater than 1. In this case, the government project is
less attractive than it would be if the government’s revenue constraint could be met with
lump-sum taxes only. It is clear from (22) that the MCF,. is greater as imported goods
are more elastically demanded. Not surprisingly, this is in accord with most of the MCF
literature. However, as indicated by Ballard and Fullerton (1992), we should not a priori

conclude that the MCF is always greater than 1.18 When the taxed good is perfectly

 

respect to the level of public goods, G, respectively.

’6 A is the social marginal cost of raising revenue, and [.1 is the marginal utility of income to the
consumer.

'7 This is shown in Appendix A.

’8 They suggest that no general conclusion can be drawn about the marginal costs of taxation used

27

inelastic in demand, so that {5pr =0, then the MCF is equal to 1. If the quantity demanded
of imported goods increases (e.g., the income effect dominates the substitution effect)
with price, then the MCF; can actually be less than one. We may not rule out this
possibility in the presence of (import) commodity taxes. For Giffen goods, by definition,
an increase in price induces an increase in quantity demanded, and vice versa, although
the Giffen case has to be considered very unusual. For some luxury goods (domestic or
imported), it may be the case that the demands for high-priced goods, such as cosmetics
and fur coats, tend to increase with price because of some psychological effects, such as

bandwagon effects, Veblen effects, etc.

4-2. The MCF of Major Policy Changes in the Presence of Evasion

In the presence of tariff evasion, the MCF, may take on a different value from the
value without evasion, since the market equilibrium price differs from the price that
would obtain in the absence of evasion. In addition, each of the government policy tools
affects the price by a different amount. Thus, the MCF,- may have a different value for

each government policy tool.

4-2-1. The MCF of a Tariff Rate Increase in the Presence of Evasion

First, suppose that the government increases a tariff rate. From (15) and (19), we get,

 

to finance a marginal public project. For example, for negative uncompensated labor supply
elasticities, the MCF of a proportional income tax increase is less than one.

28

MCF, = {[(l-a) + p(1+e)a]P‘X} / {{1+(te-d)P’(e,yp/P)-ca(aor/6t)/[(l-a) + B(l+9)0t]}
x [(1-a)+ [3(1+9)a]P‘X}
= 1 / {1 + (tempkeXp/P) - cucu/ [(1-01) + [1(1+9)a]}
= 1 / {1 + (mace/P) - swam/r) / [<1-a) + B(1+9)al}
= 1 / {1 + [(f-d) / (1+c+t")]eXp - smau,(c/t) / [(l-a) + [5(1+8)a]}

= 1 / {1 + [(te'd) / (1+C+te)]8XP ' €ca£ai(C/te)}a (23)

where ecu is the elasticity of concealing costs with respect to the proportion of imports
undeclared, and ea, is the elasticity of the proportion of imports undeclared with respect
to a tariff rate. The MCF, depends not only on exp, but also on em and 8a,. In this case,
even if exp= 0, the MCF, is greater than 1, because the importer incurs direct resource
costs to evade the tariff. From equation (4), as a tariff rate increases, the proportion of
declared imports decreases (i.e., tariff evasion increases). It thus follows that, as tariff
evasion increases, concealing costs incurred by the importer will be increased, since we
assume ca > 0. To finance a certain amount of additional revenue in the presence of
evasion, a tariff rate will be increased by a larger amount than in its absence to offset the
revenue loss, which further increases the distortions of the tariff, other things being equal.
The importer again incurs additional concealing costs. Thus, with evasion, we have
another source of distortions, costs of concealment, which tend to increase the value of
the MCF,.

However, except for the unusual case when 8’pr = exp 2 0, we cannot explicitly
compare the MCF, with the MCF}. If 8pr = EXP 2 O, the MCF, with evasion is

unambiguously greater than in its absence, since the second term of the denominator of

29

equation (22), [t /(l+t)]aXpA], is greater than in (23), [(te-d) / (I+C+le)]8,yp. It should be
noted that, in general, expa ¢ exp, and expe, exp < 0, so that we cannot say that the MC F ,
with evasion is always greater than in its absence. Even if we assume ‘6pr = EXP < 0, this
result holds, since [I /(l+t)] > [(te-d) / (1+c+t")]. In the presence of evasion, we have a
different value of the consumer price, and thus a different amount of demand.
Distortionary effects on the demand for taxed goods will be different, since an increase in
tariff rates has a different effect on the price and thus on the quantity demanded. These
results can be explained in terms of the marginal tax revenue. With evasion, t‘ is less
than t, which lowers the marginal tax revenue. However, in the presence of evasion, the
tax base (the quantity demanded for imported goods) decreases by a smaller amount than
in its absence, other things being equal. Therefore, it is not clear whether the MCF, with

evasion is greater or smaller.

4-2-2. The MCF of a Penalty Rate Increase in the Presence of Evasion

Second, if the government uses penalty rates as an instrument, from (16) and (20),

MCF. = [BatPVG / {I + (flame/P) - ce(6a/69)/[Bat]] [BarrP‘Xi
= 1 I {1 + (f-oP‘ee/P) - ewe/[Bath
= 1 / {1 + (Mace/P) - EocenefC/GJ/[Bwli

= 1 / {1 + [(te-d) /(I+C+le)]8xp - aca8a9(c/0)/[Bat]}, (24)

where age is the elasticity of the proportion of imports undeclared with respect to the

30

penalty rate. In this case, the MCFg depends on exp, am, and Sag. Unlike the MCF,, the
third term in the denominator of equation (24) works toward a lower MCF. From
equation (5), we know that an increase in 0 lowers tariff evasion. Then, costs of
concealment decrease. We cannot explicitly compare the MCFg with the MCF: except in
the case when 8pr = EXP = 0, in which the MCFg has a lower value than the MCF;
However, comparing it with the MCF,, it is clear that the MCFg has a lower value, since
the third term in the denominator in the MCFO is negative, whereas it is positive in the
MCF,. This is intuitively appealing, considering that an increase in 9 lowers concealing
costs through a decrease in evasion, while an increase in t increases these costs through
an increase in evasion. This can be also explained in terms of the marginal tax revenue.
In the case of an increase in the penalty rate, the tax base (the quantity demanded for
imported goods) decreases by a smaller amount than in the case of an increase in the
tariff rate, other things being equal. This is because an increase in the consumer price in
response to an increase in the penalty rate is smaller than in the case of an increase in the
tariff rate. We know that 6P/60 is less than 6P/6t. This result is quite natural, in the
sense that BP/at includes additional concealing costs, while 6P/60 includes the decreased
amount of these costs. Therefore, the government project will be more attractive when

we use the penalty rate to raise additional revenue.

4-2-3. The MCF of an Enforcement Expenditure Increase in the Presence of Evasion

Third, when the government enhances the probability of detection by increasing

expenditures on detection, then from (17) and (21 ),

31

MCFd = [BAHBWP’X] / {11+<t-d>P‘(e.e/P> - [c.<aa/ad)+11/ma<1+e>ar1}
x [Bd(1+0)at]P‘X
= 1 / {1 + (re-d)P’(sXp/P)—[caaar+1]/[Bd(1+0)ort]}
= 1 / {1 + (mace/P) — [s..s..<c/d>+11/[Bx1+e)ar1}

= l / {l + [(te-d) /(I+C+te)]€xp— [emsuAc/d)+l]/[Bd(l+0)0tt]}, (25)

where and is the elasticity of the proportion of imports undeclared with respect to
administrative costs. The MCFd depends on EXP, em , and sad. Unlike those in the MCFt
and the MCFg, the sign of the third term in the denominator of the MCFd is ambiguous.
Because of this ambiguous sign of the third term, it is not easy to compare the MCF.) with
the others. As enforcement expenditures increase, the portion of undeclared imports will
decrease. Then, an increase in tariff compliance lowers the amount of direct resource
costs per unit of undeclared imports that are incurred by the importer to evade the tariff
(slope of c in terms of on). These effects may work toward lowering the MCF. However,
unlike the other government policy tools, the government must use its own direct
resources on enforcement in this case. This makes enforcement spending less attractive,
compared with the other policy variables that have no enforcement costs.

Because of these counteracting effects, we cannot explicitly compare MCF.) with
MCF, or MCFg, unless we make very specific assumptions about the effectiveness of the
government efforts to detect evasion. The MCF.) can have a lower value only when the
third term in the denominator is negative, which depends on the sign of [caafil] =
[emMc/®+l]. The first term in the bracket, caad, shows a decrease in concealing costs

from a one-dollar increase in government spending, and the second term, 1, is the

32

additional government spending to detect evasion. We can consider three cases regarding
the sign of [caafil], which represents the relative effectiveness of additional government
spending on tracking down customs fraud. First, the additional government spending can
be so effective that the resulting decrease in concealing costs is greater than the increase
in the government expenditures, such that clad > 1. In this case, MCF}; is
unambiguously smaller than MCF,. However, it is still unclear whether MCFd is greater
or smaller than MCFg, without specifying the parameter values. Second, if the decrease
in concealing costs is exactly equal to the increase in government expenditures, such that
-ca0td = 1, then MCFd is smaller than MCF,, whereas it is greater than MCFg. Third, it is
possible that additional expenditure to detect customs offenses may not be so effective, so
that the direct resource costs incurred by the importer decrease by a smaller amount than
the additional government spending, such that -caad < 1. In this case, MCFd is greater
than MCFg. However, it is ambiguous whether MCF.) is greater or smaller than MCF,.
Relative magnitudes of the MCFs that do change according to the effectiveness of
enforcement are summarized in Table 3. It should be noted that the relationships among
some MCFs, MCF,’ 2 MCF,, MCF,‘ > MCFe, and MCF, > MCFg, do not change,

regardless of the effectiveness of enforcement.

33

Table 1. Comparison of the Models of the MCF in the Presence of Evasion

 

 

 

 

 

 

 

 

 

 

F ortin and Cremer and
Usher (1986) . _ This Study
Lacrorx (1994) Gahvarr (1999)
Excise tax Excise Tax Tariff
Type of Tax Income tax .
(ad valorem) (unit tax) (ad valorem)
Model Audit ([3) _ Exogenous Exogenous B = B (a, c, d)
Concealing Cost (c) c = c (or, d) — C = C (00 c = c (or)
Enforcing Cost (6!) Exogenous d = d (or, 0) —— Exogenous
MCF of Tax Rate Yes Yes Yes Yes
Formula
MCF of Penalty No Yes No Yes
for MCF
MCF of Audit Rate No Yes No Yes

 

Note: A measure of evasion, a, is different across models.

34

 

Table 2. Different Effects on Price, Utility, and Tariff Revenue

 

Factor

Relative Order

 

Consumer Price

Marginal Utility Loss (MUL)
Total Tariff Revenue (TR)
Marginal Tariff Revenue (MTR)

 

 

51376: > zap/(31> aP/ae > 0, aP/ad > 0
MUL,’ > MUL, > MULe > o, MULd > 0

771‘ > TR
Ambiguous

 

35

 

Table 3. Relationships among MCFs

 

[Goad + I]

Comparison of the MCF

 

Effective Enforcement

('Caad > 1)

MCF,‘ > MCFd

 

MCF, > MCFd, MCFe 2 MCFd

 

Neutral Enforcement

('Caad = 1)

MCF,‘ > MCFd

 

MCF, > MCFd> MCFg

 

Ineffective Enforcement

('Caad < 1)

 

MCF,‘ 2 MCFd

 

 

MCFe < MCFd, MCF, 2 MCFd

 

36

 

CHAPTER IV
THE MARGINAL COST OF PUBLIC FUNDS ASSUMING

EXOGENOUS AUDIT AND DETECTON

1. Theoretical Framework for Tariff Evasion

We modify the theoretical framework for tariff evasion in chapter 11, by assuming that
the probability of inspection (audit), A, and the probability of detection given the
inspection, D, are exogenously determined. This modification is in accordance with our
empirical analysis in chapter V, since we use lagged values for these two variables. With
an explicit consideration of the detection rate given audit separately from the audit rate,
we implicitly assume that customs officials may select the wrong imported goods for
inspection, which are not actually involved in any of customs fraud. All the other
assumptions and notations are the same as those in chapter 11 and III.

The importer chooses or to maximize its expected profit, given t, O, A, and D, which

are determined by the government. Thus, the firm maximizes

Err = {(1-A)[P-P’-(1-a)tP’-cP’]

+ A[(1-D)(P-P'-(l-a)tP’-cP’) + D(P-(1+t)P‘-eazP‘-cp‘)] }X.

Simplifying the firm’s expected profit, we get

Err = {P - P‘ - (i-ayP’- cP‘ - AD(1+e)arP‘ }X. (26)

37

The first- and second-order conditions are as follows.

[1 - c..- AD(1+0)t]P’X= 0 (27)

-c.,.,P’X < 0. (28)

If we further assume P‘ > 0 and X > 0, then equation (27) can be rewritten as follows.

[1 - AD(1+0)]t = ca. (27)'

Tariff evasion takes place if the derivative of the importer’s expected profit with respect
to or, 6E7t/6or, is positive when evaluated at or = 0. At or = 0, an increase in or would
increase the importer’s expected profit. When evaluating aErt/aa = [t - ca- AD(1+O)t]P.X

> 0 at or = 0, assuming that P. > 0 and X > O, we derive the following condition for

interior solution,

t- AD(1+6)t > 0 or AD(1+0) < 1. (29)

Now we examine the effects of a change in each of the policy variables (t, 0, A, and D)
on the fraction of imports undeclared, or. From the firm’s profit maximization problem,
we can implicitly derive the optimal value of or = a(t, 9, A, D).

First, substituting the optimal choice of or into the first-order condition and

differentiating it with respect to t, we obtain

38

act/at = [1 - AD(1+8)] mm. (30)

It can be easily shown that 6a/6t is positive, so that an increase in tariff rates will
stimulate tariff evasion. The numerator of the right-hand side of equation (30), [1 - AD
(1+0)] is positive from equation (27)'. The denominator,-c(m is positive from the second-
order condition. Thus, the sign of 601/6! is positive. This implies that, as tariff rates
increase, the proportion of imports undeclared increases, since the additional benefit, 1,
from an underdeclaration outweighs an increase in the expected penalty, AD(1+9).
Second, substituting the optimal choice of or into the first-order condition and

differentiating it with respect to 0, we obtain

601/60 = -ADt / cm. (31)

The sign of 60/60 can be shown to be negative, so that an increase in the penalty rate
decreases the fraction of true imports undeclared. This implies that an increase in the
penalty rate has a deterrent effect on tariff evasion, since the expected net profit from an
additional dollar of tariff evasion, -ADt, is negative, other things being equal.

Third, substituting the optimal choice of a into the first-order condition and

differentiating it with respect to A, we obtain

601/6A = -D(l+0)t / cm. (32)

From equation (32), it is clear that the sign of 6oz/6A is negative, so that an increase in the

39

inspection rate will decrease tariff evasion.
Fourth, substituting the optimal choice of or into the first-order condition and

differentiating it with respect to D, we obtain

6a/6D = -A(1+0)t / cm. (33)

As with an increase in the inspection rate, it is predicted that an increase in the detection

rate would decrease tariff evasion.

40

2. The Marginal Cost of Public Funds in the Presence of Tariff Evasion

Now the government has four policy options available in the presence of evasion. The
government can raise revenues for public spending by increasing tariff rates, by
increasing the probability of inspection or audit, by increasing the probability of detection
given audit, or by raising penalty rates. First, we examine the additional government
revenue and the consruner’s utility losses from each of these government policy tools.

Then, we derive the MCF), where i = t, 9, A and D, using the formula,

MCF,- = - (change in consumer welfare) / (additional government revenue).

2-1. Marginal Tariff Revenues

Let

f a [(1-11) +AD (1+9)0t]t . (34)

denote the firm’s expected tariff payment per dollar of imports. The market equilibrium

occurs at

P=P‘+cP‘ +t‘P’ = [1 +c+f]P‘, (35)

where both c and t‘ are evaluated at the firm’s optimal value of or.

Differentiating (3 5) with respect to t, 9, A, and D, and using equation (27)', we get,

41

6P/6t = ca(60t/61)P. + {[(1-a)+AD(1+0)0t] + [-1+AD(1+0)] (Ga/60!”).
= ca(6or/6t)P’ + {[(1-a)+A D(1+0)0t] - ca(aa/at)}P‘
= [(1-6) + AD(1+8)01] P‘ (36)
6P/69 = ca(60t/60)P. + [ADOtt - ca(60t/60)]P’
=ADatP‘ (37)
6P/6A = ca(60t/6A)P‘ + [D(1+9)at - ca(60t/6A)]P‘
= D(1+e)atP‘. (38)
6P/6D = ca(6a/6D)P‘ + [A(1+0)at - ca(6a/6D)]P.

= A(1+0)ortP'. (39)

In the presence of tariff evasion, total tariff revenue (TR) is given by

TR = fP‘X

= [(1 -(1) + AD(1+9)a]tP'X . (40)

Differentiating (40) with respect to t, 0, A and D, and using derivatives of the equilibrium
price (equations (36), (37), (38) and (39)), we can derive formulas for the marginal tariff

revenue (MT R) fi'om each of the policy variables.

First, the MT R from an increase in the tariff rate is given by

MTR, = {[(1-a) + AD(1+6)a] - ca(60z/6t) } P‘X + fP‘(6X/6P)(6P/6t)
= {[(1-a) + AD (1+0)a] - camel/at) } R‘X +fP’(6X/6P)(P/X)[(1-oc) + AD(1+0)01]

x P’(X/P)

42

= {[(1-6) + AD(1+0)or] - ca(60t/6t)}P'X+ tcP'(8Xp/P)[(l-a) + AD(1+9)a] P‘X

= {1 + fP‘(eXp/P) - ca(60t/6t)/[(1-0t) + AD(1+9)a]} [(1-6) + AD(1+0)a]P‘X. (41)

Second, the MT R from an increase in the penalty rate can be shown as

MTRe = [-ca(60t/60) + ADatjP’X + teP’(6X/6P)(6P/60)
= [mama/69) + ADatjp‘X + t‘P’(6X/6P)(P/X)(X/P)[ ADatP']
= {-ca(60t/66) + ADat + t‘P‘(eXp/P)[ ADat]}P'X

= {1 + fP’(eXP/P) - cu(601/69)/[ADOtt]} [ADat]P’X. (42)

Third, the MT R from an increase in the inspection rate is given as

MTRA = [-ca(6or/6A) + D(1+0)at]P‘X + t‘P’(6X/6P)(6P/6A)
= ['Ca(aa/6A) + D(1+9)or.t]P'X+fP’(6X/6P)(P/X)(X/P)D(1+6)atP'
= [-ca(60t/6A) + D(1+0)ort + fP‘(eXp/P)D(1+9)az]P‘X.

= {[1 + {Fan/P) — [came/5.4)] / [D(1+9)at]} [13(1 +0)at]P’X. (43)

Fourth, the MT R from an increase in the detection is derived as

MTRD = [-ca(6or/6D) + A(1+9)at]P’X + fP’(6X/6P)(6P/6A)
= [-ca(6a/6D) + A(1+9)ort]P‘X+t‘P'(6X/6P)(P/X)(X/P)A(1+9)atP‘
= [-ca(60t/6D) + A(1+9)0Lt + fP‘(e,-p/P)A(1+e)w]P‘X

= {[1 + fP.(8xp/P) — [ca(60t/6A)] / [A(1+O)at]}[A(1+O)at]P‘X. (44)

43

2-2. Consumer Welfare Losses

In the presence of tariff evasion, the marginal utility losses (MUL) of the

representative consumer from each of the government policy tools are as follows:

MUL, = (6P/6t)X

= [(1-a) +AD(1+9)a]P‘X (45)
MULO = (6P/60)X

=ADatP’X (46)
MULA = (6P/6A)X

= D(1+6)atP‘X. (47)
MULD = (6P/6D)X

=A(1+e)atP‘X. (48)

2-3. The Marginal Cost of Public Funds of Major Policy Changes

The MCFs can be obtained as the absolute value of the ratio of the marginal utility
losses of the consumer to the additional government revenue, so that the MCF,- =
MUL/MTR), where i = t, 6, A and D. In the presence of tariff evasion, each of the
government policy tools affects the price by a different amount. Thus, the MCF,- may
have a different value for each government policy tool.

First, suppose that the government increases a tariff rate. From (41) and (45), we get,

44

MCF, = {[(1-6) + AD(1+0)a]P'X} / {[1+fP‘(eXp/P)-ca(aa/ar)/[(1-6) + AD(1+0)a]]
x [(1-6) + AD(1+e)a]P‘X}

= 1 / {1 + [f / (1+c+te)]sXp - scasa,(c/t")}. (49)

Second, if the government uses penalty rates as an instrument, from (42) and (46),

MCFg = [ADatP’XJ/ {[1 + teP‘(eXp/P) - ca(6a/60)/[A Dat]] [ADat]P’X}

= l / {1 + [t"’ / (1+c+t")]e,yp - 8m8a9(C/9)/[AD(1I]}. (50)

Third, the MCF of the inspection rate in the presence of evasion is derived, using (43)

and (47),

MCF, = [D(1+e)azP'X[ / {[1+z"P‘(eXp/P) — cu(601/6A )/[D(1+9)at]]
x [D(1+9)at]P‘X}

= 1 / {1 + [f / (1+c+f)]exp - emaMc/[AD(1+0)0tt]}, (51)

where 80,; is the elasticity of the proportion of imports undeclared with respect to the
inspection rate.

Fourth, the formula for the MCF associated with the detection rate is derived, using

(44) and (48),

MCFD = [A(1+9)atP‘XJ / {[1+t'-’P‘(eXP/P) — ca(6o/6D)/[A(l+9)at]]

x [A(1+6)az[P‘X}

45

= I / {I + [fe / (1+C+fe)]8xp — SCQEQDC/[AD(I+9)(11]}, (52)

where Sap is the elasticity of evasion with respect to the detection rate given inspection.

46

CHAPTER V

THE EMPIRICAL ANALYSIS OF TARIFF EVASION

1. Empirical Framework

An econometric model to explain tariff evasion behavior can be specified as

EVASION,- = flo + '61 TARIFF,- + flzDETECTIONi + ,63AUDIT; + ARAWFUELS;
+ ,65 CAPGOODS; + ,66 LOG(IMPORT)) + ,6; EXCHANGE,- + ’63 NONPROFIT ,-

+ figINDIVIDUALi + ,810 FIRMAGE + ,6” BRANCH,- + flu OECD; + u,, (53)

where:
E VASION = A measure of tariff evasion,
TARIFF = An effective tariff rate including internal taxes levied on imports,

DETECTION = The probability of detection, conditional on the occurrence of audit,

AUDIT = The probability of audit for the firm,
RAWFUELS = A dummy variable equal to 1 when the imported good belongs to
Raw Materials & Fuels,

CAPGOODS = A dummy variable equal to 1 when the good belongs to Capital Goods,

LOG(IMPORT) = The logarithm of an amount of imports in US. dollars,

EXCHANGE = An exchange rate of Korean won to US. dollar expressed in W 100,

NONPROFIT = A dummy variable equal to 1 when the importer belongs to the non-
profit organization group,

INDIVIDUAL : A dummy variable equal to 1 when the importer belongs to the

47

individual group,

FIRMAGE = Years after the establishment of the firm,
BRANCH = A dummy variable equal to 1 when the importer is a branch,
OECD = A dummy variable equal to 1 when the imported good is originated
from a well-developed country”, and
u = The disturbance term.

Equation (53) describes the major determinants of tariff evasion, where i denotes each
import declaration.

In specifying an empirical framework of tariff evasion, it should be noted that a
substantial portion of import declarations is not involved in tariff evasion. Thus, like
most empirical works on tax evasion, such as Clotfelter (1983), Feinstein (1991), and
Erard (1997), I use a Tobit estimation method to account for the large portion of imports
with non-evasion.20 The Tobit model for tariff-evasion behavior can be formulated using

a latent variable as follows.

y: = 1‘13 + ui (54)

h=mumah. on

where x,- is a vector of exogenous variables, such as tariff and enforcement variables,

 

'9 We measure “well-develop country” by OECD membership.

2° Detailed explanations and econometric treatments of variables like this, called corner solution
outcomes, are given in Wooldridge (2001, Chapter 16). Amount of life insurance coverage
chosen by an individual, charitable contributions, family contributions to an individual retirement
account, and firm expenditures on research and development are examples of corner solution
outcomes.

48

commodity classes, firm-specific characteristics, and a country variable for each
declaration i, B is a vector of coefficients, and y; is a latent variable for the propensity of
the importer to evade tariff payments when he or she files an import declaration i. In
equation (27), u,- is assumed to be distributed N (0, 02). Equation (28) implies that the
observed 'variable (actual tariff payment or the ratio of evaded tariff to true tariff
payment), y), equals y; when y: > 0, and y,- = 0 otherwise.

One problem in the econometric analysis of evasion is that our measure of tariff
evasion, explained in the next section, includes not only understatements of tariff
payments, but also overstatements of tariff payments. Table 4 provides statistics on
underreporting and overreporting of tariff payments.

For Consumer Goods and Materials & Fuels, the frequency and the average amount of
understated tariff payments far exceed the corresponding statistics for overstatements of
tariffs. Thus, the overreporting of tariffs in these two categories can be neglected.
However, for Capital Goods, the average amount of understated tariff payments are
slightly less than the overstatements of tariffs. Thus, one may argue that some part of the
underdeclarations of tariffs in Capital Goods are due to unintentional mistakes rather than
attempting to evade tariffs.2l

However, it may not be the case that unintentional mistakes are included in
understatements of tariffs. Thus, there may not be a symmetric relationship between

underreporting and overreporting. Furthermore, overstatements may not always be

 

2' However, our estimation results for total tariff evasion behavior in a later section rarely change,
even if we exclude Capital Goods in the estimation of the Tobit model for tariff evasion behavior.
The significance of each variable does not change. Only the magnitudes of some coefficients
change a little bit.

49

related to unintentional mistakes.

First, in most cases, professional customs brokers assist the importers in preparing
import declaration forms, including commodity classification.22 In addition, more than
88% of importers belong to for-profit corporations, which have some experience and
skills in filing imports. Most of them may well try to maximize their profits in the legal
system. Thus, there is little possibility that the importers are involved in unintentional
mistakes by overreporting their true tariff payments, and that they just throw their money
out of pockets by pure mistakes. In addition, most importers (especially for-profit
corporations) understand the possible disadvantages they will face when caught for
underreporting, even if it is just due to unintentional errors, since it is very difficult for
customs officers to distinguish an intentional understatement from an unintentional one.
Thus, as with overreporting, it is less likely that unintentional errors are included in the
underreporting.

Second, sometimes, even for-profit corporations and customs brokers may have
difficulties in understanding customs regulations or clearance procedures, because they
require specific technical knowledge in trade. When they are not sure about the Customs
Law or other regulations, including commodity classification, they tend to be
conservative in declaring imports. That is, when there is confusion about the
classification or tariff rates, they may tend to declare the item in a category for which a

higher tariff rate applies, or to overstate their tariff payments. By doing so, the importers

 

22 Article 137-3 of Korea Customs Act prescribes that Customs declaration for import or export
can be submitted under the name of the owner of goods or customs brokers. According to KCS,
about 95% of all export & import declarations are made in the name of customs brokers. It is
more likely that almost all import declarations are submitted under the name of customs brokers.
This is because imports are typically involved in tariff payments and, thus, professional
knowledge of commodity classifications and import regulations is required.

50

and customs brokers can avoid unnecessary conflict with the customs service (e.g., being
named in the black list) or sanctions.23 In addition, these mistakes of overstatements can
be corrected without any difficulty, when they (or customs officials) finally figure out the
true Harmonized Commodity Description and Coding System (HS) code or tariff rates,
which are favorable to them. In most cases, these corrections are made by the customs
before the settlement of the import declaration (or right after the declaration), when the
importers raise questions to verify the appropriateness of the commodity code or tariff
rates. Thus, when they are not sure about the commodity classification or tariff
payments, they may take the safe way, since there is nothing to lose.

Therefore, it is less likely that unintentional mistakes are included in the
understatements of tariffs. In addition, we may not rule out the possibility of intentional
overstatements. For these reasons, the measured level of tariff evasion (y,) is set equal to

zero for import declarations that overstate tariff payments.24

 

23 When the importers are found to be involved in customs fraud, customs brokers who assist the
importers also face various kinds of disadvantages, such as being required to get additional
education, and an increased rate of audit for the imports assisted by that broker. This is often the
case, even if the brokers do not know of the fraud of their customers when they assist the
importers. In particular, an increase in the audit rate for import declarations that are assisted by
specific brokers is considered to be a serious problem for those brokers. When the importers
realize this situation, no one will get assistance from those brokers, and, as a result, brokers may
not continue to do business.

24 In most of the literature on tax evasion, for example, Clotfelter (1983) and Feinstein (1991), the
measure of tax evasion is set to zero for income-tax returns that overstate tax payments, to
maintain the Tobit structure.

51

2. Data Description

In the empirical analysis, I use the most detailed and reliable data sources for customs
fraud, including tariff evasion, from the Korea Customs Service, which is described as
follows.

First, I get the total number of 13,695 forms of the import declarations randomly
selected for investigation at the time of import in 1998. Each observation is a particular
transaction of import, which includes
(i) Firm code including the ‘type of firm’ and ‘year of establishment’,

(ii) Date of import,

(iii) Amount of import in Korean won and US. dollars,

(iv) Declared commodity code (HS) and corresponding pure tariff rate,

(v) Corrected commodity code (HS) and its tariff rate after investigation,

(vi) Country code for origin,

(vii) Type of actions taken by the KCS when detecting violations of the Customs Law,
and

(viii) Final tariff payment (including internal taxes).

Second, I use the data for 4,881 firms on the import and audit records in 1997, of
those who have been randomly selected for inspection in 1998. This finn data will be
used to calculate the audit rate in the preceding year and the perceived penalty rate for
each firm. The data include,

(i) Number of import declarations by the importer,
(ii) Number of inspections on the importer, and

(iii) Number of detections for the importer after inspections.

52

I now describe how the variables in (26) are measured.

Dependent Variable: E VASION

For a measure of evasion, I use the ratio of the undeclared tariff (the difference
between the true tariff and the declared tariff) to the true amount of tariff. These declared
and true amounts of tariff incorporate not the only pure tariff, but also all relevant internal
taxes levied on imports at the time of import, such as VAT, special excise tax, liquor tax,
education tax, etc. We should not rule out the possibility that some importers may have
an incentive to evade internal tax payments, even if they declare pure tarifi’ payments
honestly. For some importers, internal taxes on imports are heavier than the pure tariff.25
In addition, evasion of pure tariff payment affects the amount of internal taxes on imports
as well, since internal taxes (e.g., VAT) are imposed on the sum of the imports (CIF
valuation) and pure tariff payments.26 Thus, it is more appropriate to include all evaded
internal taxes in a measure of evasion, as well as pure tariff.

However, data for tariff and internal taxes are available only for the final payment.
Thus, I use an indirect way to measure evasion using an effective tariff rate. I calculate

an effective tariff rate for each imported good, by dividing the final tariff payment

 

25 The rate of VAT is uniformly set to 10%. In addition to VAT, some goods are subject to special
excise tax, which varies from 7 ~ 30%. For example, a special excise tax of 30% shall be levied
on the jewelry (on the sum of the customs value and the related duty). VAT is not included in the
tax base for special excise tax.

2" Suppose Integrated Circuit (1C) is subject to a pure tariff at a rate of 8% and VAT at a rate of
10%. The firm imports IC with CIF valuation of $100,000. Then, pure tariff and VAT payment
are $8,000 and $10,800 respectively. If the firm declares IC as another item with a tariff rate of
4%, the pure tariff payment can be reduced to $4,000. In addition, the amount of VAT also
decreases to $10,400. Thus, the total tax and tariff payment is reduced from $18,800 to $14,400,

53

including internal taxes by the amount of imports. In the case of false declaration of
items, so that a lower tariff rate would apply, the difference between the true and the
declared effective tariff rate, divided by the true effective tariff rate, can be a good
measure of evasion. However, this measure cannot capture other types of legally
disguised smuggling, such as the importation of undeclared items, undervaluation,
underdeclaration of a quantity or weight, etc.27 For the importation of undeclared items, I
set the evasion rate to 100 %, since all tariff and tax payments on these items are totally
evaded. This is considered to be the most serious violation of Customs Law. For the
other types of evasion, I can calculate an evasion ratio or propensity to evade only when
the actual amount of the undervaluation or the underdeclaration are identified from the

descriptions of actions taken by the KCS after detecting customs fraud.28

T arifi' Variable: TARIFF

The calculation of the effective tariff rates is mentioned in the description of the
dependent variable, E VASION. Except for some elastic tariffs (special customs duties),
there is little room for each government to adjust a pure tariff rate to raise revenue or to

protect the domestic industry.

(Pure) Tariffs in effect in Korea are as follows: (1) Statutory Tariff, (2) Conventional

 

27 One typical example of the importation of the undeclared items is hiding smuggling goods in
the declared items to disguise their illegal activities. Some importers attempt to evade tariff by
not declaring the cost of transportation and/or insurance or by underreporting its unit price or
quantity (weight). . . . . .

2 Only three observations can be identified by thrs method. Thus, our estimation of tariff evasion
is focused mainly on evasion of tariff payments by declaring false description or HS code of

imported goods.

54

Tariff, and (3) Elastic Tariff. Statutory Tariff includes the Basic Tariff and the Temporary
Tariff, which are both regulated by national legislation. Conventional Tariff includes
World Trade Organization (WTO) Conventional Tariff and Preferential Tariff for
developing member countries. Elastic Tariffs can be used to regulate foreign trade or to
stabilize the domestic market. For example, when there is deemed to be an urgent need
to regulate foreign trade for protection of domestic industry, anti-dumping duty,
retaliatory duty, emergency duty, countervailing duty, and beneficial duty are levied on
some designated import goods by Presidential Decree. These Elastic Tariffs have the first
priority in applying tariff rates. However, these adjustments of basic duty rates can only
be allowed in very limited and special situations under the guidance of WTO. 29

In most cases, pure tariff rates for imported goods are determined by multilateral or
bilateral trade agreements rather than determined solely by the government alone.
Statutory Tariff regulated by the national legislation is dominated by Conventional Tariff
in priority of tariff application. Table 5 provides data on the amounts and percentage of
total tax and tariff revenue collected from Statutory, Conventional, and Elastic Tariff in
the year 1998.

Thus, the effective tariff rate including internal taxes may be the more relevant
independent variable in estimating tariff evasion behavior, which is a plausible

government instrummt in raising additional revenues. An effective tariff rate is

 

29 For example, Article 12 I of Korea Customs Act prescribes initial conditions for imposing
Emergency Duty as follows. ‘If it is confirmed through an investigation, that increased imports
of any specified product cause or threaten to cause serious injury to the domestic industry that
produces like or directly competitive products, and if it is deemed necessary to protect such
domestic industry, the customs duties may be imposed additionally in the limit necessary to
prevent or remedy such serious injury and to facilitate adjustment’. The following sections (11 ~
VIII) prescribe complicated conditions and procedures for Emergency Duty.

55

calculated as follows. Suppose that an imported car is subject to pure tariff at a rate, 8%,
and VAT at a rate, 10%. Then, an effective tariff rate, [pure tariff rate + VAT rate (1 +

pure tariff rate)], for the car is 18.8%.

Enforcement Variables: A UDI T and DETECTION

Probability of A udit (AUDIT): I use the ratio of the number of import declarations by
the importer selected for inspection to the total number of declarations by the importer in
1997 as the probability of audit. That is, (number of inspections /number of declarations)
for each importer is used. The number of inspections includes import declarations
selected by customs officers according to certain criteria, as well as those randomly
selected. It should be noted that not only actual inspection of goods, but also reviews of
declared documents by customs officers, such as import-declaration forms, invoices, and
relevant certificates, are included in the number of inspections. However, due to the data
availability, we neither account for post-audits of imports by the Audit Bureau nor
comprehensive investigations by the Investigation and Surveillance Bureau. Here, we
mainly focus on the results from review of import-documents and inspection of goods by
the Clearance Facilitation Bureau, which are done at the time of import.

Probability of Detection (DETECTION): I use the actual detection rate given audit in
1997 for each importer. Thus, the detection rate is calculated by dividing the number of
detections by the number of inspections. The number of detections includes all kinds of
customs fraud, including actual tariff evasion detected after investigation. In addition, the

number of inspections includes not only import declarations by the importer randomly

56

selected for inspection, but also those selected by customs officials in local custom-
houses and by the pre-specified criteria determined by the KCS headquarters.

Penalty Rate (PENALTY): Another important policy tool to enforce compliance is a
penalty rate change. Although the penalty rate for evasion of customs duty varies case by
case, in our data set, all cases except for one are subject to a penalty of 10% of the duties
evaded.30 Thus, we cannot directly estimate the effects of the penalty rate on tariff
evasion, since there are no variations in the penalty rates faced by the importers. It is
fixed to 10% of the evaded tariffs and taxes by legislation. Accordingly, the penalty rate

is excluded from the explanatory variables in the empirical estimation.

Firm-Specific Variables: IMPORT, NONPROFIT, INDIVIDUAL, FIRIWAGE, and

BRANCH

Amount of Imports (IMPORT): I consider the amount of imports valued at transaction
value (unit price x quantity or weight) plus the cost of transportation and insurance to the
frontier of Korea (CIF valuation) in US. dollars.3 ’

Type of the Importers (NONPROFIT, INDIVIDUAL): The Korea Customs Service

classifies importers into six types according to their nature: (1) for-profit corporations, (2)

 

3° Article 180 I (i) of Korea Customs Act prescribes that any person who files a false or no report
on the customs value or the rate of customs duties with the intention of affecting the
determination of taxes shall be punished by imprisonment not exceeding three years, or by a fine
equivalent to the amount not exceeding the larger amount of money between five times the duties
evaded and the cost of the goods. However, due to the difficulty and related costs in proving
intentional evasion in court, most firms detected at the time of import are subject to a penalty of
10% of undeclared tariffs.

3' In most countries including Korea, imports are valued at transaction value plus the cost of
transportation and insurance to the frontier of the importing country or territory.

57

non-profit corporations, (3) government agencies including public enterprises, (4) foreign
corporations, (5) self-employed individuals, and (6) individuals with exemption from
internal taxes. I recategorize these 6 types into 3 broad groups: (1) for-profit and foreign
corporations, (2) non-profit organizations, and (3) individuals. This is because some of
types drop when I estimate the equation for tariff evasion (customs fraud) by commodity
categories.32 I include 2 dummies for each group of the firms, making the for-profit
corporations as the basis.

Years After Establishment of The Firm (FIRMAGE): We can examine the effect of
years after the establishment by including the variable, (1999 — year of establishment).

Branch (BRANCH): Some firms have branch offices other than their headquarters.
From the importer-code, we can differentiate the headquarters from its branch, for each
importer who has at least one branch. Except for the 1 digit in the code, the branch has
the same firm code as its headquarters. I include a dummy variable for the branch to
examine the difference in tariff evasion behavior between the headquarters and its

branches.

Country Variable: OECD

I divide countries into two categories, developed countries and developing countries,

using OECD countries as criteria for deveIOped countries. However, among the 29

OECD countries, I exclude 5 countries that recently joined the OECD: Mexico, the Czech

 

32 Some types of firms drop because they perfectly predict tariff evasion or perfectly do not
predict evasion.

58

Republic, Hungary, Poland, and Korea. In general, these countries are not considered as
developed countries. The dummy variable is equal to one when the imported good is
originated or produced in one of the OECD countries excluding those 5 countries, and
zero when the good originates in one of the non-OECD countries including those 5
countries}3

It should be noted that the country of origin is a different concept from the country of
the last shipment. Suppose that the US. exports products originated in the US. to Hong
Kong, and Hong Kong re-exports these goods to Korea. In this case, the KCS regards the

US. as the country of origin and Hong Kong as the country of the last shipment. This is

the often the case for via third country trade.
Commodity Classification

Commodity Category (C ONGOODS, RAWFUELS, and CAPGOODS): As
recommended by the World Customs Organization (WCO), the KCS classifies imported
goods on the basis of the Harmonized Commodity Description and Coding System (HS),
which consists of 97 2-digit, 1,241 4-digit, 5,113 6-digit, and 11,096 10-digit codes.
However, HS codes are designed mainly to make imposition of tariff easy, so that it may

not be easy to understand the item descriptions of HS codes, which are too detailed.

 

33 The OECD countries that are considered as developed countries in this paper include Australia,
Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Iceland, Ireland, Italy,
Japan, Luxembourg, The Netherlands, New Zealand, Norway, Portugal, Spain, Sweden,
Switzerland, Turkey, United Kingdom, and United States. The original 20 member-countries that
established the OECD in the year 1961 are located in Western countries of Europe and North
America. After the year 1961, Japan (1964), Finland (1969), Australia (1971), and New Zealand
(1973) joined the OECD.

59

Thus, it may not be useful to use HS codes directly to classify goods for our analysis,
even at the 2-digit level.

Instead, I use Import End-Use codes from the KCS, which are designed for the
publication of trade statistics. 1 match Korea HS codes to these Import End-Use codes on
the basis of HS lO-digit codes.34 Imported goods are classified into the categories and

classes given in Table 6.

Other Variables

Foreign Exchange Rate (EXCHANGE): Article 9-13 of Korea Customs Act prescribes
that, in order to determine the customs value in Korean currency of the value expressed in
foreign currency, the Administrator shall determine the exchange rate on the average of
the selling rate of foreign exchange applied during the immediately preceding calendar
week. Thus, various foreign exchange rates are applied to each of the imported goods,
such as exchange rates of Korean won to Japanese yen. I unify these foreign exchange
rates into that of Korean won to US. dollar, which is calculated by dividing imports
valued in Korean won by imports in US. dollars.

Table 7 provides short descriptions of the variables. In the estimation of tariff-evasion
behavior and customs fraud in later sections, a dummy variable for Consumer Goods
(C ONGOODS) and a dummy variable for for-profit corporations (F ORPROFIT) are

excluded as the basis.

 

3" HS 6-digit codes are unified over all the countries, under the guidance of the WCO. For the
remaining digits, each country makes and uses its own codes for specific purposes.

60

From summary statistics given in Table 8, an average fraction of undeclared tariffs is
0.28%. More than 10% of import declarations are involved in various kinds of customs
fraud, including tariff evasion. In terms of the number of import declarations, the shares
of Consumer Goods, Crude Materials & Fuels, and Capital Goods are 16.8%, 35.0%, and
48.1%, respectively. For-profit corporations are responsible for 88.6% of the total
number of imports. More than 72% of the total number of imports are originated or
produced in developed countries.

The mean value of each variable by commodity category is presented in Table 9.
These values are quite different across commodities, especially between the mean values
for Consumer Goods and those for the other two categories. On average, it is more likely
that Consumer Goods are involved in evasion and customs fraud than the other two
categories. The average audit rate on Consumer Goods is more than 20%, while those on
Materials & Fuels and Capital Goods are around 10%. Thus, despite a relatively higher
audit rate in the previous year, the evasion ratio and the fraction of customs fraud are
higher in the case of Consumer Goods, on average.

Self-employed individuals import 19.44% of Consumer Goods, while they import less
than 10% of the other goods. On average, the years of establishment of importers that
import Consumer Goods are relatively short, about 13 years, while those that import

Materials & Fuels and Capital Goods are about 18 and 19 years, respectively.

61

3. Empirical Results

3-1. Total Estimation

3-1-1. Estimation Results

I estimate the Tobit model for tariff-evasion behavior of importers for all observations.
The estimation results for the Tobit model are presented in Table 10. For comparison,
OLS and Probit estimates are reported along with Tobit. However, the magnitudes of the
coefficients are not directly comparable across models. Thus, I compute adjustment
factors for unconditional expectation, by evaluating the standard normal cumulative
distribution function (cdf) at the mean values of the x,-, <D(xBA/oA), which is about 0.0065.
By multiplying the Tobit coefficients by the estimated scale factor, 0.0065, they are made
roughly comparable to the OLS estimates. That is, the partial effect of x, on E(y Ix) is
Obtained from 6E(y I X)/5xj = [80(XBA/0'A) = 0.0065B,~.35 For the Probit estimates, I also
derive the partial effects of the variables xj’s on the response probabilities using the
sample averages of the xj’s, which are comparable to the OLS estimates. The scaled
coefficients for Probit and Tobit are also reported in Table 10.

There are big differences in the magnitudes and statistical significance of the
coefficients across models, while they show the same signs of coefficients. We compare

and discuss the estimation results in detail in the following sub-sections.

 

3’ Wooldridge (2001, Chapter 16) points out that, in comer-solution applicatiqns, interest centers
on E(y I x, y>0) and E(y Ix), while, in data censoring applications, interest lies in E(y Ix). Thus, I mainly
focus on the partial effects on E(y Ix) rather than E(y' Ix).

62

It should be noted that each firm can be viewed as a cluster, since firms make more
than one import declaration. The observations within a cluster are likely to be correlated,
due to unobserved cluster effects. Thus, I correct the standard errors of the above
estimates, specifying that the observations are independent across firms (clusters), but not
necessarily. independent within firms (clusters). However, even if we assume that all
observations are independent with each other, so that there is no clustering effect, it does
not make much difference to the standard errors, and the significance levels of the
coefficients do not change. This may be because we have so many clusters (5713 firms
out of 13695 observations) in our data. Thus, although the number of observations within
a cluster usually differs across clusters, varying from 1 to 472, the average number of
import declarations per firm is only 2.40. The estimation results without considering the

cluster effect are reported in table 11 for comparison.

3-1-2. Specification Tests

An informal diagnostic to test the appropriateness of the Tobit model, especially for
comer-solution applications, is to compare the Probit estimates (B) to the Tobit estimates
(B/o).36 As shown in Table 10, the estimates from the Tobit and the Probit are the same
in their signs, and, except for BRANCH, the same variables are statistically significant in
both models.37 Furthermore, the magnitudes of the Tobit estimates are almost the same

as those of the Probit estimates. Thus, we can roughly verify the appropriateness of the

 

3" o“ in the Tobit model is 0.9801.
37 The t-statistic and p-value for the coefficient of BRANCH in the Probit model is -l .608 and
0.108 respectively.

63

Tobit model, since there are no statistically significant sign changes between the two
models. However, there are some differences between the OLS estimates and the Tobit
and Probit estimates in the statistical significance of the corresponding variables. One of
the major differences is that the coefficient of TARIFF is insignificant in OLS, while it is
statistically significant in the Tobit and the Probit model. In addition, the coefficient of
log(IMPORT) is significant in OLS, while it is not statistically different from zero in the
Tobit and the Probit model.

The Tobit model relies crucially on assumptions of normality and homoskedasticity in
the latent variable model. In the presence of heteroskedasticity or nonnormality, the Tobit
estimator BA is inconsistent for B. Furthermore, in comer-solution applications,
heteroskedasticity or nonnormality will entirely change the functional forms for E(y I x)
and E(y Ix, y>0).

First, the Lagrange multiplier (LM) statistic is used to test the null hypothesis of
homoskedasticity, assuming Var(u I x) = ozexp(zf>), where z is a 1 x Q subsector of x.
The LM test can be carried out without estimating the unrestricted model. As shown in
Table 12, the hypothesis of homoskedasticity is rejected, based on the LM statistic.
However, based on the LR test, we cannot reject the hypothesis. The log-likelihood from
the restricted model (a standard Tobit model) is —608.725, while that from the
unrestricted model (a Tobit model with heteroskedasticity) is —601.570. Then, the LR
statistic is 2[-601.570 - (-608.725)] = 14.310, which is less than the critical value.

As shown in Table 12, the test results from the LM and LR statistics for

homoskedasticity in our Tobit model are contradictory.38 This result may not be
surprising. Numerous Monte Carlo experiments suggest that the outer product of the
score LM statistic has severe size distortions.39 Typically, the null hypothesis is rejected
much more often than the nominal size of the test. That is, these LM tests often have a
size far in excess of their nominal asymptotic size. Thus, we can reasonably argue that
the homoskedasticity assumption is satisfied based on the LR test. Nevertheless, the
estimation results of the Heteroskedatic Tobit model are shown in Table 13 for
reference.40

Second, the conditional moment-based test (Pagan and Vella, 1989) is used to evaluate
the nonnormality in the latent-variable model. The test result is presented in Table 12.
Based on the LM statistic, which is chi-squared with 2 degrees of freedom, the null

hypothesis of normality cannot be rejected.

 

38 The LM and LR statistics are both calculated with respect to the full set of regressors except a
constant. They are asymptotically distributed as chi-squared with 12 degrees of freedom.

39 Wooldridge (2001) and Davidson and Mackinnon (1993) provide several references for this
problem. Davidson and Mackinnon (p.477, 1993) suggest that, in most cases, it is safe to
conclude that a restriction is compatible with the data if a test statistic computed using the outer
product of the gradient estimator fails to reject the null hypothesis. However, they also indicate
that it is generally not safe to conclude that a restriction is incompatible with the data if this
statistic rejects the null.

4° It should be noted that the standard errors for Heteroskedastic Tobit estimates are not adjusted
to account for cluster effect. However, it does make any problem, since the maginitudes of the
coefficients are the same with and without cluster effect.

65

3-1-3. Implications

Tarifi' and Enforcement Variables

First, the tariff and enforcement variables, TARIFF and AUDIT, are found to be
significantly different from zero at the 5% level, while DETECTION is insignificant.

As the tariff rate increases by 1% point, tariff evasion increases by 0.00002. In other
words, a one-percentage-point increase in the tariff rate increases the fraction of the
undeclared tariff payment by 0.002% point. This is a relatively small amount, since the
average of the dependent variable, E VASION, is 0.00279. In other words, the average
value of the fraction of the undeclared tariff payment is 0.279%. The elasticity of the
expected tariff evasion with respect to the tariff rate is 0.114 at the average value of both
variables. Thus, for the importer with an average tariff rate for the sample, 18.70%, a
one-percent increase in the tariff rate (to 18.89%) would result in an expected 0.114%
increase in tariff evasion. In other words, if the tariff were to increase from 18.70% to
18.89%, these estimates suggest that tariff evasion would increase from 0.279% to about
0.2793%. This finding is consistent with the prediction of our theoretical framework.

Although there are no empirical works at all for the effect of a tariff rate on tariff
evasion, we can get some rough idea of the elasticity through comparison with the

estimates of Clotfelter (1983)."1 He finds that the elasticities of income tax evaded with

 

“ There is an extensive review on the theoretical and empirical studies of income-tax evasion in
Andreoni, Erard, and Feinstein (1998). Almost all major issues in tax compliance are fully
described and discussed in this paper. However, they focus mainly on the individual income-tax
evasion problem.

66

respect to the income-tax rate varies from about 0.515 to 0.844 for three classes of
taxpayers (Non-Business, Non-Farrn Business, and Farm), by estimating a standard Tobit
model of tax evasion using data from the 1969 Taxpayer Compliance Measurement
Program. Although the overall elasticity of evasion with respect to the tariff rate, 0.114,
is much smaller than Clotfelter (1983), as shown in a later section, some of the elasticities
for the three commodity categories (Consumer Goods, Raw Materials & Fuels, and
Capital Goods), which vary from 0.102 to 1.026, are not far from his estimates.
However, his estimates of elasticities for ten audit classes show greater variation, ranging
from about 0.5 to over 3.0, which are much larger than our estimates for three commodity
categories.

Tobit estimates for A UDIT show that as the probability of audit or inspection increases
by 1% point, tariff evasion decreases by 0.00005. However, considering the average
value of the expected tariff evasion, 0.00279, practically, an increase in the probability of
inspection has only a small deterrent effect on tariff evasion. The elasticity of this
variable is —0.232 at the average values, implying that, for the importer with an average
probability of being inspected for the sample, 11.994%, a one-percent increase in the
inspection rate would decrease tariff evasion by 0.232% in the following year. In other
words, if the audit rate were to increase from 11.994% to 12.114%, these estimates
suggest that tariff evasion would decrease from 0.279% to about 0.278%. This finding is
also consistent with the prediction of our theoretical framework as well as the general
notion that a higher probability of audit discourages evasion.

However, unlike our expectation, the probability of detection, given audit, has no

significant efiects on tariff evasion. The implication seems to be that being audited is

67

costly, whereas being detected and penalized is not so costly. This is plausible,
considering the fact that the penalty rate is so low, 10% of the evaded tariff payments in
most cases. Thus, whether the importer has the previous experience of being detected,
conditional on the occurrence of audit, does not make any difference, when they decide to
undertake tariff evasion or when they choose the amount of evasion. On the contrary,
importers are likely to care more about being audited, since being audited by customs is
pretty costly. When imports are selected for inspection, importers have to bear all
necessary expenses, such as costs of loading and unloading, costs of unpacking and
repacking, etc. In addition, an inspection of goods can delay the settlement of import
declaration, which may lead to delay of imports to be used as elements of production or

to be sold to another firms or consumers.

Commodity Categories

Second, according to our estimates, all dummy variables for commodity categories are
significantly different from zero. That is, Crude Materials & Fuels and Capital Goods are
less likely to be the target of evasion, compared with Consumer Goods. This may be
caused by relative differences in the degree of ease in evasion according to the nature and
the characteristics of imported goods. In particular, for some agricultural products
(Cereal or Direct Consumer Goods), compared with manufactured consumer goods or
Capital Goods, it is relatively easier to hide the true nature of products. Thus, these items
are more likely to be involved in illegal activities, such as false item declaration so that a

lower tarifi’ rate applies.

68

F irm-Specific Variables

Third, except for LOG(IMPORT) and FIRMA GE, all the other firrn-specific variables
are found to be significant.

Compared with for-profit corporations, individuals have a greater tendency to evade
the tarifi payment. The intuition behind this result is that corporations may care more
about their reputations than individuals do. It is likely that those individuals who are
involved in importing are usually risk-loving. On the contrary, firms will not take the
risk of evasion, since their reputation would be greatly damaged in the case of detection
by customs. Unlike our expectations, non-profit organizations are more likely to be
involved in evasion than for-profit corporations.

The amount of imports valued in US. dollars and years after the establishment have
no significant effects on evasion decision. These results suggest that there are no
differences in evasion behavior between the firms that import a relatively large volume in
CIF valuation and those that import a small volume, other things being the same. Thus,
the amount of import, unit import-price x quantity or weight, does not influence the
evasion decision. Rather, it is the objectives of import, such as Consumer Goods, Raw
Materials & Fuels, and Capital Goods, that affect the level of tariff evasion. In addition,
the firms with a relatively long history do not show any differences in the evasion
decision from the newly established firms.

However, branches are found to have different attitudes from their headquarters in
evasion behavior. They have less tendency to evade tariff or tax payments on their

imports.

69

Country of Origin

Fourth, the country-of-origin variable is found to be significantly different from zero.
This implies that goods originated or produced from the well-developed countries
(OECD) have significant differences in evasion pattern, when compared with less affluent
countries. We usually expect that it is relatively easier for importers to evade tariff
payments for goods imported from the less-developed countries by misinvoicing
(undervaluing, underweighing, false items, etc.) with the collusion of exporters. Foreign
exporters of the less-developed countries would be rarely detected in their home countries
because these countries usually have weak internal control, less-developed information
technology system in customs procedures, etc.

However, the coefficient on the country of origin variable is positive, implying that
goods from the well-developed countries are more likely to be involved in evasion, an
opposite finding to our general expectation. This may be because, for some goods (e.g.,
complicated machines) newly designed and manufactured in the well-developed
countries, it may be relatively easy to declare the good to be another item, to evade tariffs
or some internal taxes. For some customs officials with little knowledge about the new
product, this false declaration of items or description or goods can be often overlooked,
and import declarations may be settled without detecting anything.42

I also estimate tariff evasion, using the G7 countries as criteria for developed

 

“2 The Central Customs Laboratory and four regional customs laboratories of the KCS are
responsible for technical services necessary for effective customs administration, such as tariff
classification. Thus, when customs officials have difficulties in deciding the commodity
classifications of imports, they can ask the laboratories to provide comprehensive and technical
analyses on these goods based on the International Agreement on the Harmonized Commodity
Description and Coding System.

70

countries, which include Canada, France, Germany, Italy, Japan, the United Kingdom,
and the United States. However, there is no change in the sign and significance of the
country-of-origin variable (a dummy variable). The magnitude of the coefficient is
almost the same. In addition, the signs and significance of the other variables do not
change, and the magnitudes of the other coefficients rarely change. It is not surprising
that the results for G7 are similar to the results for OECD, since G7 is a subset of OECD.
When I use top three importing countries of Korea in 1998 (the United States, Japan, and
China), as dummy variables, I get the same results.43 That is, there are no changes in the
signs and significance of the other variables, and the magnitudes of the other variables

are almost the same.

Foreign Exchange Rate

An increase in the foreign exchange rates implies an increase in the unit import price,
other things being the same. That is, equation (10) in part 111 can be expressed in terms
of Korean won, P” = {l + c + t‘?}Pt x exchange rate. Thus, it is predicted that, as the unit

import price increases, importers are more likely to commit tariff evasion to reduce their

increased financial burden.

The coefficient on the foreign exchange rate is significantly different from zero. A
one-hundred won increase in the exchange rate of won to dollar increases tariff evasion

by 0.00064. The elasticity of the expected tariff evasion with respect to the foreign

 

‘3 The amount of imports from these three countries accounts for about half of total Korea imports
in the year 1998. The share of imports from the United States, Japan, and China is 21.9%, 18.1%,
and 7.0%, respectively.

71

exchange rate is 3.162 at the average value of both variables. Thus, for the importer with
an average foreign exchange rate for the sample, W1,380.414 per US. dollar, a one-
percent increase in the foreign exchange rate (to W1,394.218) would result in an expected
3.162% increase in tariff evasion. In other words, if the exchange rate were to increase
from W1,380.414 to W1,394.218, these estimates suggest that tariff evasion would

increase from 0.279% to about 0.288%.

72

3-2. Estimation by Commodity Categories

3-2-1. Estimation Results

I divide the imported goods into 3 categories (Consumer Goods, Raw Materials &
Fuels, and Capital Goods) and estimate a Tobit model for each of the categories
separately. The Tobit estimates for each of the commodity categories are given in Table
14, 15, and 16. For comparison, OLS and Probit estimates are reported along with those
of the Tobit. As in the previous section, I compute adjustment factors for the
unconditional expectations for each of the commodity categories, evaluating the standard
normal cdf at the mean values of the xj’s, <D(xBA/oA). The estimated scale factors for
Consumer Goods, Raw Materials & Fuels, and Capital Goods are 0.0102, 0.0064, and
0.0047 respectively. For the Probit estimates, I also derive the partial effects of the
variables xj on the response probabilities using the sample averages of the xj’s. The
scaled coefficients for Probit and Tobit are also reported in each of these tables. The
standard Tobit estimates by commodity categories are rearranged in Table 17.

It should be noted that the standard errors of above estimates are also adjusted for
cluster sampling, allowing that the observations are independent across firms (clusters),
but not necessarily independent within firms (clusters). The average number of
observations within a cluster usually differs across commodity categories, varying from
1.5 to 2.8, as shown in Table 18. However, the standard errors are very similar to those
computed without accounting for the cluster effect. The estimation results without

considering the cluster effect are reported in table 18 for comparison.

73

3-2-2. Specification Tests

The results of specification tests for homoskedasticity and normality by commodity
categories are presented in Table 20.

First, both the LM and LR statistics are used to test the null hypothesis of
homoskedasticity against the alternative hypothesis of multiplicative heteroskedasticity.44
As in the previous section, the test results for heteroskedasticity are contradictory. Based
on the LR test, we cannot reject the null hypothesis in any of the commodity categories.
However, based on the LM test, the hypothesis can be rejected. Considering that the
outer product of the score LM statistic tends to reject much more often than the nominal
size of the test, we can reasonably argue that the homoskedasticity assumption is satisfied
in each of the categories. The estimates of the heteroskedastic Tobit model by commodity
categories are given in Table 21, 22, and 23 for comparison with those of the
homoskedastic Tobit model.

Second, based on the LM statistic (Pagan and Vella, 1989), which is chi-squared, with
2 degrees of freedom, the null hypothesis of normality cannot be rejected for any of the

commodities.

 

4" Both the LM and LR statistics are calculated with respect to the full set of regressors except a constant.
They are asymptotically distributed as chi-squared with 10 degrees of freedom.

74

3-2-3. Implications

As indicated in Table 17, the Tobit estimates show almost consistent signs among
commodity categories. However, except for the effective tariff rate, the statistical

significance of the coefficients is quite different across commodity categories.

Tarifl and Enforcement Variables

The coefficients for effective tariff rates are found to be significantly different from
zero at the 10% level for all commodity categories. The elasticities of expected tariff
evasion with respect to the effective tariff rate in Consumer Goods, Crude Materials &
Fuels, and Capital Goods are 0.102, 0.852, and 1.026, respectively, at the average values
of the relevant variables. Comparing these values with the overall elasticity of evasion
with respect to the effective tariff rate, 0.114, imported goods falling into Crude Materials
& Fuels and Capital Goods have relatively higher elasticities, while the elasticity for
Consumer Goods is actually very close to the overall elasticity.

These different levels of elasticities across commodity categories can be explained in
terms of the slope of evasion in terms of the tariff rate (the derivative of or with respect to
t) and the ratio of the tariff rate to evasion (t/or). For the case of Consumer Goods, the
slope of evasion and the ratio of the tariff rate to evasion are 0.000025 and 4109.84
(=25.07/0.0061), respectively, both of which are far smaller than those of Crude Materials
& Fuels (0.000111 and 7804.35 (=17.95/0.0023), respectively) and Capital Goods

(0.000117 and 8957.89 (=17.02/0.0019), respectively). This is because, for certain range

75

of or and t, or is a concave function of I so that or, > O and or" < 0. It thus follows that
Consumer Goods has the lowest elasticity of evasion with respect to the tariff rate, 8.1,:
(60r/6t)(t/or), and Capital Goods has the greatest elasticity of evasion. Table 24 shows the
slope of evasion with respect to the tariff rate, the ratio of the tariff rate to evasion, and
the elasticity of evasion with respect to the tariff rate for each of the commodity
categories.

Unlike the prediction of the theoretical model in chapter IV, the Tobit estimates of
DETECTION imply that an increase in the probability of detection would have no
significant deterrent effects on evasion in any of the commodity categories, although it is
of the predicted sign.

The probability of inspection is found to be statistically significant at the 5% level
only in Consumer Goods. The Tobit estimate for A UDIT shows that, as the probability of
inspection increases by a one-percentage point, tariff evasion in Consumer Goods
decreases by 0.0001. The elasticity of tariff evasion with respect to the probability of
audit is -0.416 at the average values. Thus, for firms which import goods classified in
Consumer Goods with an average probability of being inspected for this sample,
21.307%, a one-percent increase in the inspection rate would decrease tariff evasion by
0.416% in the following year. In other words, if the audit rate were to increase from

21.307% to 21.520%, these estimates suggest that tariff evasion would decrease from

0.615% to about 0.612%.

76

F irm-Specific Variables

For the firm-specific variables, we find that there are big differences in the statistical
significance of the independent variables across commodity categories.

The amount of import is found to have no significant effects on evasion behavior in
any of the commodity categories.

The Tobit estimates show that individuals have a higher tendency to evade tariff
payments than for-profit corporations in Raw Materials & Fuels. However, in Consumer
Goods and Capital Goods, there are no differences in evasion behavior between
individuals and for-profit corporations. In all categories, the evasion behaviors of non-
profit organizations are not statistically different from those of for-profit corporations.

Only in Crude Materials & Fuels does the number of years after establishment of the
firm have a significant effect on evasion behavior. The positive coefficient of this
variable implies that firms with a relatively long history have a greater tendency to
commit illegal activities in this category.

However, BRANCH is insignificant in all categories. This implies that branches have

no differences in evasion behavior from their headquarters.

Country of Origin

Only for Capital Goods does the country of origin have significant effects on the

evasion decision. Tobit estimates for this variable show that items classified in Capital

Goods originated from or produced in one of the OECD countries are more likely to be

77

involved in evasion. For Consumer Goods and Crude Materials & Fuels, there are no
differences in evasion propensities, whether they are imported from the well-developed

countries or from the less-developed countries.

Foreign Exchange Rate

In both Consumer Goods and Crude Materials & Fuels, the foreign exchange rates are
significantly different from zero. Thus, as the unit import price increases, firms who
import goods in these categories are more likely to commit evasion to reduce their
financial burden. The elasticities of the expected tariff evasion with respect to the foreign
exchange rate for Consumer Goods and Crude Materials & Fuels are 4.010 and 3.163 at
the average value of both variables, respectively. Thus, for the importer with an average
foreign exchange rate for the sample, a one-percent increase in the foreign exchange rate
would result in an expected 4.010% increase in tariff evasion when importing Consumer
Goods and an expected 3.163% increase in tariff evasion when importing Crude
Materials & Fuels. However, for Capital Goods, an increase in the foreign exchange
rates and, thus, an increase in the unit price, has no significant effects on evasion.

When we compare the elasticity of evasion with respect to the foreign exchange rate
with the elasticity with respect to the tariff rate, the former far exceeds the latter in all
commodities, as summarized in Table 25. Thus, the importers seem much more sensitive
to exchange rates than to tariff rates.

This is because an increase in the foreign exchange rate would give importers

additional incentives of evasion. When the exchange rate increases, not only do firms

78

make higher tariff payments, but they also pay more for the good itself. As a result, the
firm suffers more financial pressure from an increase in the exchange rate than from an
increase in the tariff rate. A one-percent increase in the tariff rate raises tariff payments
by a one-percent. So does the foreign exchange rate. There is no difference in the
amount of changes in tariff payments between an increase in the foreign exchange rate
and an increase in the tariff rate. However, when the foreign exchange rate increases,
importers incur additional costs of imports, since they have to buy the same amount of
foreign currency in exchange for more Korean Won.

This can be illustrated with a simple example. Suppose that the firm imports the
goods of $1000 in the presence of the foreign exchange rate, W1,000 per US. dollar and
the tariff rate of 10% of imports. If a tariff rate increases by 10% (from 10% to 11%),
then the tariff payment would increase by W 10,000. If a foreign exchange rate increases
by 10% (from W1,000 to W 1,100), the total additional burden to the firm is W 110,000. It
is the sum of the increased tariff payment, W10,000, and the additional cost of buying the
foreign currency in the foreign exchange market, W100,000. Thus, it is most likely that
the importer is more sensitive to an increase in the foreign exchange rate than to an

increase in the tariff rate, other things being the same.

79

Table 4. Understatements and Overstatements of Tariff Payments by Commodity

 

 

 

 

 

Understatements Overstatements
Category Observation Under-tariff Avg. Import Observation Over-tariff Avg. Import
Consumer goods 35 1,496 15,331 6 369 14,678
Materials & Fuels 39 3,037 36,303 12 316 9,839
Capital goods 39 583 22,776 26 766 26,382
Overall average _ 1,713 25,139 _ 589 20,274
Total 1 13 193,533 2,840,670 44 25,930 892,063

 

 

 

 

 

 

 

Note: Observation is the number of cases of under- or over-statements. Understatements or overstatements
of internal taxes levied on imported goods are included in these statistics. Under-tariff (Over-tariff) denotes
the average understated (overstated) tariff payments. Avg. import denotes the average amount of imports
(CIF valuation). The unit of Under— and Over-tariffs and Avg. import is one thousand Korean won.

80

 

Table 5. Tax and Tariff Revenue from Statutory, Conventional, and Elastic Tariff

 

 

 

 

 

Revenue
Tariff (%) VAT (%) Total (%)
Statutory tariff 5.089 (75.5) 8,747 (67.8) 13,836 (70.4)
Conventional tariff 902 (13.4) 2,554 (19.8) 3,456 (17.6)
Elastic tariff 748 (11.1) 1,608 (12.4) 2,356 (12.0)
Total 6,739 (100) 12,909 (100) 19,648 (100)

 

 

 

 

Note: The unit of Tariff and VAT is one billion Korean won.

81

 

Table 6. Commodity Classification

 

 

Variable Commodity Description Sub Commodity
«1 Category 1 0 Consumer Goods
- (31355 1 - Cereal wheat. rice, soya bean, corn & fodder
- Class 2 - Direct Consumer Goods tobacco, beverages, meat, fish, etc.
- Class 3 - Durable Consumer Goods electro-machine for home use, car, etc.
- Class 4 - Non-Durable Consumer Goods printed book & product, clothing, etc.
0 Category 2 0 Crude Materials & Fuels
- Class 5 - Fuels crude oil, coal & coke, gas. oil products
- Class 5 - Minerals crude mineral, iron ore, etc.
- Class 7 - Light-Industrial Crude Materials raw sugar, rubber, W009, pUIP, raw cotton
- Class 3 - Oil & Fat oil & fat of cattle, vegetable oil & fat
- Class 9 - Fibre textile yarn & thread, woven fabrics, etc.
- Class 10 - Chemicals (in)organic compounds, fertilizer, etc.
- Class 11 - Iron & Steel Products pig iron, coil. lump. plates & sheets. ete-
- Class [2 - Non-Ferrous Metal copper, aluminum, lead, zinc, tin. nickel
- Class 13 - Others cement, glass & glassware, paper board
0 Category 3 0 Capital Goods _
- Class 14 - Machinery & Precision Equipment machinery, PTCCISIO" equipment
, Class 15 , Electric & Electronic Machinery generator, computer, semi-conductor, etc.
, Class 16 _ Transport Equipment railway vehicles. aircraft, ships & boats
- Class 17 - Others

 

 

Note: Passenger cars are classified into Durable Consumer Goods. Road motor vehicles excluding
passenger cars are included in Transport Equipment.

82

Table 7. Variable Descriptions

 

 

Variable Description

E VASION the ratio of undeclared amount of tariff to true tariff payment

FRAUD 1 if detected for customs fraud afier investigation, 0 otherwise.

TAFIFF an effective tariff rate which includes all internal taxes levied on imported goods at
the time of import as well as a pure tariff

DETECTION the actual detection rate for each importer in 1997, conditional on the audit

AUDIT the ratio of the number of import declarations inspected to the total number of
declarations by the importer in 1997

CONGOODS 1 if the imported good is classified as Consumer Goods (categoryl), 0 otherwise

RAWFUELS 1 if the imported good is classified as Crude Materials & Fuels (categoryZ), 0
otherwise

CAPGOODS 1 if the imported good is classified as Capital Goods (category3), 0 otherwise

IMPORT amount of imports valued in US. dollars

EXCHANGE foreign exchange rate of Korean won for US. dollars

F ORPROF IT 1 if the importer falls into either the for-profit corporation category or the foreign
corporation category

NONPROFIT 1 if the importer falls into either the category of non-profit corporation or
government agency including public enterprise

INDIVIDUAL 1 if the importer falls into either the category of self-employed individuals or an
individual exempted from internal taxes

F IRMAGE years after establishment of the firm, making 1999 as the base year

BRANCH 1 if the importer is a branch, 0 otherwise

OECD 1 if the country of origin of the imported good is one of the OECD countries

 

(except for Mexico, the Czech Republic, Hungary, Poland, or Korea), 0 otherwise

 

83

Table 8. Summary Statistics

 

 

 

Variable Mean Std. Deviation Minimum Maximum
EVASION 0.0028 0.0413 0 1
FRAUD 0.1043 0.3057 0 1
TARIFF (%) 18.700 16.656 0 584.500
DETECTION (%) 11.865 13.957 0 100
AUDIT ' (%) 11.994 15.270 0 100
CONGOODS 0.168 0.374 0 1
RAWFUELS 0.350 0.477 0 1
CAPGOODS 0.481 0.500 0 1
IMPORT (3) 33.216 213,011 1 7,900.819
EXCHANGE (W) 1.380 124.80 638 2,217
FORPROFIT 0.886 0.3 18 0 1
NONPROFIT 0.012 0.110 ' 0 1
INDIVIDUAL 0.102 0.302 0 1
FIRMAGE (year) 17.780 13.042 1 80
BRANCH 0.238 0.426 0 1
OECD 0.721 0.448 0 1

 

 

 

 

 

84

 

Table 9. Mean Values by Commodity Category

 

 

 

Variable Consumer Goods Materials & Fuels Capital Goods
E VASION 0.0061 0.0023 0.0019
FRAUD 0.144 0.096 0.096
TARIFF (%) 25.072 17.954 17.015
DETECTION (%) 14.186 10.276 12.210
A UDIT (%) 21.307 10.975 9.478
IMPORT (8) 21,548 55,299 21,214
EXCHANGE (W) 1,383 1,381 1,379
F ORPROFIT (%) 78.959 89.019 91.701
NONPROFIT (%) 1.605 1.021 1.244
INDIVIDUAL (%) 19.436 9.960 7.055
FIRMAGE (year) 12.674 18.439 19.085
BRANCH (%) 10.152 22.901 29.313
OECD (%) 60.0 73.536 75.376

 

 

 

 

Note: The mean values for dummy variables are calculated by multiplying average values of these variables

by 100%. Thus, for these variables, each mean value represents its fraction in total sample.

85

 

Table 10. Regression Results of Tariff-Evasion Behavior with Cluster Effect

 

Dependent variable: E 1 'ASION

 

 

 

 

 

 

 

 

 

 

Independent Probit Tobit
Variable OLS x 100 [1 x 100 g(xB)B x 100 [1 x 100 13 d>(xB/o) x 100
TARIFF 0.0085 0.267'" 0.0049‘" 0.264’" 0.0017‘"
(0.0063) (0.081) (0.0015) (0.081) (0.0005)
DETECTION -0.0006 -0291 -0.0053 -0.256 -0.0017
(0.0022) (0.252) (0.0046) (0.245) (0.0016)
AUDIT -0.0069"‘ -0876‘” -0.0161“‘ -0828" -00054"
(00025) (0302) (0.0053) (0.291) (0.0019)
RAWFUELS 0.3656” -24.622“ -0.4I7I” -25303'” -01655‘“
(0.1515) (9.768) (0.1538) (9.881) (0.0642)
CAPGOODS -0.4219'” .35.645”' -0.6602'" -35.707”‘ -0.2335”‘
(0.1466) (9.686) (0.1806) (9.961) (0.0647)
LOG(IMPORT) 0.0366“ -1.081 -0.0198 -1507 -00099
(0.0180) (1.711) (0.0312) (1.658) (0.0108)
EXCHANGE 0.0624" 10.420‘” 0.1910‘“ 9.769‘” 0.0639‘”
(0.0298) (2.231) (0.0423) (2.177) (0.0142)
NONPROFIT 0.7785 46.628" 1.5184” 45.776“ 0.2994"
(0.6865) (22.803) (1.1621) (22.829) (0.1484)
INDIVIDUAL 0.3037‘ 22.498" 0.5181“ 22.214” 0.1453“
(0.1734) (10.445) (0.2896) (10.40) (0.0676)
FIRMAGE 0.0047 0.146 0.0027 0.207 0.0014
(0.0039) (0.354) (0.0065) (0.347) (0.0023)
BRANCH 01933“ -17092 -0.2819 -18182‘ -01189‘
(0.0852) (10.679) (0.1579) (10.706) (0.0696)
OECD 0.2105'” 20.920” 0.3447" 21.428” 0.1402"
(0.0650) (9.403) (0.1377) (9.340) (0.0607)
CONSTANT -02359 -364523‘” _. -348692‘” -22807‘”
(0.4245) (34.582) _ (39.853) (0.2590)
Observations 13.695 13.695 13.695
Clusters 5.713 5.713 5,713
Log-Likelihood value N/A -6l8.542 -608.725
Pseudo R2 0.005 0.048 0.050

 

Note: W the estimates are reported after multiplying by 100 to avoid the confusion from rounding. The partial effects
of the dummy variables in the Probit estimation are for discrete changesufrom 0 to 1. The values in parentheses below
the estimates are the corrected standard errors for cluster sampling.
10% levels respectively. The pseudo R-squared for the OLS is the usual R2. For the Probit and Tobit. the pseudo R2 is
the measure based on the log likelihoods.

86

, , and ' denote significance at the 1%, 5%, and

 

Table 11. Regression Results of Tariff-Evasion Behavior without Cluster Effect

 

Dependent variable: E IASION

 

 

 

 

 

 

 

 

 

 

Independent Probit Tobit
Variable OLS x 100 [1 x 100 g(xB)B x 100 [1 x 100 [3 <D(xB/o) x 100
TARIFF 0.0085 0.267‘" 0.0049“ 0.264'" 0.0017'“
(0.0062) (0.10) (0.0019) (0.098) (0.0007)
DETECTION -0.0006 -0291 -0.0053 -0.256 -0.0017
(0.0022) (0.305) (0.0056) (0.296) (0.0019)
AUDIT -00069‘“ 0876'“ 0016!” 0828‘” -0.0054"'
(0.0026) (0.308) (0.0056) (0.304) (0.0020)
RAWFUELS 03656" -24622" 0.4171" -25.303‘” -0.1655“‘
(0.1490) (9.679) (0.1534) (9.598) (0.0627)
CAPGOODS .0.4219‘“ -35.645“‘ -0.6602‘” -35.707"‘ 0.2335'“
(0.1433) (9.632) (0.1822) (9.756) (0.0629)
LOG(IMPORT) 00366“ -1.081 -0.0198 -1507 -0.0099
(0.0181) (1.90) (0.0348) (1.844) (0.0121)
EXCHANGE 0.0624“ 10.420’” 0.l910”’ 9.769‘” 0.0639‘"
(0.0285) (2.466) (0.0444) (2.556) (0.0164)
NONPROFIT 0.7785 46.628" 1.5184” 45.776“ 0.2994"
(0.6721) (22.428) ( 1.1467) (21.990) (0.1444)
INDIVIDUAL 0.3037‘ 22.498" 0.5181" 22.214” 0.1453"
(0.1732) (10.659) (0.3021) (10.495) (0.0688)
FIRMAGE 0.0047 0.146 0.0027 0.207 0.0014
(0.0040) (0.325) (0.0060) (0.316) (0.0021)
BRANCH 0.1933" 47.092 -0.2819 -18.182' .0.1189’
(0.0867) (10.631) (0.1568) (10.495) (0.0680)
OECD 0.2105‘” 20.920" 0.3447“ 21.428” 0.1402"
(0.0656) (8.808) (0.1294) (8.779) (0.0565)
CONSTANT 02359 -364.523“' _. -348692‘“ -2.2807‘”
(0.4102) (38.955) __ (48.875) (0.3522)
Observations 13,695 13,695 1 3.695
Log-Likelihood value N/A -618.542 -608.725
Pseudo R2 0.005 0.048 0.050

 

Note: All the estimates are reported after multiplying by 100 to avoid the confusion from rounding. The partial effects
of the dummy variables in the Probit estimation are for discrete changes from 0 to 1. For the OLS estimates, the values
in parentheses below the estimates are heteroskedasticity-robust standard errors. For Probit and Tobit, the values in
parentheses are the usual standard errors.
The pseudo R-squared for the OLS is the usual R2. For the Probit and Tobit, the pseudo R2 is the measure based on the

log likelihoods.

87

m. u, and ' denote significance at the 1%, 5%, and 10% levels respectively.

 

Table 12. Results of Specification Tests in the Tobit Model

 

 

 

 

Homoskedasticity Normality
LM test LR test LM test
Test statistic 28.156 14.310 0.617
P-value 0.005 0.281 0.735
Critical value 21 .026 21 .026 5.991
Test result rejected not rejected not rejected

 

 

 

 

Note: The critical values for the LM and LR tests of homoskedasticity are derived from the 95th percentile
. in the chi-squared distribution with 12 degrees of freedom. The critical value for the LM test of normality
is derived from the 95th percentile in the chi-squared distribution with 2 degrees of freedom.

88

 

Table 13. Tobit Estimates of Tariff-Evasion Behavior

 

Dependent variable: E II’ASION

 

 

 

 

 

 

 

Independent Homoskedastic Tobit Heteroskedastic Tobit
Variables (3 x 100 11 curd/F“) x 100 11 x 100 0 0(x0’Io”) x 100
TARIFF 0.26?" 0.0017'" 0.071 -00315
(0.081) (0.0005) (0.550) (0.0623)
DETECTION -0.256 -00017 -1501 -0.1014
(0.245) (0.0016) (2.704) (0.2219)
A UDIT -0828" 00054" -3114 -01795
(0.291) (0.0019) (2.060) (0.1627)
RAWFUELS -25303‘” -0.1655"' -14.048 0.6085
(9.881) (0.0642) (26.206) (2.4942)
CAPGOODS -35.707"‘ .0.2335“’ -50.715 -1.8022
(9.961) (0.0647) (37.259) (3.5899)
LoG(IMPOR7) -1.507 -00099 -0.821 0.1005
(1.658) (0.0108) (8.376) (0.7345)
EXCHANGE 9.769‘" 0.0639‘“ 12.036 0.2379
(2.177) (0.0142) (9.965) (0.6286)
NONPROFIT 45.776" 0.2994" 41.182 0.3550
(22.829) (0.1484) (102.175) (9.4369)
INDIVIDUAL 22.214“ 0.1453" 58.090‘ 4.4031
(10.40) (0.0676) (35.003) (3.8738)
FIRMAGE 0.207 0.0014 -1 . 164 .0.1240
(0.347) (0.0023) (1.656) (0.1334)
BRANCH -18.182° -01189‘ 2.732 0.1711
(10.706) (0.0696) (54.607) (4.5254)
OECD 21.428” 0.1402” 23.614 0.2609
(9.340) (0.0607) (36.057) (2.8815)
CONSTANT -348692‘“ -22807‘“ 339.481" -1.8487
(39.853) (0.2590) (171.686) (9.1004)
a 0.9801 0.9175
Observations 1 3,695 1 3,695
Log-Likelihood value -608.725 -601.570
Pseudo R2 0.05 —

 

 

 

Note: All the estimates are reported after multiplying by 100 to avoid the confusion from rounding. The values in
parentheses below the Homoskedastic Tobit estimates are the standard errors, corrected for cluster effect. The values in

parentheses below the Heteroskedastic Tobit estimates are the standard errors, not corrected for cluster effect.
and denote significance at the 1%, 5%, and 10% levels respectively.

89

.0. O.

 

Table 14. Regression Results of Tariff-Evasion Behavior for Consumer Goods

 

Dependent variable: E I 'ASION

 

 

 

 

 

 

 

Independent Probit Tobit
variable OLS x 100 (3 x 100 g(xB)B x 100 [1 x 100 [3 map/6) x 100
TARIFF 0.0076 0.221' 0.0058' 0.245r 0.0025'
(0.0067) (0.101) (0.0027) (0.112) (0.0011)
DETECTION 0.0027 0.350 0.0092 -0.338 0.0035
(0.0049) (0.432) (0.0109) (0.453) (0.0046)
AUDIT 0.0099" -1.153°‘ 0.0302" 4.175“ 0.0120"
(0.0047) (0.443) (0.0104) (0.446) (0.0045)
LOG(IMPORT) 0.0319 0.714 0.0187 -1 . 10 0.0113
(0.0447) (2.475) (0.0652) (2.665) (0.0272)
EXCHANGE 0.1682‘ 17.118'” 0.4488‘” 17.419‘” 0.1783‘”
(00953) (4533) (0.1343) (4.450) (0.0454)
NONPROFIT 0.6474 67.151' 3.7665‘ 65.697 0.6724
(1.2485) (37.437) (3.6622) (38.713) (0.3949)
INDIVIDUAL 0.2870 29.501‘ 0.9647‘ 29.162 0.2985
(0.3452) (18.020) (0.7049) (18.825) (0.1920)
FIRMAGE 0.0087 0.093 0.0024 0.162 0.0017
(0.0179) (0.735) (0.0192) (0.830) (0.0085)
BRANCH 0.6597 .27.453 05630 01.971 0.3272
(0.5067) (35.313) (0.5330) (38.973) (0.3975)
OECD 0.4601' 21.939 0.5514 25.286 0.2588
(0.2354) (16.053) (0.3789) (17.612) (0.1796)
CONSTANT -1.7713 458.515‘” _ 482.764” .4941 1‘"
(1.3221) (66.898) _ (75.976) (0.7750)
Observations 2,305 2.305 2.305
Clusters 1.551 1,551 1.551
Log-Likelihood value N/A -l62.019 -l66.699
Pseudo R2 0.007 0.085 0.076

 

 

 

 

Note: All the estimates are reported after multiplying by 100 to avoid the confusion from rounding. The partial effects
of the dummy variables in Probit are for discrete changes from 0 to 1. The values in parentheses below the estimates

are the standard errors. adjusted for cluster sampling.

, ., and ' denote significance at the 1%, 5%, and 10% levels

respectively. The pseudo R-squared for the OLS is the usual R2. For Probit and Tobit, the pseudo R2 is the measure

based on the log likelihoods.

90

 

Table 15

lndep

7m

BETH“

.ILIDIT

g LCXEU

Eltfif

I NONI
”D! I

I FIR)
Blit.

; art

I (If)

Table 15. Regression Results of Tariff-Evasion Behavior for Raw Materials & Fuels

 

Dependent variable: E I'ASION

 

 

 

 

 

 

 

Independent Probit Tobit
variable 01.5 x 100 13 x 100 g(xB)B x 100 0 x 100 0 (D(xB/o) x 100
TARIFF 0.0254 2.141" 0.0390f ”29¢ 0.0111"
(0.0178) (0.879) (0.0169) (0.720) (0.0046)
DETECTION 0.0008 0.415 0.0076 0.299 0.0019
(0.0041) (0.428) (0.0077) (0.348) (0.0022)
AUDIT 0.0021 0.749 0.0137 0.541 0.0035
(0.0044) (0.657) (0.0116) (0.506) (0.0032)
LOG(IMPORT) 0.0466 -3024 0.0551 -2915 0.0188
(0.0421) (2.956) (0.0528) (2.573) (0.0165)
EXCHANGE 0.1107‘ 9.766’“ 0.1779‘” 8.322" 0.0536“
(0.0605) (3.665) (0.0660) (3.267) (0.0209)
NONPROFIT 0.2705 48.916 1.6334 37.619 0.2422
(03330) (44.454) (2.3785) (34.594) (0.2214)
INDIVIDUAL 0.5606‘ 38.216“ 1.0335“ 33.053" 0.2128“
(0.3239) (17.095) (0.6172) (15.082) (0.0965)
FIRMAGE 0.0094 0.867 0.0158 0.754' 0.0049‘
(0.0061) (0.528) (0.0093) (0.454) (0.0029)
BRANCH 0.1977’ -15.625 0.2574 04.428 0.0929
(0.1166) (17.277) (0.2547) (14.282) (0.0914)
OECD 0.1162 11.865 0.2022 9.944 0.0640
(0.1005) (14.755) (0.2351) (11.947) (0.0765)
CONSTANT -1.5702‘ 407.323’“ _ 034.621‘” .2.1547‘”
(0.9021) (64.614) _ (69.069) (0.4420)
Observations 4,799 4.7 99 4,799
Clusters 2.560 2,560 2,560
Log-Likelihood value N/A -214.957 -204.360
Pseudo R2 0.007 0.051 0.057

 

 

 

 

Note: All the estimates are reported after multiplying by 100 to avoid the confusion from rounding. The partial effects
of the dummy variables in Probit are for discrete changes from 0 to 1. The values in parentheses below the estimates

are the standard errors, adjusted for cluster sampling. ...

, and ' denote significance at the 1%, 5%, and 10% levels

respectively. The pseudo R-squared for the OLS is the usual R’. For Probit and Tobit. the pseudo R2 is the measure

based on the log likelihoods.

91

 

Table 16. Regression Results of Tariff-Evasion Behavior for Capital Goods

 

 

 

 

 

 

 

 

Dependent variable: E I'ASION
Independent Probit Tobit
variable OLS x 100 [1 x 100 g(xB)B x 100 [3 x 100 [3 ¢(xB/o) x 100
TARIFF 0.0307' 2.406'" 0.0334“ 2.457'" 0.01 17'"
(0.0175) (0.678) (0.0107) (0.772) (0.0036)
DETECTION 0.0003 0. 181 0.0025 0. 170 0.0008
(0.0026) (0.455) (0.0064) (0.453) (0.0021)
AUDIT 0.0079‘ 0.690 0.0096 0.810 0.0038
(0,0043) (0.617) (0.0085) (0.662) (0.0031)
LOG(IMPORT) 0.0240 0.090 0.0013 0.601 0.0029
(0.0161) (2.857) (00397) (2.716) (0.0128)
EXCHANGE 0.0143 5.728' 0.0795‘ 4.864 0.0231
(0.0210) (3.446) (0.0486) (3.221) (0.0151)
NONPROFIT 1 .4020 48.699 1.271 1 59.038 0.2803
(1.3807) (41.846) (1.7692) (45.869) (0.2156)
INDIVIDUAL 0.0421 08.459 0.2099 05.595 0.0740
(0.2265) (25.597) (0.2378) (25.586) (0.1203)
FIRMAGE 0.0003 0.569 0.0079 0.494 0.0023
(0.0041) (0.510) (0.0072) (0.495) (0.0023)
BRANCH 0,0390 - 10.096 -0. 1331 01.298 -0.0536
(0.0928) (13.560) (0.1713) (13.819) (0.0650)
OECD 0.1527‘ 35.268' 0.3993‘ 37.034“ 0.1758‘
(0.0780) (18.369) (0.1673) (19.536) (0.0918)
CONSTANT 0.0436 -381.645"° _ 069.878‘“ 0.7561’“
(0.4662) (58.304) _ (68.312) (0.3211)
Observations 6,591 6.591 6.591
Clusters 2.339 2.339 2.339
Log-Likelihood value N/A -229251 -224.849
Pseudo R2 0.004 0.041 0.041

 

 

 

 

Note: All the estimates are reported afier multiplying by 100 to avoid the confusion from rounding. The partial effects
of the dummy variables in Probit are for discrete changes from 0 to 1. The values in parentheses below the estimates

are the standard errors. adjusted for cluster sampling. ...

, u. and ° denote significance at the 1%, 5%, and 10% levels

respectively. The pseudo R-squared for the OLS is the usual R2. For Probit and Tobit, the pseudo R2 is the measure

based on the log likelihoods.

92

 

Table I7. Tc

_____
l

. lndepender
) variables

‘1' DETECTIO

l .0017
l

l
LOG(IMPC

l
1 EXCHAN

l N'O—VPRC

l

l l.\Dll'll

l

l FIRMA.

1
1
l

' 88m

l OECL

Table 17. Tobit Estimates of Tariff Evasion by Commodity Category

 

 

 

 

 

 

 

 

 

Dependent variable: E VASION
Consumer Goods Raw Materials & Fuels Capital Goods
Independent [3 (pup/0) B (D(xB/o) B <D(xB/o)
variables B x 100 B x 100 [3 x 100
x 100 x 100 x 100
TARIFF 0245* 0.0025‘ 1.729" 0.01 11" 2.457" 0.0117'"
(0.1 12) (0.001 1) (0.720) (0.0046) (0.772) (0.0036)
DETECTION 0.338 0.0035 0.299 0.0019 0.170 -0.0008
(0.453) (0.0046) (0.348) (0.0022) (0.453) (0.0021)
A UDIT -1 .175“ 0.0120” 0.541 0.0035 0.810 0.0038
(0.446) (0.0045) (0.506) (0.0032) (0.662) (0.0031)
LOG(IMPORT) -1 . 10 0.0113 -2915 0.0188 0.601 0.0029
(2.665) (0.0272) (2.573) (0.0165) (2.716) (0.0128)
EXCHANGE 17.419‘“ 0.1783’” 8.322" 0.0536“ 4.864 0.0231
(4.450) (0.0454) (3.267) (0.0209) (3.221) (0.0151)
NONPROFIT 65.697 0.6724 37.619 0.2422 59.038 0.2803
(38.713) (0.3949) (34.594) (0.2214) (45.869) (0.2156)
INDIVIDUAL 29.162 0.2985 33.053" 0.2128“ -15.595 0.0740
(18.825) (0.1920) (15.082) (0.0965) (25.586) (0.1203)
FIRMAGE 0.162 0.0017 0.754‘ 0.0049' 0.494 0.0023
(0.830) (0.0085) (0.454) (0.0029) (0.495) (0.0023)
BRANCH -31.971 0.3272 -l4.428 0.0929 -11.298 0.0536
(38.973) (0.3975) (14.282) (0.0914) (13.819) (0.0650)
OECD 25.286 0.2588 9.944 0.0640 37.034' 0.1758"
(17.612) (0.1796) (11.947) (0.0765) (19.536) (0.0918)
CONSTANT -482764'” -49411‘” -334621‘” -21547'“ -369.878“‘ -l.756l“'
(75.976) (0.7750) (69.069) (0.4420) (68.312) (0.3211)
6" 1.0960 0.8141 1.0060
Observations 2,305 4.799 6,591
Clusters 1.551 2.560 2,339
Log-Likelihood 466.699 -204.360 -224.849
Pseudo R2 0.076 0.057 0.041

 

 

 

 

Note: All the estimates are reported after multiplying by 100 to avoid the confusion from rounding. The values in
parentheses below the estimates are the standard errors, adjusted for cluster sampling. , , and denote significance
at the 1%, 5%. and 10% levels respectively.

93

 

Table 18. Observations per Firm across Commodity

 

 

 

 

Category Minimum Average Maximum
Consumer Goods l 1.5 33
Raw Materials 1 1.9 41
Capital Goods 1 2.8 472

Overall 1 2.4 472

 

 

 

 

94

 

Table 19. Tobit Estimates of Tariff Evasion by Commodity without Cluster Effect

 

 

 

 

 

 

 

 

 

Dependent variable: E l'ASION
Consumer Goods Raw Materials & Fuels Capital Goods
Independent 0 ¢(x13/o) 13 <D(xB/0') B <D(xB/o)
variables B x 100 B x 100 B x 100
x 100 x 100 x 100
TARIFF 0.245fi 0.0025' 1.729” 0.01 11‘" 2.457'" 0.01 17'"
(0.127) (0.0013) (0.638) . (0.0043) (0.821) (0.0041)
DETECTION 0.338 0.0035 0.299 0.0019 0.170 0.0008
(0.577) (0.0059) (0.429) (0.0027) (0.542) (0.0026)
AUDIT -1 .175“ 0.0120” 0.541 0.0035 0.810 -0.0038
(0.528) (0.0050) (0.498) (0.0032) (0.657) (0.0031)
LOG(IMPORT) -1.10 0.0113 -2.915 0.0188 0.601 0.0029
(4.098) (0.0419) (2.625) (0.0168) (3.112) (0.0148)
EXCHANGE 17.419'" 0.1783‘” 8.322“ 0.0536" 4.864 0.0231
(6.056) (0.0609) (3.559) (0.0226) (4.338) (0.0204)
NONPROFIT 65.697 0.6724 37.619 0.2422 59.038 0.2803
(42.292) (0.4381) (35.589) (0.2294) (40.348) (0.1936)
INDIVIDUAL 29.162 0.2985 33.053“ 0.2128" .15.595 0.0740
(19.438) (0.1998) (14.715) (0.0944) (26.175) (0.1237)
FIRMAGE 0.162 0.0017 0.754' 0.0049‘ 0.494 0.0023
(0.747) (0.0076) (0.405) (0.0026) (0.580) (0.0027)
BRANCH -3l.971 0.3272 -14.428 0.0929 -1 1.298 0.0536
(32.344) (0.3301) (13.911) (0.0890) (16.205) (0.0766)
OECD 25.286 0.2588 9.944 0.0640 37.034‘ 0.1758"
(17.639) (0.1784) (11.892) (0.0763) (18.960) (0.0836)
CONSTANT -482.764‘” 49411” -334.621”' 21547” -369878‘” -1.756l‘”
(112.982) (1.2944) (73.802) (0.5348) (85.991) (0.4566)
o" 1.0960 0.8141 1.0060
Observations 2,305 4,799 6,591
Log-Likelihood -166.699 -204.360 -224.849
Pseudo R2 0.076 0.057 0.041

 

 

 

 

Note: All the estimates are reported afier multiplying by 100 to avoid the confusion from rounding. The values in

parentheses below the estimates are the usual standard errors. ..

levels respectively.

95

., u, and ' denote significance at the l°/o, 5%, and 10%

 

Table 20. Results of Specification Test in the Tobit Model by Commodity Categories

 

 

 

 

 

 

 

Homoskedasticity Normality
Commodity LM test LR test LM test

Test statistic 58.772 10.528 0.618
Consumer Goods P- value 0.000 0.395 0.734
Critical value 18.307 18.307 5.991

Test result rejected not rejected not rejected
Test statistic 81.703 13.750 4.695
Raw Materials & P- value 0.000 0.185 0.096
Fuels Critical value 18.307 18.307 5.991

Test result rejected not rejected not rejected
Test statistic 61.373 17.798 3.580
Capital Goods P- value 0.000 0.058 0.167
Critical value 18.307 18.307 5.991

Test result rejected not rejected not rejected
Test statistic 28.156 14.310 0.617
Total P- value 0.005 0.281 0.735
Critical value 21.026 21.026 5.991

Test result rejected not rejected not rejected

 

 

 

 

 

Note: The critical values for LM and LR tests of homoskedasticity in commodity categories are derived
from the 95th percentile in the chi-squared distribution with 10 degrees of freedom. The critical values for
LM tests of normality in commodity categories are derived from the 95th percentile in the chi-squared

distribution with 2 degrees of freedom.

96

 

Table 21. Heteroskedastic Tobit Estimates of Tariff Evasion for Consumer Goods

 

Dependent variable: E l IdS/ON

 

 

 

 

 

 

 

Independent Homoskedastic Tobit Heteroskedastic Tobit
variables 13 x 100 [1 <1>(x13‘/€‘) x 100 13 x 100 B ¢(xB‘/o’) x 100

TARIFF 0.245' 0.0025‘ 0.011 0.0282
(0.112) (0.0011) (1.006) (0.1095)

DETECTION 0.338 0.0035 -3.066 0.2412
(0.453) (0.0046) (6.020) (0.7371)

A UDIT 4 . 175" 0.0120” -2.766 0.1736
(0.446) (0.0045) (4.203) (0.4187)

LOG(IMPORT) 4.10 0.0113 6.480 0.8961
(2.665) (0.0272) (21.174) (2.9525)

EXCHANGE 17.419‘“ 0.1783’" 22.742 0.6490
(4.450) (0.0454) (22.122) (1.7582)

NONPROFIT 65.697 0.6724 42.512 4.3994
(38.713) (0.3949) (132.857) (16.7105)

INDIVIDUAL 29.162 0.2985 70.005 6.7044
(18.825) (0.1920) (84.846) (12.3777)

FIRMAGE 0.162 0.0017 0.022 0.0629
(0.830) (0.0085) (4.088) (0.4325)
BRANCH -31 .971 0.3272 499.822 4 3.1926
(38.973) (0.3975) (570.305) (41.6823)

OECD 25.286 0.2588 14.482 1.1692
(17.612) (0.1796) (64.765) (9.0899)

CONSTANT 482.764'" .4941 1’” -570.61 1 -3.7640
(75.976) (0.7750) (367.028) (42.8390)

0 1.0960 1.5004
Observations 2.305 2.305
Log-Likelihood value 466.699 461.435
Pseudo R2 0.076

 

 

 

Note: All the estimates are reported after multiplying by 100 to avoid the confusion from rounding. The values in
parentheses below the Homoskedastic Tobit estimates are the standard errors, adjusted for cluster effect. The values in

.0. .0

parentheses below the Heteroskedastic Tobit estimates are the standard errors, not adjusted for cluster effect. , ,

and ' denote significance at the 1%, 5%, and 10% levels respectively.

Heteroskedastic Tobit is about 0.0158.

97

The computed adjustment factor for

 

Table 22. Heteroskedastic Tobit Estimates of Tariff Evasion for Raw Materials & Fuels

 

 

 

 

 

 

 

 

Dependent variable: E MSION
Independent Homoskedastic Tobit Heteroskedastic Tobit
variables B x 100 B CD(xB/o) x 100 B x 100 B ¢(xB/o) x 100
TARIFF 1.729? 0.0111" 1.337 -0.1385
(0.720) (0.0046) (1.873) (0.1923)
DETECTION 0.299 0.0019 0.737 0.0274
(0.348) (0.0022) (2.524) (0.1934)
AUDIT 0.541 0.0035 -2.098 0.0811
(0.506) (0.0032) (3.183) (0.2306)
LOG(IMPORT) -2.915 -0.0188 4.416 0.0642
(2.573) (0.0165) (13.902) (1.0008)
EXCHANGE 8.322“ 0.0536" -2.346 0.6322
(3.267) (0.0209) (27.737) (1.9918)
NONPROFIT 37.619 0.2422 -77.581 -7.1104
(34.594) (0.2214) (1885.730) (102.3258)
INDIVIDUAL 33.053” 0.2128" 43.172 -3.9426
(15.082) (0.0965) (143.443) (7.1280)
FIRMAGE 0.754‘ 0.0049’ 4.791 0.0887
(0.454) (0.0029) (1.625) (0.1849)
BRANCH 44.428 0.0929 12.980 0.0430
(14.282) (0.0914) (51.955) (4.8545)
OECD 9.944 0.0640 03.320 -7.3637
(1 1.947) (0.0765 ) (81.611) (11.8204)
CONSTANT -334.621”‘ -2.1547"‘ 410.418 0.1385
(69.069) (0.4420) (470.943) (6.7234)
a 0.8141 0.2009
Observations 4.799 4.799
Log-Likelihood value -204.36O -l97.485
Pseudo R2 0.057

 

 

 

Note: All the estimates are reported after multiplying by 100 to avoid the confusion from rounding. The values in
parentheses below the Homoskedastic Tobit estimates are the standard errors, adjusted for cluster effect. The values in
arentheses below the Heteroskedastic Tobit estimates are the standard errors, not adjusted for cluster effect. m, ”. and
denote significance at the 1%, 5%. and 10% levels respectively. The computed adjustment factor for Heteroskedastic

Tobit is about 0.0005.

98

 

Table 23. Heteroskedastic Tobit Estimates of Tariff Evasion for Capital Goods

 

 

 

 

 

 

 

 

Dependent variable: E VASION
Independent Homoskedastic Tobit Heteroskedastic Tobit
variables 13 x 100 13 <D(xB/o) x 100 B x 100 13 <D(xB/o) x 100
TARIFF 245'?" 0.0117'" 1.587 0.0119
(0.772) (0.0036) (2.215) (0.1424)
DETECTION 0.170 0.0008 0.161 0.0002
(0.453) (0.0021) (2.151) (0.0961)
A UDIT 0.810 -0.0038 0.759 0.0926
(0.662) (0.0031) (1.897) (0.1318)
LOG(IMPORT) 0.601 0.0029 4.997 0.0641
(2.716) (0.0128) (17.837) (0.9623)
EXCHANGE 4.864 0.0231 5.992 0.1895
(3.221) (0.0151) (18.014) (0.8966)
NONPROFIT 59.038 0.2803 -53.729 -3.9493
(45.869) (0.2156) (4402.850) (142.8151)
INDIVIDUAL 45.595 0.0740 432.547 4.5809
(25.586) (0.1203) (237.817) (10.1305)
FIRMAGE 0.494 0.0023 4.882 0.0711
(0.495) (0.0023) (3.875) (0.1831)
BRANCH 41.298 0.0536 -28.260 0.4649
(13.819) (0.0650) (152.349) (6.4055)
OECD 37.034‘ 01758 407.286 -8.2191
(19.536) (0.0918) (76.504) (7.5435)
CONSTANT -369.878"° 4 .7561 482.696 0.4804
(68.312) (0.3211) (278.052) (6.1346)
a 1.0060 0.4497
Observations 6.591 6.591
Log-Likelihood value -224.849 -215.950
Pseudo R2 0.041

 

 

 

Note: All the estimates are reported after multiplying by 100 to avoid the confusion from rounding. The values in
parentheses below the Homoskedastic Tobit estimates are the standard errors, adjusted for cluster effect. The values in
arentheses below the Heteroskedastic Tobit estimates are the standard errors, not adjusted for cluster effect. M, ,and
denote significance at the 1%. 5%, and 10% levels respectively. The computed adjustment factor for Heteroskedastic
Tobit is about 0.0001.

99

 

Table 24. Different Elasticity of Evasion across Commodities

 

 

 

Classification Oct/at t/a 30,,
Consumer Goods 0.000025 4109.84 (=25.07/0.0061) 0.102
Raw Materials 0.000111 7804.35 (=17.95/0.0023) 0.852
Capital Goods 0.000117 8957.89 (=17.02/0.0019) 1.026

 

 

 

 

100

 

Table 25. Elasticity of Evasion with Respect To the Foreign Exchange Rate

 

 

 

 

Classification Tariff Elasticity Exchange Elasticity
Consumer Goods 0.102 4.010
Raw Materials 0.852 3.163
Capital Goods 1.026 1 .642
Overall 0.114 3.162

 

 

 

101

 

CHAPTER VI

THE EMPIRICAL ANALYSIS OF CUSTOMS FRAUD

1. Empirical Framework

I use a similar econometric model explaining tariff-evasion behavior to examine the

major determinants of customs fraud as follows.

FRA UD. = B, + B. TARIFF,- + B2 DETECTION, + B3 A UDIT,- + B. RAWFUELS,-
+ B5 CAPGOODS,- + B. LOG(IMPORT)) + B EXCHANGE,- + £8 NONPROFIT,-

+ Bo INDIVIDUALi + B... F1RMAGE.+ .611 BRANCH,- + [3.2 OECD, + u), (56)

where F RA UD is a dummy variable set to 1 if import declaration is involved in customs
fraud. For the other variables, I use the same data as in the estimation of tariff evasion in
the previous chapter. Equation (29) describes the major determinants of customs fraud,
where i denotes each import declaration.

One problem with using a Tobit model to estimate tariff evasion in the preceding
chapter is that we cannot capture all types of customs fraud, other than the case where
import declarations are involved in actual evasion of tariff payments. For some imported
goods, the importers must have a permit or approval relating to the importation of goods
under laws and regulations other than the Customs Law. Importers may avoid these
regulations by declaring false items. In addition, some importers tend to be wrongly
benefited by falsely or not marking the country of origin, which may lead to a confusion

or misunderstanding among ultimate consumers. The effects of these customs frauds are

102

not directly measured in terms of the amount of tariff evaded or money value.
Thus, accounting for other types of customs fraud, 1 also estimate the behaviors of
importers that are associated with committing customs fraud, using a binary dependent

model, such as the following Probit model.

yr. = XIB + 11;
y,- = 1 (customs fraud) if y,’ > 0

y,- = 0 (compliance) otherwise, (57)
where customs fraud (y, = 1) accounts for all import declarations involving the violation

of Customs Law as well as other regulations, which include false description of imported

goods to avoid the import regulations, false marking of the country of origin, etc.

103

2. Data Description

Except for the dependent variable, I use the same data set used in the estimation of
tarifl evasion. Table 26 provides major types of customs fraud occurring at the time of
import.

The total number of customs frauds detected after random inspection in 1998 is 1,475.
The most frequent type of customs fraud is the violation of the country-of-origin rule,
which includes differences in country of origin between actual marks on goods and those
on declaration forms, and false marking of country of origin (681 observations).

It should be noted that the total number of customs frauds includes declaration of
wrong items, which, to my knowledge, neither affects the tariff payment nor is related to
any of the import restrictions or regulations (299 observations). These may be regarded
as minor mistakes, which may not be serious violations of the Customs Laws and other
relevant regulations. Sometimes, however, these mistakes cause serious problems in the
calculation of trade statistics for imports by item and by country. These incorrect trade
statistics can lead to inappropriate international trade policies, or to wrong decision-
makings for business. In this respect, the KCS has put much emphasis on the need for
correct declaration of commodity description and HS code to collect reliable and useful

trade statistics. Thus, we include these figures in the estimation of customs fiaud.

104

3. Empirical Results

3-1. Total Estimation

3-1-1. Estimation Results

I estimate the Probit model for customs fraud committed by importers. For
comparison, OLS and Logit estimates are reported along with those of the Probit in Table
27. Since the magnitudes of the coefficients are not directly comparable across the
models, I derive the partial derivatives comparable to the OLS estimates. The effects of
the variables x,- on the response probabilities P(y=1 Ix) are estimated at the average values
of the independent variables in the sample. The scaled coefficients for Logit and Probit
are also reported in Table 27.

The estimates from the three models are the same in their signs, and the same
variables are statistically significant in all models. In addition, the OLS estimates, the
partial effects in the Logit model, and the partial effects in the Probit model are very
similar in their magnitudes. Comparing with estimates of the Tobit model for tariff
evasion, we find some interesting results. The effective tariff rates are found to have no
effects on customs fraud, while they are significant in the tariff-evasion equation. It is
not surprising that fraud is not affected by tariff rates, since fraud is ofien motivated for
reasons other than tariff evasion. It is predicted that an increase in the probability of
detection has deterrent effects on fraud, while they are not statistically different from zero

in the tariff-evasion model. More detailed interpretations and implications of these

105

estimates are discussed in a later section.

As with the estimation of tariff evasion in the previous chapter, the standard errors of
the above estimates are adjusted for cluster sampling, allowing that the observations are
independent across firms (clusters), but not necessarily independent within firms
(clusters). However, even if we assume that all observations are independent with each
other so that there is no clustering effect, it does not make much difference in the
standard errors, and the significance levels of the coefficients do not change. The
estimation results without considering cluster effect are reported in table 28 for

comparison.

3-1-2. Tests for Homoskedasticity

To test the hypothesis of homoskedasticity against the alternative hypothesis of
heteroskedasticity in the Probit and Logit model, both the LM and LR statistics are used.
The more general model (unrestricted model) is assumed to be the model with
multiplicative heteroskedasticity in which Var(u) = [exp(z§)]2. The test results for
homoskedasticity are presented in Table 29. Based on the LR and LM statistics, the
hypothesis of homoskedasticity in the Probit can be rejected. In addition, we can also
reject the assumption of homoskedasticity in the Logit based on these two statistics. In
contrast to the test statistics for the Tobit model in the previous chapter, the LM statistics
are much smaller than the LR statistics. Considering the general notion that the outer
product of the score LM statistic has severe size distortions, it is interesting that LM

statistics are much smaller than LR statistics.

106

We can compare the estimates of the standard Probit (Logit) model to those of the
Probit (Logit) model with multiplicative heteroskedasticity in order to determine the
extent to which heteroskedasticity in the latent variable model affects the partial
derivatives. As shown in Table 30 (for the Probit) and Table 31 (for the Logit), the
effects of heteroskedasticity on B are extremely large in both models. However,
comparing the effects of the variables X} on the response probabilities P(y=1 Ix), they are
not much different. Especially, the presence of heteroskedasticity does not have huge
impacts on the partial effects in the Probit model. Except for TARIFF and FIRMAGE,
there are no huge differences between the partial effects of explanatory variables in the
homoskedastic Probit model and those in the heteroskedastic Probit model. For the Logit
model, the partial effects in the heteroskedastic Logit model are not far from those in the
standard Logit model either, except for a few variables, such as TARIFF. In addition, the
introduction of the more complicated model (heteroskedastic Probit and Logit Model)
does not improve the percentage correctly predicted. Instead, due to the large standard
errors, all the partial effects in the heteroskedastic Probit and Logit models are not
statistically different from zero.

Despite test results from the LM and LR statistics, we may not reject the hypothesis,
since the effects of heteroskedasticity on the partial effects are not so huge. Based on this

reasoning, I mainly focus on the estimation results of the standard Probit model hereafter.

107

3-1-3. Implications

Tarifl and Enforcement Variables

DETECTION and AUDIT are found to be significantly different from zero at the 5%
level, while TARIFF is insignificant.

Unlike the estimation results of tariff evasion, in which an increase in tariff rates has
significant effects on tariff-evasion behavior, the coefficients are not statistically different
from zero. The intuition behind this difference in the statistical significance of effective
tariff rates is that the importer who commits customs fraud can be wrongly benefited by
misleading consumers and avoiding other regulations, rather than by evading actual tariff
payments. As noted in the previous section, customs fraud includes not only tariff
evasion, but also falsely marking or not marking the country of origin, avoidance of
regulations, infringement of trademark, etc. Thus, from the types of customs fraud, it is
obvious that importers who commit fraud have different incentives compared with those
who are attempting to evade tariffs.

The Probit estimate shows that, unlike the Tobit estimate in tariff evasion, an increase
in the probability of detection has statistically significant effects on customs fraud. As
the detection rate increases by 1% point, the probability of committing customs fraud
decreases by 0.0012. Thus, an increase in the detection rate has deterrent effects.
However, the effect is relatively small, since the average of the dependent variable,
FRAUD, is 0.1043. The elasticity of customs fraud with respect to the detection rate is

-0.133 at the average value of both variables. Thus, for the importer with an average

108

detection rate for the sample, 11.865%, a one-percent increase in the detection rate would
result in an expected 0.133% decrease in the probability of committing customs fraud. In
other words, if the detection rate were to increase from 11.865% to 11.984%, these
estimates suggest that the probability of customs fraud would decrease from 0.1043 to
about 0.1042.

The Probit estimate for AUDIT shows that, as the inspection rate increases by 1%
point, customs fraud decreases by 0.0020. As with DETECTION, considering the
average value of the probability of committing customs fraud, practically, an increase in
the probability of inspection has only a small deterrent effect on fraud. The elasticity of
customs fraud with respect to the audit rate is -0.230 at the average value of both
variables. Thus, for the importer with an average audit rate for the sample, 11.994%, a
one-percent increase in the penalty rate would result in an expected 0.230% decrease in
the probability of committing customs fraud. In other words, if the audit rate were to
increase from 11.994% to 12.114%, these estimates suggest that the probability of

customs fraud would decrease from 0.1043 to about 0.1041.

Commodity Categories

All dummy variables for commodity categories are significantly different from zero
with negative signs. Thus, in most cases, it is predicted that Raw Materials & Fuels and
Capital Goods are less likely to be the target of customs fraud, compared with Consumer
Goods. In particular, imported goods classified in Cereal, Direct Consumer Goods,

Durable Consumer Goods, or Non-Durable Consumer Goods have a higher probability of

109

committing fraud compared with the other imported goods.

It may be the case that Cereal (e.g., wheat, rice, corn) and agricultural products in
Direct Consumer Goods (e.g., garlic, sesame seed) are goods for which it is relatively
easy to disguise their true nature or to undervalue quantity or weight. Durable or Non-
Durable Consumer Goods, such as domestic electro-machines, household appliances,
etc., tend to be the target of false marking of origin, as well as declaration of false items.
For these consumer goods, the country of origin is one of the most important factors in
attracting consumers. It may also have great impact on consumer price. Thus, some
importers may well attempt to be wrongly benefited by falsely marking or not marking
the country of origin on these Durable or Non-Durable Consumer Goods, which leads to
a confusion or misunderstanding among the ultimate consumers.

For example, in many cases, the Foreign Trade Act of Korea requires each imported
good to be marked in a conspicuous place as legibly, indelibly, and permanently, and the
country of origin should appear preceded by “made in”, “product of”, or other words of
similar meaning. However, some importers would mark with adhesive stickers or tags, in
attempt to purposely destroy or alter the country of origin after the customs clearance.
They can also mark the origin in a manner intentionally leading to confusion among
consumers, using obscure words, such as “stylized in”. In this case, the name of country

preceded by “stylized in” is different from true country of origin.45

 

‘5 For example, consumers who buy cloth think that they are imported from Italy, when “stylized
in Italy” is marked on each cloth in a place easily seen, from which true country of origin is
different.

110

F irm-Specific Variables

The coefficient on the amount of imports is not statistically different from zero. Thus,
the probability of committing customs fraud is not affected by the amount of imports.
Compared with for-profit corporations, it is predicted that importing individuals have a
higher probability of committing customs fraud. Years after the establishment of the firm
are shown to be negatively related to the probability of fraud. Thus, when the firm has a
long history of its business, it is less likely to commit fraud at the time of import.46 In
addition, compared with their headquarters, branches have a lower probability of

committing fraud.

Country of Origin

The country-of-origin variable has significant effects on customs fraud. This implies

that goods originated from or produced in developed countries (OECD countries) are

more likely to be the target of fraud.

Foreign Exchange Rate

The coefficient on the foreign exchange rate is found to be significantly different from

 

‘6 In the previous chapter, for Raw Materials & Fuels, the years-after—establishment of the firm
has a positive effect on evasion, which is a counter-intuitive result. As with the estimation results
here, it is more likely that the history of the firm is negatively related to the prospensity to
commit customs fraud as well as to evasion.

lll

zero. That is, a one-hundred Korean won increase in the exchange rate of won to dollar
increases the probability of customs fraud by 0.025. The elasticity of customs fraud with
respect to the foreign exchange rate is 3.309 at the average value of both variables. Thus.
for the importer with an average foreign exchange rate for the sample, W 1,380.414 per
US. dollar, a one-percent increase in the foreign exchange rate would result in an
expected 3.309% increase in the probability of committing customs fraud. In other
words, if the exchange rate were to increase from W1 ,380.414 to W1,394.218, these
estimates suggest that the probability of customs fraud would increase from 0.1043 to

about 0.1078.

112

3-2. Estimation by Commodity Categories

3-2-1. Estimation Results

I divide the imported goods into 3 categories (Consumer Goods, Raw Materials &
Fuels, and Capital Goods) and estimate a Probit model for each of the categories
separately. The Probit estimates for each of the commodity categories are given in Table
32.

When we compare the signs and significances of the estimates across commodity
categories, most variables are consistent in their signs and significance. In particular,
DETECTION, AUDIT, and EXCHANGE are significant in all commodity categories and
have the same signs across the categories. It is shown that an increase in the probability
of detection given audit is predicted to have significant deterrent effects on customs fraud
in all categories. The probability of audit also has significant deterrent effects on fraud.
In addition, an increase in the exchange rate would induce more customs fraud in all
categories.

The standard errors of Probit estimates are also adjusted for cluster sampling, allowing
that the observations are independent across firms (clusters), but not necessarily
independent within firms (clusters). However, the standard errors are very similar to
those computed without accounting for cluster effect. The estimation results without

considering the cluster effect are also reported in table 33 for comparison.

”3

3-2-2. Tests of Homoskedasticity

Both the LM and LR statistics are used to test the hypothesis of homoskedasticity
against the heteroskedasticity in the Probit model by commodity categories. As shown in
Table 34, based on the LR and LM statistics, the hypothesis of homoskedasticity can be
rejected in all categories at the 95% level.

However, for Consumer Goods, the hypothesis cannot be rejected at the 99% level
with the critical value of 23.21 based on LM statistic. In addition, the statistical tests may
not be sufficient to determine whether the homoskedasticity assumption is satisfied.
Thus, we compare the estimates of the standard Probit model to those of the Probit model
with multiplicative heteroskedasticity to examine the effects of heteroskedasticity in the
latent variable model on the partial derivatives. As presented in Table 35, 36, and 37, the
effects of heteroskedasticity on B seem to be extremely large in all commodity groups.
However, comparing the effects of the independent variables on the response
probabilities, they are not much different except in the case of Consumer Goods.47
Especially, there are no huge differences in the partial effects of our key variables, such
as TARIFF, PENALTY, and AUDIT in Raw Materials & Fuels and Capital Goods.
Despite test results from the LM and LR statistics at the 95% level, we may not reject the
hypothesis, since the effects of heteroskedasticity on the partial effects are not so large.
Based on this reasoning, I mainly focus on the estimates of the standard Probit to

interpret the estimation results.

 

‘7 The estimates (B‘and g(xBA)B”) in the heteroskedastic Probit model for Consumer Goods may not
be reliable, since there is abnormal exit from initial iterations during the estimation. However,
based on the LM statistic at the 99%, the homoskedasticity assumption can still be supported.

114

3-2-3. Implications

Tarifl and Enforcement Variables

The coefficients on effective tariff rates are found to be significantly different from
zero at the 10% level in Raw Materials & Fuels and Capital Goods, while insignificant in
Consumer Goods. Thus, as the tariff rate increases by 1% point, the probability of
committing customs fraud increases by 0.0016 in Crude Materials & Fuels and 0.0029 in
Capital Goods. The elasticities of customs fraud with respect to the tariff rate in Raw
Materials & Fuels and Capital Goods are 0.299 and 0.514, respectively. Thus, for
importers of Raw Materials & Fuels, if the tariff rate were to increase from 17.954% to
18.124%, these estimates suggest that the probability of committing customs fraud would
increase from 0.096 to about 0.0963. For Capital Goods, if the tariff rate were to increase
from 17.015% to 17.185%, these estimates suggest that the probability of committing
customs fraud would increase from 0.096 to about 0.0965.

The Probit estimate for DETECTION shows that an increase in the probability of
detection has deterrent effects on customs fraud in all commodity groups. A 1% point
increase in the detection rate is predicted to decrease the probability of committing
customs fraud by 0.0021 in Consumer Goods, 0.0007 in Crude Materials & Fuels and
0.0012 in Capital Goods. Thus, an increase in the detection rate can be a relatively
effective tool to deter fraud for Consumer Goods, although the overall deterrent effects of
the detection rate are small. The elasticities of customs fraud with respect to the penalty

rate in Consumer Goods, Raw Materials & Fuels, and Capital Goods are -0.207, -0.075,

115

and -0.152, respectively. Thus, for importers of Consumer Goods, if the detection rate
were to increase from 14.186% to l4.328%, these estimates suggest that the probability
of committing customs fraud would decrease from 0.144 to about 0.1437. For importers
of Raw Materials & Fuels, if the detection rate were to increase from 10.276% to
10.379%, these estimates suggest that the probability of committing customs fraud would
decrease from 0.096 to about 0.0959. For Capital Goods, if the detection rate were to
increase from 12.210% to 12.332%, these estimates suggest that the probability of
committing customs fraud would decrease from 0.096 to about 0.0959.

Like the probability of detection, the probability of inspection also has deterrent
effects on customs fraud. The Probit estimate for AUDIT shows that as the inspection
rate increases by 1% point, customs fraud decreases by 0.002] in Consumer Goods,
0.0022 in Crude Materials & Fuels and 0.0024 in Capital Goods. The effects of this
instrument on fraud are very similar across commodities. The elasticities of customs
fraud with respect to the inspection rate in Consumer Goods, Raw Materials & Fuels, and
Capital Goods are -0.311, -0.252, and -0.237, respectively. Thus, for importers of
Consumer Goods, if the audit rate were to increase from 21.307% to 21.520%, these
estimates suggest that the probability of committing customs fraud would decrease from
0.144 to about 0.1436. For importers of Raw Materials & Fuels, if the audit rate were to
increase from 10.975% to 11.085%, these estimates suggest that the probability of
committing customs fraud would decrease from 0.096 to about 0.0958. For Capital
Goods, if the audit rate were to increase from 9.478% to 9.573%, these estimates suggest
that the probability of committing customs fraud would decrease from 0.096 to about

0.0958.

116

F irm-Specific Variables

For the firm-specific variables, we can find that there are some differences in the
statistical significance and signs of the independent variables across commodity
categories.

The amount of import is found to have significant effects on customs fraud in Raw
Materials & Fuels and Capital Goods, while its effect is insignificant in Consumer Goods.
However, the amount of imports is negatively related to the probability of committing
customs fraud in Raw Materials & Fuels, while positively related in Capital Goods.

The Probit estimates show that individuals are more likely to be involved in fraud only
for Raw Materials & Fuels, compared with the for-profit corporations and non-profit
organizations. For Consumer Goods and Capital Goods, there are no differences in the
probability of committing fiaud between the individuals and the for-profit corporations.
In any of the categories, the possibilities of customs frauds of the non-profit organizations
are not statistically different from those of the for-profit corporations.

The number of years after establishment of the firm has significant effects on fraud in
all commodities. The negative coefficient of this variable implies that firms with
relatively long histories have a smaller tendency to commit illegal activities when they
import.

In addition, compared with their headquarters, branches have a lower probability of

committing fraud in all commodity categories.

117

Country of Origin

Except in the case of Consumer Goods, the country of origin has significant effects on
customs fraud. Probit estimates for this variable show that items classified in Raw
Materials & Fuels or Capital Goods originated from or produced in one of the OECD

countries are more likely to be involved in illegal activities at the time of import.

Foreign Exchange Rate

In all commodity groups, the foreign exchange rates are significantly different from
zero. That is, a one-hundred Korean won increase in the exchange rate of won to dollar
increases the probability of committing customs fiaud by 0.050 in Consumer Goods, by
0.015 in Raw Materials & Fuels, and by 0.022 in Capital Goods. Thus, as the unit import
price increases, firms are more likely to commit fraud to make compensation for their
additional financial burden. The elasticities of customs fraud with respect to the foreign
exchange rate in Consumer Goods, Raw Materials & Fuels, and Capital Goods are 4.802,
2.158, and 3.160, respectively. Thus, for the importer with an average foreign exchange
rate for the sample, a one-percent increase in the foreign exchange rate would result in an
expected 4.802% increase in the probability of committing customs fraud in Consumer
Goods, a 2.158% increase in Raw Materials & Fuels, and a 3.160% increase in Capital

Goods.

118

Table 26. Types of Customs Fraud

 

 

 

Type Reason Method (related to type)
o Undervaluation - To evade customs duty Reduction of the quantity or unit
price
0 False Declaration of Items - To claim lower duty rates False description of items
- To claim lower special excise taxes False declaration of HS code

- To avoid import regulations

 

 

 

0 Violation of Country-of - To claim lower duty rates Not marking of origin
- Origin Rule - To deceive final consumers False marking of origin
- To avoid import regulations False declaration of origin
o lnfrmged Trademarks - To deceive final consumers False marking of trade mark

 

 

 

119

 

Table 27. OLS, Logit, and Probit Estimates of Customs Fraud with Cluster Effect

 

Dependent variable: FRAUD

 

 

 

 

 

 

 

Independent Logit Probit
variable OLS x 100 B" x 100 g(xB‘)B" x 100 13" x 100 g(xtl“‘)[3" x 100
TARIFF 0.0171 0.147 0.0121 0.098 0.0163
(0.0158) (0.115) (0.0094) (0.069) (0.01 14)
DETECTION 0.1 146‘" 4 .423'” 0.1 170’” 0.708‘“ -0.1 173'"
(0.0176) (0.275) (0.0226) (0.133) (0.0223)
A UDIT 0.1983'" -2415‘” 0.1985‘" 4 .195'” -0. I98 I
(0.0197) (0.318) (0.0261) (0.148) (0.0247)
RAWFUELS -5.6715‘” -55585‘” 4.5672'” -28.624“' 4.7441'”
(0.9263) (8.654) (0.7112) (4.641) (0.7l06)
CAPGOODS -5.4144“' -51599'" 42397’” -27541 4.5647‘”
(0.8826) (8.238) (0.6770) (4.433) (0.7150
LOG(IMPORT) 0.1492 4 .593 0.1309 0.913 -0.15l4
(0.1424) (1.594) (0.1310) (0.840) (0.140)
EXCHANGE 2.8096’” 27.432‘“ 2.2540‘” 14.959‘“ 2.4793'"
(0.2724) (2.108) (0.1733) (1.160) (0.2011)
NONPROFIT 2.0527 18.206 1.4959 9.506 1.5755
(2.4502) (26.266) (2.1586) (13.851) (2.5806)
INDIVIDUAL 3.3633‘" 24.208'" 1.9891 13.373'" 2.2164'”
(1.0385) (9.079) (0.7461) (4.892) (0.9285)
FIRMAGE 0.1480‘” 4.698‘" 0.1396‘” 0.834‘” 0.1382‘”
(0.0317) (0.420) (0.0345) (0.209) (0.0331)
BRANCH 4.3147” 44.832‘" -3.6837”' -22.561“‘ -3.7392'”
(0.8037) (11.376) (0.9349) (5.607) (0.7826)
OECD 2.4948‘” 30.516‘" 2.5074‘“ 16.226‘” 2.6893‘”
(0.5875) (7.315) (0.6012) (3.803) (0.5603)
CONSTANT 47.8058‘“ 492.369‘“ 40.4564'” 279.999'” _
(3.9747) (33.414) (2.7460) (18.072) _
Observations 13.695 13.695 13.695
Clusters 5.713 5.713 5.713
Correctly Predicted _ 89.57% 89.57%
Log-Likelihood _ 4326.393 4328.337
Pseudo R2 0.037 0.056 0.055

 

 

 

 

Note: All the estimates are reported after multiplying by 100 to avoid the confusion from rounding. The partial effects
of the dummy variables in the Logit and the Probit model are computed at the mean values. The values in parentheses
below the estimates are the standard errors, adjusted for cluster sampling. m
5%, and 10% levels respectively. The pseudo R-squared is the usual R2 for the OLS and the measure based on the log
likelihoods for Logit and Probit.

120

, and ' denote significance at the 1%,

 

Table 28. OLS, Logit, and Probit Estimates of Customs Fraud without Cluster Effect

 

Dependent variable: FRAUD

 

 

 

 

 

 

 

Independent Logit Probit
variable OLS x 100 a“? 100 g(xB’)l3" x 100 (3’3 100 g(xl3")11' x 100
TARIFF 0.0171 0.147 0.0121 0.098 0.0 I 63
(0.0148) (0.137) (0.0113) (0.079) (0.0132)
DETECTION 0.1 146'" 4 .423'“ 0.1 170‘" 0.708'“ -0.1 173‘“
(0.0176) (0.260) (0.0212) (0.127) (0.0210)
AUDIT 0.1983’” -2.415"‘ 0.1985'” 4.195‘” -0. 1981'”
(0.0182) (0.256) (0.0206) (0.124) (0.0203)
RAWFUELS -5.6715"‘ -55.585"’ 4.5672'” -28.624"‘ 4.7441’“
(0.8829) (8.244) (0.6741) (4.414) (0.7306)
CAPGOODS -54144'" -51 .599‘” 4.2397‘” -27.541‘” -4.5647'”
(0.8602) (7.878) (0.6449) (4.252) (0.7037)
LOG(IMPORT) 0.1492 4.593 0.1309 0.913 -0. 1514
(0.1343) (1.526) (0.1254) (0.796) (0.1319)
EXCHANGE 2.8096'“ 27.432'” 2.2540'” 14.959'“ 2.4793‘”
(0.2462) (1.988) (0.1595) (1.088) (0.1783)
NONPROFIT 2.0527 18.206 1.4959 9.506 1.5755
(2.3745) (25.935) (2.1310) (13.597) (2.2536)
INDIVIDUAL 3.3633‘” 24.208‘” 1.9891'” 13.373‘” 2.2164‘"
(1.0269) (8.831) (0.7262) (4.810) (0.7977)
FIRMAGE 0.1480‘” 4 .698‘“ 0.1396’” 0.834‘” 0.1382”
(0.0240) (0.285) (0.0232) (0.143) (0.0236)
BRANCH .3.3147"‘ 44.832‘” -3.6837°" -22561'” -3.7392“'
(0.6434) (8.875) (0.7227) (4.395) (0.7253)
OECD 2.4948‘“ 30.516'” 2.5074'” 16.226‘" 2.6893'"
(0.5593) (6.758) (0.5524) (3.496) (0.5777)
CONSTANT 47.8058‘” 492.369‘” 40.4564‘“ 279.999‘” .—
(3.6676) (31.1682) (2.5248) (16.820) _
Observations 13,695 13,695 13,695
Clusters 5.713 5.713 5.713
Correctly Predicted — 89.57% 89.57%
Log-Likelihood _ -4326.393 4328.337
Pseudo R2 0.037 0.056 0.055

 

 

 

 

Note: All the estimates are reported after multiplying by 100 to avoid the confusion from rounding. The partial effects
of the dummy variables in the Logit and the Probit model are computed at the mean values. For the OLS, the values in
parentheses below the estimates are heteroskedasticity-robust standard errors. For Logit and Probit, the values in

parentheses are the usual standard errors.

, ”. and ' denote significance at the 1%, 5%, and 10% levels respectively.

The pseudo R-squared is the usual R2 for the OLS and the measure based on the log likelihoods for Logit and Probit.

121

 

Table 2‘.

T7351?
. P- 1
'1 Critic;
l Test

distributi

Table 29. Tests for Homoskedasticity in Probit and Logit

 

Homoskedasticity

 

 

 

 

Probit Logit
LM test LR test LM test LR test
Test statistic 93.619 240.460 99.490 253.588
P- value 0.000 0.000 0.000 0.000
Critical value 21.026 21.026 21 .026 21.026
Test result Rejected Rejected Rejected Rejected

 

 

 

 

 

Note: The critical values for LM and LR test are derived from
distribution with 12 degrees of freedom.

122

the 95th percentile in the chi-squared

 

Table 30. Heteroskedastic Probit Estimates of Customs Fraud

 

 

 

 

 

 

 

 

Dependent variable: F RA UD
Independent Homoskedastic Probit Heteroskedastic Probit
Variables [3’ x 100 G(x13")p“ x 100 B” x 100 g(xl3’)lf‘ x 100
TARIFF 0.098 0.0163 0.246 0.1670
(0.069) (0.01”) (4.233) (0.1667)
DETECTION -6703“ 0.1 173'” 47.444'” 0.0707
(0133) (0.0223) (1 1.949) (0.1937)
A UDIT 4 .195‘” -0. 198 I -72.250‘” 0.2094
(0.148) (0.0247) (20.690) (0.3707)
RAWFUELS -28624‘” 4.7441 -206480“ 4.8140
(4.641) (07 l06) (102.306) (2.3792)
CAPGOODS -27.541 45647” 414.288 4 .7490
(4.433 ) (0.7151) (86.367) (2.1027)
LOG(IMPORT) 0.913 0.1514 -2939 0.1930
(0.840) (0.140) (17.709) (0.3384)
EXCHANGE 14.959‘“ 2.4793‘” .21.993 2.5041
(1.160) (0.2011) (37.656) (2.3329)
NONPROFIT 9.506 1.5755 452.836 1.5820
(13.851) (25806) (542.609) (9.4005)
INDIVIDUAL 13.373'" 2.2164‘“ 153.556" 4 .8205
(4.892) (0.9285) (77.308) (1.9238)
FIRMAGE 0.834‘" 0.1382'” 41.328” 0.0234
(0.209) (0.0330 (5.036) (0.0861)
BRANCH -22.561‘” -3.7392‘” -821657'” -22299
(5.607) (0.7826) (310.276) (5.4526)
OECD 16.226'” 2.6893'” 30.590 3.0027
(3.803) (0.5603) (63.175) (2.8549)
CONSTANT 279.999‘” _ ~78.513 0.6738
(18.072) _ (454.502) (4.6681)
Observations 1 3,695 1 3,695
Correctly Predicted 89.57 % 89.57 %
Log-Likelihood value 4328.337 4208.107

 

 

 

 

Note: All the estimates are reported after multiplying by 100 to avoid the confusion from rounding. The partial effects
of the dummy variables are computed at the mean values. The partial effect of the Heteroskedastic Probit model,
g(xB§BA, is equal to 6P(y=1 lx) / 6x, = ¢[xB/exp(z5)]x[(Bj-bijyexp(zb)]. The values in parentheses below the
Homoskedastic Probit estimates are the standard errors, adjusted for cluster effect. The values in parentheses below the
Heteroskedastic Probit estimates are the usual standard errors. m, ", and ' denote significance at the 1%. 5%, and 10%
levels respectively.

123

Table 31. Heteroskedastic Logit Estimates of Customs Fraud

 

 

 

 

 

 

 

 

Dependent variable: F RA UD
Independent Homoskedastic Logit Heteroskedastic Logit
variables B” x 100 g(xB')Bi< 100 B” x 100 g[xB7/exp(zb)]B"x100
TARIFF 0.147 0.0121 0.750 0.0021
(0.115) (0.0094) (10.111) (0.0426)
DETECTION 4 .423‘“ 0.1 170‘“ - 102.871 0.2846
(0.275) (0.0226) (33.916) (3.8521)
AUDIT -2415‘” 01985” -200278‘" 0.5541
(0.318) (0.0261) (59.601) (7.4953)
RAWFUELS -55.585”' 4.5672‘“ 435.111’ 4.2037
(8.654) (0.7112) (233.040) (16.2559)
CAPGOODS -51.599"' 4.2397‘” -245.695 0.6797
(8.238) (0.6770) (200.143) (9.1611)
LOG(IMPORT) 4.593 0.1309 —6.512 0.0180
(1.594) (0.1310) (41.578) (0.2469)
EXCHANGE 27.432‘” 2.2540‘“ -65.342 0.1808
(2.108) (0.1733) (93.232) (2.2973)
NONPROFIT 18.206 1.4959 4342.096 4.7129
(26.266) (2.1586) (1461.849) (50.3036)
INDIVIDUAL 24.208‘” 1.9891‘“ 349.80‘ 0.9677
(9.079) (0.7461) (181.314) (13.1295)
FIRMAGE 4.698‘" 0.1396’” -20.091 ' 0.0556
(0.420) (0.0345) (10.951) (0.7522)
BRANCH 44.832‘” -3.6837"‘ -2224.082“‘ -6.1530
(11.376) (0.9349) (847.899) (83.2537)
OECD 30.516’” 2.5074'“ 52.807 0.1461
(7.315) (0.6012) (148.916) (2.0784)
CONSTANT 492369” 404564“ 16.888 0.0467
(33.414) (2.7460) (1 132.738) (2.6594)
Observations 1 3 .695 13.695
Correctly Predicted 89.57 % 89.57 %
Log-Likelihood value -4326.393 4199.599

 

 

 

 

Note: All the estimates are reported after multiplying by 100 to avoid the confusion from rounding. The partial cm
of the dummy variables are computed at the mean values. The values in parentheses below the Homoskedastic Logit
estimates are the standard errors, adjusted for cluster effect. The values in parentheses below the Heteroskedastic Logit
estimates are the usual standard errors. m, ", and ° denote significance at the 1%. 5%, and 10% levels respectively.

124

Table 32. Prob

fr.
1

lndependem
l variables

l 11le F

i
l
‘ DETECTIO?)

.4 1 DH

 

1

l

LOG(IMP
1 E1014 .\'1
.. :‘JOVPM
1 18131111

l FIRMA

l

l alt-ts

L. OEC

Table 32. Probit Estimates of Customs Fraud by Commodity Category with Cluster Effect

 

Dependent variable: F RA UD

 

 

 

 

 

 

 

 

Consumer Goods Raw Materials & Fuels Capital Goods
In(”Pendent .. gtxtflli’ . (200313” . can")?
variables 13 X 100 l3 X 100 B X 100
x 100 x 100 x 100
TARIFF 0.008 0.0017 0.985' 0.1568' 1.901'" 0.2901'"
(0.080) (0.0162) (0.530) (0.0848) (0.527) (0.0777)
DETECTION 4.035'" 0.2097‘" 0.439“ 0.070" 0.782‘" 0.1193'”
((1.231) (0.0470) (0.215) (0.0343) (0.229) (0.0355)
AUDIT 4.062‘“ 0.2150‘” 4.368'” -0.2178'" 4.592‘” 0.2429’”
(0.199) (0.0385) (0.330) (0.0511) (0.278) (0.0432)
LOG(IMPORT) 0.293 0.0594 .6284’” 4.0004‘” 2.733” 0.4169"
(1.671) (0.3386) (1.414) (0.2277) (1.185) (0.1806)
EXCHANGE 24.591'” 4.9808‘“ 9.531‘” 1.5172'” 14.482'” 2.2093‘”
(2.598) (0.5246) (1.895) (0.3027) (1.679) (0.2731)
NONPROFIT 6.738 1.4168 28.625 5.4655 18.603 3.2082
(34.691) (7.5707) (23.757) (5.3093) (17.654) (3.3724)
INDIVIDUAL 8.635 1.8032 25.105’“ 4.5637‘“ 4.893 0.7684
(8.761) (1.8834) (8.047) (1.6442) (3845) (1.3971)
FIRMAGE 0.703‘ 0.1423‘ 0.399 0.0636 4.260’“ -0.l923'”
(0.374) (0.0757) (0.267) (0.0422) (0.323) (0.0449)
BRANCH -25.595‘ 4.5873’ 42.569 4.9106 -26.225‘” -3.7l68'"
(13.940) (2.1865) (7.680) (1.1110) (7.719) (1.0236)
OECD 6.528 1.3123 13.009“ 1.9871" 20.459’“ 2.9059'”
(7.155) (1.4288) (6.381) (0.9297) (5.610) (0.7378)
CONSTANT 411.249'” _ -210.512"‘ _ 452.965‘“ _
(37.907) _ (32.489) _ (28.520) _
Observations 2,3105 4,7 99 6,591
Clusters 1.551 2,560 2,339
CorrectlyPredicted 85.29 90.35 90.34
Log-Likelihood -866.252 4458.680 4957.361
Pseudo R2 0.087 0.041 0.064

 

 

 

 

Note: All the estimates are reported after multiplying by 100 to avoid the confusion from rounding. The partial effects
of the dummy variables are for discrete changes from 0 to 1. The values in parentheses below the estimates are the
standard errors, adjusted for cluster sampling. ‘

respectively.

125

, and . denote significance at the 1%, 5%, and 10% levels

 

Table 33. Probit Estimates of Customs Fraud by Commodity without Cluster Effect

 

Dependent variable: FRA UD

 

 

 

 

 

 

 

 

Consumer Goods Raw Materials & Fuels Capital Goods
Injaerli’zggm 11“ X 100 gal) 113 13" x 1 00 ngB 113 11‘ X 100 glxli )B
x 100 x 100 x 100
TARIFF 0.008 0.0017 0935‘ 0.1568' 1901'“ 0.2901'"
(0.092) (0.0187) (0.532) (0.0846) (0.449) (0.0684)
DETECTION 4.035'” 0.2097‘” 0.439” 0.070" 0.782‘“ 0.1193'"
(0.259) (0.0517) (0.209) (0.0333) (0.211) (0.0321)
A UDIT 4 .062‘” 0.2150” 4 .368‘” 0.2178‘” -1 .592‘” -0.2429”'
(0.186) (0.0371) ((1,243) (0.0390) (0.250) (0.0377)
LOG(IMPORT) 0.293 0.0594 -6284'" 4 .0004‘” 2.733“ 0.4169“
(1.747) (0.3539) (1.387) (0.2192) (1.197) (0.1824)
EXCHANGE 24.591'“ 4.9808‘“ 9.531'“ 1.5172'" 14.482‘” 2.2093”
(2.489) (0.4961) (1.929) (0.3047) (1.593) (0.2408)
NONPROFIT 6.738 1.4168 28.625 5.4655 18.603 3.2082
(27.127) (5.9152) (24.357) (5.4508) (21.874) (4.2223)
INDIVIDUAL 8.635 1.8032 25.105“‘ 4.5637'” 4.893 0.7684
(8.818) (1.8978) (8.110) (1.6625) (8.406) (1.3585)
FIRMAGE 0.703" 0.1423" 0.399‘ -0.0636' 4.260'” 0.1923'”
(0.342) (0.0692) (0.228) (0.0363) (0.219) (0.0332)
BRANCH 45.595 4.5873' 42.569' 4 .9106' -26225‘” -3.7168“'
(13.577) (2. 1249) (7.206) (1.0438) (6.201) (0.8101)
OECD 6.528 1.3123 13.009” 1.9871 “ 20.459‘” 2.9059‘”
(7.179) (1.4312) (6,145) (0.8988) (5.534) (0.7268)
CONSTANT 41 1.249‘“ _ -210.512‘“ _ -352.965"' _.
(37.040) _ (31.685) _ (24.932) __
Observations 2,305 4,799 6,591
Clusters 1.551 2.560 2.339
Correctly Predicted 85.29 90.35 90.34
Log-Likelihood -866.252 4458.680 4957.361
Pseudo R2 0.087 0.041 0.064

 

 

 

 

Note: All the estimates are reported after multiplying by 100 to avoid the confusion from rounding. The partial effects
of the dummy variables are for discrete changes from 0 to 1. The values in parentheses below the estimates are the

usual standard errors.

9

O

126

', and ' denote significance at the 1%, 5%, and 10% levels respectively.

 

 

Table 34. Tests for Homoskedasticity in the Probit Model by Commodity Categories

 

 

 

 

 

Homoskedasticity
Consumer Goods Raw Materials & Fuels Capital Goods
LM test LR test LM test LR test LM test LR test
Test statistic 19.943 65.926 65.970 123.224 50.698 83.208
P- value 0.030 0.000 0.000 0.000 0.000 0.000
Critical value 18.307 18.307 18.307 18.307 18.307 18.307
Test result Rejected Rejected Rejected Rejected Rejected Rejected

 

 

 

 

 

 

Note: The critical values for LM and LR tests are derived from the 951h percentile in the chi-squared
distribution with 10 degrees of freedom.

127

 

Table 35. Heteroskedastic Probit Estimates of Customs Fraud for Consumer Goods

 

Dependent variable: FRA UD

 

 

 

 

 

 

 

Independent Homoskedastic Probit Heteroskedastic Probit
variables 8" x 100 g(xl3")8” x 100 (37 x 100 g(x13‘)13‘ x 100
TARIFF 0.008 0.0017 -7317 0.0037
(0.080) (0.0162) (14.473) (0.4694)
DETECTION 4 .035‘“ 0.2097‘” -77.60 0.0031
(0.231) (0.0470) (56.680) (0.7818)
AUDIT 4.062‘” 02150” -260.053 0.0197
(0.199) (0.0385) (162.938) (1.5405)
LOG(IMPORT) 0.293 0.0594 -70.384 0.0357
(1.671) (0.3386) (131.276) (4.5829)
EXCHANGE 24.59 1 49808” 130.295 0.8514
(2.598) (0.5246) (206.360) (117.3061)
NONPROFIT 6.738 1.4168 4828030 -88.3565
(34.691) (75707) (23523 500.0) (9070.4524)
INDIVIDUAL 8.635 1.8032 450.763 0.2444
(8.761) (18834) (481.876) (34.1802)
FIRMAGE 0.703‘ 0.1423’ -53.647 0.0147
(0.374) (0.0757) (47.236) (1.7688)
BRANCH -25.595‘ 4.5873‘ 4 580.240 0.1452
(13.940) (2. 1865) (2393.31) (27.1630)
OECD 6.528 1.3123 -514.802 0.5656
(7.155) (1.4288) (589.405) (75.2271)
CONSTANT 41 1 .249‘“ _ 4947.970 0.4188
(37.907) _ (2733.410) (66.2810)
Observations 2.305 2,305
Clusters 1,551 1,551
Correctly Predicted 85.29 % 85.64 %
Log-Likelihood value -866.252 -833.289

 

 

 

Note: All the estimates are reported after multiplying by 100 to avoid the confusion from rounding. The partial effects
of the dummy variables are computed at the mean values. The partial effect of the Heteroskedastic Probit model,
g(xBA)BA, is equal to 6P(y=l lx) / 8x, = ¢[xB/exp(28)]x[(Bj-bijyeXMZbfl. The values in parentheses below the
Homoskedastic Probit estimates are the standard errors, adjusted for cluster effect. The values in parentheses below the

Heteroskedastic Probit estimates are the usual standard errors. ..

levels respectively.

128

., ", and ° denote significance at the 1%, 5%, and 10%

 

Table 36. Heteroskedastic Probit Estimates of Customs Fraud for Raw Materials & Fuels

 

 

 

 

 

 

Dependent variable: F RA UD
Independent Homoskedastic Probit Heteroskedastic Probit
Variables 13“ x 100 g(x13")13*x 100 13"x 100 gerbil)” x 100
TARIFF 0.985' 0.1568' 0.899 0.2244
(0.530) (0.0848) (7.265) (0.4749)
DETECTION 0.439“ 0.070" -26740‘ 0.0142
(0.215) (0.0343) (14.528) (0.2170)
AUDIT 4 .368‘“ 0.2178‘” 06.218" 0.2139
(0.330) (0.0511) (30.608) (0.6856)
LOG(IMPORT) -6284‘“ 4 .0004‘" 44369 -()_3379
(1.414) (0.2277) (19.751) (1.3414)
EXCHANGE 9.531'” 1.5172'“ -2746) 1,5996
(1.895) (0.3027) (39.613) (3.0695)
NONPROFIT 28.625 5.4655 162.549 0.4344
(23,757) (5.3093) (229.561) (7.3121)
INDIVIDUAL 25.105‘” 45637“ 110.754 -0.2718
(8.047) (1.6442) (79.674) (2.3875)
FIRMAGE 0.399 0.0636 4 .736 0.0356
(0.267) (0.0422) (3.335) (0.1031)
BRANCH 42.569 4 .9106 -343.796 0.1188
(7.680) (1.11 10) (209.807) (3.9810)
OECD 13.009" 1.9871” 99.751 1.7922
(6.381) (0.9297) (70.228) (4.2168)
CONSTANT -210512’" _ 124.107 1.4534
(32.489) _ (500.651) (3.5391)
Observations 4,7 99 4,7 99
Clusters 2.560 2,560
Correctly Predicted 90.35 % 90.37 %
Log-Likelihood value 4458.680 4397.068

 

 

 

 

 

 

 

 

Note: All the estimates are reported after multiplying by 100 to avoid the confusion from rounding. The partial effects
of the dummy variables are computed at the mean values. The partial effect of the Heteroskedastic Probit model,
g(xB3BA, is equal to 6P(y=l lx) / ax, = ¢[xB/exp(z8)]x[(Bj-ijByexp(zb)]. The values in parentheses below the
Homoskedastic Probit estimates are the standard errors, adjusted for cluster effect. The values in parentheses below the

Heteroskedastic Probit estimates are the usual standard errors. m, ", and ' denote significance at the 1%, 5%, and 10%
levels respectively.

129

Table 37. Heteroskedastic Probit Estimates of Customs Fraud for Capital Goods

 

 

 

 

 

 

 

 

Dependent variable: FRA UD
Independent Homoskedastic Probit Heteroskedastic Probit
variables B" x 100 g(x13")p‘ x 100 B” x 100 g(xBTB‘ x 100

TARIFF 1.901'" 0.2901'" 9.680 0.3453
(0.527) (0.0777) (1 1.469) (0.5805)

DETECTION 0.782‘” 0.1 193'” 43.538” 0.0760
(0.229) (0.0355) (17.068) (0.3119)

A UDIT 4 .592'” 0.2429‘” 49.454" 0.1930
(0.278) (0.0432) (27.118) (0.5601 )

LOG(IMPORT) 2.733" 0.4169" 25.360 0.240 1

(1.185) (0.1806) (25.762) (0.7986) F

EXCHANGE 14.482‘” 2.2093‘” 4 .822 2.2833 1
(1.679) (0.2731) (46.608) (3.5012) , ,

NONPROFIT 18.603 3.2082 104.152 0.3364
(17.654) (3.3724) (514.544) (11.9146)

INDIVIDUAL 4.893 0.7684 190.798 4.2105
(8.645) (1.3971) (122.483) (4.8427)

FIRMAGE 4 260'“ 0.1923‘” 44.289‘ 0.0802
(0.323) (0.0449) (8.355) (0.1972)

BRANCH -26.225“‘ 4.7168‘” 027.616‘ 4.4954
(7.719) (1.0236) (537.041) (10.9984)

OECD 20.459'” 2.9059‘” 0.337 3.5305
(5.610) (0.7378) (92.588) (4.9972)

CONSTANT 452.965‘” _ 473.360‘ -7.5661
(28.520) _ (427.756) (18.8249)

Observations 6.5 91 6,5 91
Clusters 2,339 2,339
Correctly Predicted 90.34 % 90.34 %
Log-Likelihood value 4957.361 4915.757

 

 

 

 

Note: All the estimates are reported after multiplying by 100 to avoid the confusion from rounding. The partial effects
of the dummy variables are computed at the mean values. The partial effect of the Heteroskedastic Probit model.
g(xB3BA, is equal to 6P(y=1 lx) / 6x,- = ¢[xB/exp(zfi)]x[(Bj—ijByexMzbfl. The values in parentheses below the
Homoskedastic Probit estimates are the standard errors, adjusted for cluster effect. The values in parentheses below the
Heteroskedastic Probit estimates are the usual standard errors. m, ", and . denote significance at the 1%, 5%, and 10%

levels respectively.

 

130

CHAPTER VII

COMPUTATION OF THE MARGINAL COST OF PUBLIC FUNDS

l. The Empirical Studies on the MCFs

Until recently, little work has been done to measure the MCF of tariff rates, not to
mention the MCF, in the presence of tariff evasion. The only study done in this area,
Clarete and Whalley (1987), finds that the MCF of tariffs on import substitutes in the
Philippines are in the range of 1.28 to 6.99 for the tariff rates of 5 ~ 30%.48 We can also
get some idea of the MCF.) of tariffs by comparing it with the MCF with internal taxes,
such as sales taxes. Clarete and Whalley (1987) calculate the MCF of commodity taxes
on import substitutes in the Philippines, which is in the range of 0.93 to 1.11 for the tax
rates of 5 ~ 30%. Ballard, Shoven, and Whalley (1985) calculate the MCF associated
with consumer sales taxes, which is in the range of 1.256 to 1.388.49

There are only two studies to compute the MCF in the presence of evasion, F ortin and
Lacroix (1994) and Poapongsakom, et. al (2000). However, both of these empirical
studies of the MCF with evasion have been focused on the MCF of the income tax. No
work has been done to measure empirically the MCF of indirect taxes, such as
commodity taxes and tariffs, in the presence of evasion.

Fortin and Lacroix (1994) find that the MCF of an income-tax rate in the presence of

 

48 For their case, when a tariff rate is 10%, the MCF is 1.46.

49 Their central case is the value of 1.388 with elasticities of 0.4 for saving and 0.15 for labor
supply. It should be noted that they use a Cobb-Douglas utility function that imposes the
assumption of unitary elasticities. This may overstate the demand elasticity for many goods,
especially for alcohol, tobacco, and gasoline, which are the goods with the highest tax rates.
Thus, they may overstate the MCF for sales taxes.

131

evasion is in the range of 1.444 and 1.529, using a simultaneous model of labor supply in
the regular and irregular sectors. These values are greater than the corresponding values
of the MCF in the absence of evasion (no irregular market), which ranges from 1.393 to
1.508. Fortin and Lacroix (1994) also explicitly consider the penalty as one of the
effective policy tools to raise revenues in the presence of evasion. They find that the
MCF with respect to penalty rates is 1.33, which is far lower than MCFs of tax rates with
and without evasion. Fortin and Lacroix (1994) find that the MCF associated with a
probability of audit is 1.47, assuming that it could be raised at no cost to the government
and is determined exogenously. As they point out, taking enforcement expenditures into
account is likely to increase the MCF.

Recently, Poapongsakom et a1. (2000) find that the MCF from raising income-tax
rates is 1.043 at their base case labor supply elasticitiy, 0.02. They also compute the
MCF from additional tax enforcement (taxpayer survey). With intermediate risk
aversion, the MCF of an increase in the taxpayer survey is in the range of 1.40 and 11.60,
depending on the survival rates of firms.

Previous studies on the MCF in the presence of income-tax evasion and the related
studies on the MCF of commodity tax or tariff in the absence of evasion are summarized

in Table 38.

132

 

 

2. The MCF in the Absence of Tariff Evasion

We can get a numerical value of the MCF without tariff evasion, MCF,‘, using the
average effective tariff rate and the elasticity of import demand with respect to the price
for each of the commodity categories given in Tables 38 and 39.

First, I calculate the average effective tariff rate for each of the commodity categories,
by just summing up all the rates and dividing them by the number of observations in each
commodity category. These are not weighted-average tariff rates, since I do not account
for the amount of imports. However, despite their limitations, considering the data
availability, they can still give some useful information about average tariff rates. In
addition, this problem can be solved by sensitivity analysis, using different values of
effective tariff rates and elasticities. Actual amount of imports and average effective
tariff rates by commodity categories in 1998 are presented in Table 39.

Second, according to a trade index published by the KCS, the average import price in
U.S. dollars was 17.82% lower in 1998 than in 1997. The average quantity or weight
also decreased by 21.51%. As a result, the total amount of imports (unit import price x
quantity or weight) decreased by 35.5% in U.S. dollars in 1998. However, if we convert
the measurement of the unit price from U.S. dollar to Korean won, using an average
exchange rate in each year, the average import price in terms of Korean won increases by
16.65%. This is mainly due to an increase in the exchange rate.50 In terms of the Korean
won, the import volume decreases only by 8.42%. Then, dividing percentage quantity

changes by percentage price changes in the Korean won, we can roughly estimate

 

5° The average exchange rate used to determine the customs value was 1,011 Korean won per
U.S. dollar in 1997 and 1,435 W/$ in 1998. Thus, the exchange rate increased by 41.94%.

133

elasticities of demands for imported goods with respect to the price, Expo = (AX/X) /
(AP/P), other things being the same. Import price elasticities by commodity categories
are provided in Table 40.

It should be noted that, in calculating demand elasticities based on observations from
1997 and 1998, I control for the change in income (e.g., Gross National Product), since
the observed changes in the quantity of imports are accounted for not only changes in
import prices, but also changes in income. The GNP in Korea decreased by 5.8% in 1998
compared with that in 1997. Thus, I adjust the 1998 quantities by 5.8% to account for a
decrease in income.5| If I do not consider the reduction in income, I would have higher
elasticities in absolute terms, other things being equal. This adjustment may not reflect
the true effects of the income decrease on quantities on import, since I adjust changes in
quantities by the same proportions for all commodities, without accounting for the
different effects of a decrease in income on different commodities. However, despite this
limitation, our adjustment may generate price elasticities of imports that are close to the
true values, since changes in income affect the quantity of imports, and there is no other
way to consider the effects of income except for this method.

It is shown in Table 39 that the overall elasticity of Korean imports is —0.94. The
import elasticities for Consumer Goods, Crude Materials & Fuels, and Capital Goods are

—1.17, —0.83, and —1 .01, respectively. All of these are fairly close to unity. Considering

 

5' By adjusting 1998 quantities by 5.8%, I implicitly assume that the elasticity of imports with
respect to income is unity. This is not far from existing estimates of the elasticity of Korean
imports. For example, Senhadji (1998) finds that the income elasticity of Korean imports is 1.32,
using annual data from 1960 to 1993. Giorgianni and Milesi-Ferretti (1997) also find that the
income elasticity of Korean imports is in the range of 1.24 — 1.40, using quarterly data for the
period of 1973 -— 1995. With annual data, the elasticity is estimated to be 1.20. Mah (1993)
shows that the income elasticity of Korean imports is 0.658, using quarterly data for the period of
1971-1988.

134

the Korean import regime (high weight of Raw Materials & Fuels and Capital Goods),
these estimates are reasonable. Because of the lack of natural resources, Korea depends
heavily on foreign trade, by importing most crude materials and capital goods,
assembling and making goods with these raw materials and capital goods, and then
exporting these goods.52 Thus, for the case of Korea, it is not surprising that raw
materials and capitals goods account for more than 90% of total imports, which are used
mainly for the production of exports. In this context, it can be reasonably argued that
imports, especially Crude Materials & Fuels and some of Capital Goods, are not so
elastic in demand with respect to the price. In addition, these elasticities are in
accordance with studies for the elasticity of imports for Korea, although there are no

empirical studies on these elasticities by commodity category using micro data.53

It should be noted that, for some commodities, the price elasticities of Korean imports
are greater than zero. Non-durable consumer goods, Minerals, and Electric & Electronic
machinery have the positive elasticities. For non-durable consumer goods, we can
reasonably expect that the demands for high-priced goods, such as cosmetics and fur
coats, tend to increase with price because of some psychological effects, such as

bandwagon effects, Veblen effects, etc. For Minerals and Electric & Electronic

 

52 The shares of Consumer Goods, Crude Materials & Fuels, and Capital Goods in total imports in
1998 are 9.54%, 54.23%, and 36.23%, respectively.

53 Senhadji (1998) shows that the price elasticity of Korean imports is —0.84, using annual data
over the period 1960-1993, where the ratio of the import deflator to the GDP deflator is used as
the relative price of imports. Giorgianni and Milesi-Ferretti (1997) find that the price elasticity of
Korean imports is -1.07 when the prices are measured using the ratio of the unit value of imports
to the domestic wholesale price index, and —0.69 when the CPI-based real effective exchange rate
is used, using quarterly data for the period 1973-1995. With annual data, they find that the price
elasticity of imports is —l .19, using the ratio of the unit value of imports to the domestic
wholesale price index. Mah (1993) shows that the price elasticity of Korea imports is -1.029,
using quarterly data for the period of 1971-1988, where the ratio of the import price index to
wholesale price index is used. All of these studies show that the elasticity of Korean imports with
respect to the price is very close to unity.

135

machinery, it is hard to give any plausible explanation for the positive elasticities. For
Minerals, it is possible that a decrease in income has little effects on the quantity of
imports. For Electric & Electronic machinery, it may be the case that the quantity index

is mis-measured.

\Vith these parameter values, the MCF: without evasion can be calculated from
equation (22), MCF,’ = 1 / {1 + [r /(1+r)]e,v,..}. Table 41 shows the MCF,‘ for each
commodity category. Despite some limitations in the calculation of effective tariff rates
and elasticities, our numerical values can still provide us with valuable information about
the MCFs of effective tariff rates in the absence of evasion. As we expect, the MCF in
Crude Materials & Fuels has the lowest value, among the commodity categories, at
1.1446. This is because these goods have relatively lower import demand elasticities and
effective tariff rates. Accordingly, the Consumer Goods sector, with higher elasticities
and tariff rates, has the highest value of the MCF, 1.3064. The MCF in Capital Goods is
1.1722, which falls between the MCFs for Consumer Goods and Crude Materials &
Fuels.

Table 42 shows the MCF of tariff rate change in the absence of evasion, MCF): for
each of the commodity categories for different import demand elasticities. As indicated
above, if we do not consider changes in income, we may have higher import-demand
elasticities. However, given the effective tariff rates of each category, there are no
dramatic changes in the MCF). for lower elasticities. Our central case in the absence of
evasion is an MCF). of 1.1738 with the import demand elasticity of -0.94 and the
effective tariff rate of 18.70%. It should be noted that the average pure tariff rate

excluding internal taxes is 8.14%. The MCF for the pure tariff rate is 1.0761, which is

136

smaller than our central case, 1.1738.

Table 43 gives the MCF; for different demand elasticities and tariff rates. The MCFs
with higher tariff rates are more sensitive to changes in the demand elasticity than the
MCFs with lower tariff rates. Similarly, the MCFs with higher import-demand
elasticities are more sensitive to changes in the tariff rates than the MCFs with lower

elasticities.

137

 

3. The MCF in the Presence of Tariff Evasion

Some of the key parameter values in calculating the MCF with tariff evasion are the
elasticities of evasion. Table 44 gives the elasticities of tariff evasion with respect to the
tariff rate, the audit rate, and the detection rate, which are estimated in Chapter V.

For the calculation of the MCF), we should also specify the concealing cost, c,
incurred by the importer to evade tariff payments, and the elasticity of the concealing cost
with respect to the fraction of imports undeclared, am. These values cannot be directly
observed or estimated. Thus, I choose concealing costs, which are equal to c) = 0.9(t-te)
and c; = 0.45(t-t‘), so that they satisfy the condition, c + t" < t, in chapter III. For am, I
use various values, such as 0.1, 0.05, and 0.01, to see the sensitivity of the MCF).

As mentioned in chapter 11, concealing costs include costs of special packaging, costs
of misinvoicing, extra costs (premiums) of purchasing foreign exchange in the black
market, etc. These concealing costs and elasticities are different from commodity to
commodity, depending on their nature and characteristics. Table 45 provides concealing
costs per one million Korean Won of imports by commodity categories.

For some agricultural products, concealing costs may be relatively higher and more
elastic than those for other imported goods. For example, potato starch is a highly dutied
product (customs duty rate, 327.8%). This is often intentionally mixed with bread
crumbs to be disguised as prepared potato (duty rate, 20%). After customs clearance, the
potato starch is separated from the mixture. The bread crumbs are then discarded, and
the potato starch is sold as such. In this case, the concealing costs would not be trivial.
They would also increase by a large amount, as the amount of potato starch to be

disguised as prepared potato increases. In this case, the importer would need much more

138

bread crumbs. The concealing costs include not only the bread crumbs, but also the
necessary expenses or efforts to separate the potato starch from the mixture. Thus, an
increase in the tariff rate would increase tariff evasion. An increased amount of evaded
tariff payments again would result in an increased concealing costs incurred by the
importer. - For goods like potato starch, due to its nature and characteristics, the
concealing costs per dollar of imports, c, may be relatively large, and the elasticity of the
concealing cost with respect to tariff evasion may also be relatively large.

For the other goods, such as Capitals goods and Consumer goods, the amount of
concealing costs would be negligible, and these costs would be relatively inelastic with
respect to the degree of evasion, since tariff payments are generally evaded by
misinvoicing or counterfeiting of related documents.54 For these goods, the elasticity of
the concealing cost with respect to tariff evasion may be close to zero.

Thus, in computing the MCFs associated with different policy tools, I choose
relatively small parameter values as our base cases, such that the concealing costs = C;
and the elasticity of concealing costs with respect to evasion = 0.05.

In addition, the firm’s expected tariff payments per Korean won of imports, t" .=. [(1-01)
+ B(1+0)or]t, defined in chapter III, should satisfy the condition, [(1-01) + B(1+0)or] < 1.
As indicated in chapter III, without this condition, there is no incentive at all to evade
tariffs by incurring concealing costs. That is, I should be greater than f in all commodity

categories. To check this condition, the firm’s expected tariff payments per import

 

5‘ Because of the continuing deregulation of the foreign exchange market in Korea, the extra costs
of selling or purchasing the foreign exchange in the black market will be small. In addition, it is
more likely that special packaging, which is relatively expensive and elastic in costs, is used in
smuggling of drugs or gold & jewelry, rather than in legally disguised smuggling. The purpose of
special packaging in this case is to successfully pass X-ray inspection machines, metal detectors,
or sensors for drugs.

139

 

(expected tariff rate in the presence of evasion) by category are calculated using the
average values of a, B, and 0 in each commodity category. These results are presented in
Table 46. In all categories, the condition, [(1-or) + B(l+0)0t] < 1, is satisfied. Thus, the
true tariff rate, t, is greater than the expected tariff rate with evasion, 19.

For the elasticities of demand for imported goods in the presence of evasion, I assume
that they have the same values of elasticities as those without tariff evasion, discussed
above. Thus, the elasticities in Table 40 are used for the calculation of the MCF with

evasion.

140

I
ll
Ll
I
l
11
l
I

 

3-1. The MCF of the Tariff Rate Change in the Presence of Evasion

The MCFs of the tariff rate change in the presence of evasion are calculated using the

formula, given in equation (49),

MCF, = 1 I {1 + [1‘ /(1+c+t")]syp - ewaa,(c/te)}.

As mentioned in chapter IV, this formula is derived by assuming that the probability of
audit and the probability of detection are exogenously determined, and they could be

raised at no cost to the government, in accordance with our empirical analysis.

Among the parameters in equation (49), the values of 80“ = (Oot/OtXt/a) are derived
from the partial effects of tariff rates on evasion (Go/6t) at the average values of evasion
and tariff rates, which are estimated using a Tobit model in chapter V. Thus, in
computing the MCF,, I use the lower and upper bounds of the confidence intervals for the
partial effects of tariff rates, as well as the point estimates. In addition, I also consider the
case where an increase in the tariff rate has no effect on evasion, so that ea, = 0 or (Ont/Or)
= 0. This procedure would provide a range of likely values for the partial effects of tariff
rates and, thus, for the MCFs of the tariff rate increase in the presence of evasion. Table
47 provides possible values of the MCF, at corresponging values of ea, or (601/61). Our
central cases of the MCF, are in the range of 1.1443 to 1.3044 for three commodity
categories, with em = 0.05 and C2, It is shown that the MCFs of the tariff rate increase are

very insensitive to different values of ea, or (Oct/6t).

141

 

I also compute the values of the MCF, with different parameter values for the
elasticities of concealing costs with respect to evasion and concealing costs, to see the
sensitivity of the MCF). As given in Table 48, the MCF,’s are shown to be insensitive to
both the concealing cost and the elasticity of the concealing cost with respect to evasion.
They are robust to these parameter values as well as the elasticities of evasion with
respect to tariff rate, as shown in Table 47. Thus, variation of the MCF, in each of the
commodity categories is very small. This is because the third term in the denominator of
the formula for the MCF,, magma/re), has very little effect on the values of the MCF, due
to the very small values of the concealing cost, c. The variations in the MCF, across
commodities come mainly from the differences in the expected tariff rates and the

elasticity of import demand with respect to the price.

Comparing the MCF, with the MCFII, we find very interesting results. For all
commodity categories, regardless of parameter values, the MCFs of the tariff rate change
in the presence of evasion are strictly smaller than the correponding values of the MCFs
in the absence of evasion. These results can be explained in terms of the marginal utility
loss or the marginal tariff revenue. In the presence of tariff evasion, the marginal utility
loss from an increase in a tariff rate is smaller than in its absence, since the consumer
price will increase by a smaller amount with evasion, which works toward a lower MCF.
An increase in the tariff rate will increase the firm’s expected tariff payment, 11", by a
smaller amount. This effect works toward a lower marginal tariff revenue (MT R) and,
thus, a higher MCF. However, the amount of imports will decrease by a smaller amount,
since the market price with evasion will increase by less than 1. This effect works toward

a higher MT R and, thus, a lower MCF. In addition, when the elasticities of concealing

142

costs with respect to evasion are lower, the marginal tariff revenue will have a higher
value, which lowers the MCF, other things being the same. Thus, with a lower am, it is
more likely that the marginal tariff revenue with evasion has a higher value. When
combined with a smaller marginal utility loss, it is not surprising that the MCF,’s in the
presence of evasion have smaller values than in its absence. Even if the marginal tariff
revenue with evasion increases by a smaller amount, it is possible that we may have a
lower MCF, when the marginal utility loss from a tariff rate with evasion is so small.

However, for all commodity categories, the difference between the MCF of the tariff
rate in the presence of evasion and the MCF of the tariff rate in the absence of evasion is
very small. They are very similar in their maganitudes. Thus, even if we account for the
presence of evasion in calculating the MCF), it does not have much of an effect on the
values of the MCF,.

This is partly because the second term in the denominator of the formula for the MCF,,
[f / (1+c+t‘)]axp, is very close to that for the MCF,‘,[t /(l+[)]8xp/\, since the expected tariff
rate, te, is not far from the true tariff rate, t, the concealing cost, c, is small in magnitude,
and I assume that the import demand elasticities in the presence of evasion and in the
absence of evasion are the same.

Second, the third term in the denominator of the formula for the MCFI, amaa,(c/te), has
little effect on the value of the MCF), since [(1-or) + AD(1+0)01] is very close to 1, and the
concealing cost, c, is very small to satisfy the condition, c < (t - t“).55

All of these similarities come from the very small values of evasion, or. Greater

 

55 I choose concealing costs, c. = 0.9(t-t”) and c2 = 0.45(t-t"), so that they satisfy this condition.

143

evasion would lead to much smaller values of the expected tariff rate, and, thus, would
result in much greater differences between the values of the MCF of the tariff rate in the
presence of evasion and those in the absence of evasion. When I assume that relatively
large amounts of tariffs are evaded, such that or = 0.1, the expected tariff rate is far
smaller than the true tariff rate in each of the commodity categories ([(1-or) + AD(1+0)a]
deviates from 1 by a larger amount), as given in Table 49.

Then, as shown in Table 50, the corresponding values of the MCFS of the tariff rate in
the presence of greater evasion have much smaller values than those derived with little
evasion in Table 47. This is because, in the presence of greater evasion, the marginal
utility loss from an increase in a tariff rate would be much smaller, since the consumer
price would increase by a smaller amount than with little evasion. In addition, the
amount of imports would decrease by a smaller amount, since the market price with
evasion will increase by smaller amount, than with little evasion. These effects work
toward much lower values of the MCF in the presence of greater evasion.

Accordingly, the difference between the MCF of the tariff rate change with evasion

and those without evasion would be greater, as more tariff payments are evaded.

144

F
i.
l

1-

 

3-2. The MCF of the Penalty Rate Change in the Presence of Evasion

The MCF associated with the penalty rate in the presence of evasion can be calculated

using the following formula, equation (50),

MCF9 = 1 / (1 + [f I (1+c+r’)]e,,-p - ecaaa9(c/0)/[ADat] }.

Since the penalty rate is fixed to 10% in most cases, we cannot directly estimate the
elasticity of evasion with respect to the penalty rate. Thus, I use several values for the
elasticity, such as Sag = -0.01, -0.05, and -0.1, to see the sensitivity of the values in the
MCFe. I choose relatively small values of the elasticity of evasion with respect to the
penalty rate. This is because an increase in the penalty rate would not have huge impacts
on the evasion decision, since the legislative penalty rate is so low, 10% of the evaded
tariff and tax payments, in most cases. Thus, most importers would not be so sensitive to
a change in the penalty rate. The values of the MCFe with different values of cues, seas,
and the concealing costs are given in Table 51 through 53.

Unlike the MCF of the tariff rate with evasion, the MCng are found to be very
sensitive to both the concealing cost and the elasticity of the concealing cost with respect
to evasion. This is because the third term in the denominator of the formula for the
MCFe, emea9(c/0)/[ADort] has relatively large impact on the values of the MCFe. As
expected, an increase in the deterrent effects of the penalty rate (an increase in the

absolute values of age) reduces the values of the MCFg. As shown in chapter III and IV,

if an increase in the penalty rate has deterrent effects on evasion, the MCFg is always

145

 

 

lower than the MCF), regardless of the other parameter values. Thus, it is clear that the
MCFg’s are smaller than the MCF,’s in all categories, regardless of parameter values.
These results are intuitively obvious. The marginal utility loss from an increase in the
penalty rate is less than the marginal utility loss from an increase in the tariff rate, since
the consumer price increases by a smaller amount in response to an increase in the
penalty rate. That is, OP/OG is less than OP/Ot, since OP/Ot implicitly includes additional
concealing costs for an increase in evasion, while aP/OG includes the decreased amount of
these costs for a decrease in evasion. In addition, when the penalty rate increases, the tax
base (the quantity demanded for imported goods) decreases by a smaller amount
compared with the case of an increase in the tariff rate. Thus, as shown in Table 51
through Table 53, in the presence of evasion, the values of the MCF of a penalty rate

change are always smaller than the MCF of a tariff rate change in all commodities

146

 

 

3-3. The MCF of the Audit Rate Change in the Presence of Evasion

The MCF’s of the probability of audit in the presence of evasion are calculated using

the following formula, equation (51) in chapter IV,

MCFA = l / {1 + [1" / (1+c+t")]axp — acaauAc/[AD(1+0)01I]},

The values of the MCF, in Table 54 are computed at the point estimates of the
coefficients of the audit rate variable and the lower and upper bounds of the
corresponding confidence intervals. Our central cases of the MCFA are in the range of
0.6170 and 1.4980, depending on the commodity categories, with em = 0.05 and C;

Comparing the MCFA with the MCF; and the MCF,, the values of the MCFA are
usually smaller, except for Crude Materials and Capital Goods at the upper bound of the
confidence interval for the partial effects of the audit rate on evasion. At the upper bound
for these two categories, the partial effect of the audit rate on evasion, and, thus, the
elasticities of evasion with respect to the audit rate have positive values. These positive
values result in greater values of the MCFA. If an increase in the audit rate has deterrent
effects on evasion (if the partial effect of the audit rate is positive), then the values of the
MCFA are always smaller than the MCF of a tariff rate change with and without evasion,

with similar reasons to the MCF of the penalty rate.

It should be noted that the values of the MCFA are sometimes less than one. In
particular, all values of the MCFA are smaller than one, at the lower bound in each of the

commodity categories. At the lower bound, the partial effect of an audit rate (the

147

 

 

elasticities of evasion with respect to the audit rate) has the greatest values in absolute
terms. Thus, if an increase in the audit rate effectively deters evasion and the audit rate
can be increased without any resource costs to government, then the government can

solely depend on an increase in the audit rate to raise revenues in the presence of evasion.

I also compute the values of the MCFA with different parameter values for the
concealing costs and the elasticities of concealing costs with respect to evasion, to see the
sensitivity of the MCFA, which is given in Table 55. Like the MCFg, the values of the
MCFA’s are found to be sensitive to both the concealing costs and the elasticity of
concealing cost with respect to evasion. Compared with the MCFa, the values of the
MCFA are shown to be smaller than those of the MCFe at the corresponding parameter
values, if the penalty rate has only small deterrent effects on evasion, such as 80.9 = -0.01.
With greater deterrent effects of the penalty rate, such as Sag = -0.05 or -0. l , the values of
the MCFA are greater than those of the MCFe, at the corresponding parameter values.

Thus, the relative attractiveness of each of the policy variables depends on whether it
has greater deterrent effect on tariff evasion. If one policy variable is more effective in
discouraging evasion, then it is a more attractive policy tool to raise additional
government revenues for a certain public project in the presence of evasion.

If there are relatively much evasion, such that on = 0.1, as shown in Table 56, the
corresponding values of the MCFs of the audit rate in the presence of evasion have
usually much greater values than those derived with little evasion. In the presence of a
greater degree of evasion, even if an increase in the audit rate has a deterrent effect on
evasion, the marginal revenue would be much smaller than the marginal revenue with

little evasion, whcih work toward greater values of the MCFs. If this unfavorable effect

148

 

 

on the MCF exceeds the favorable effects on the MCF with greater evasion, such as the
smaller amount of the marginal utility loss and a smaller decrease in the amount of
imports, then the MCF of the audit rate change in the presence of greater evasion has the

greater values than those derived in the presence of little evasion.

149

3-4. The MCF of the Detection Rate Change in the Presence of Evasion

The MCF’s of the probability of detection given the audit in the presence of evasion

are calculated using the following formula, which was given in equation (52),

MCFD = 1 / {1 + [f / (1+c+re)]eyp — acaeaDc/[AD(1+0)0tt] }.

The values of the MCF D in Table 57 are also computed at the point estimates of the
partial effect of the detection rate on evasion and the lower and upper bounds of the
corresponding confidence intervals.

Our central cases of the MCFD are in the range of 0.6983 and 2.2998, depending on
the commodity categories, with em = 0.05 and c; Comparing the MCFD with the MCF).
and the MCF,, if an increase in the detection rate has deterrent effects on evasion (if the
partial effect of the detection rate is negative), then the values of the MCFD are smaller
than those of the MCF: and the MCF,. However, for all categories, the values of the
MCFD are shown to be greater than those of the MCFA. This is because the absolute
values of the elasticities of evasion with respect to the audit are greater than the

elasticities with respect to the detection rate in all commodities.

As with the MCFA, the MCFD are sometimes less than one. If an increase in the
detection rate effectively deters evasion, and it can be raised without incurring any
resource cost to the government, then the detection rate would be an attractive policy tool

to raise revenues in the presence of evasion.

150

 

 

I do the sensitive analysis with different parameter values for the concealing costs and
the elasticities of concealing costs with respect to evasion as given in Table 58. Like the
MCFg and the MCFA, the values of the MCFD’s are sensitive to the elasticity of
concealing cost with respect to evasion. In all commodity categories, the values of the
MCFD are greater than those of the MCFA and the MCFg even with Bag = -0.01 at the
corresponding parameter values.

With similar reasons to the MCFA, greater evasion leads to much higher values than

those derived with little evasion, in most cases. This result is provided in Table 59.

151

 

Table 38. Empirical Studies on the MCF in the Presence of Evasion

 

 

 

 

Empirical Studies MCF; MCF! MCFG MCFA MCF D

Ballard et al. (1985)

- Sales Tax 1.256 ~ 1.388 -— — — —
Clarete and Whalley (1987)

- Tariff 1.28 ~ 6.99 _ ._ _ __

- Commodity Tax 0.93 ~ 1.11 _ _ _ _
F ortin and Lacroix (1994)

- Income Tax 1.393 ~ 1.508 1.444 ~ 1.529 1.33 1.47 _
Poapongsakom et a1. (2000)

- Income Tax _ 1.043 _ 1.40 ~ 11.60 _
This Study

- Effective Tariff 1.145 ~ 1.306 1.144 ~ 1.304 0.950 ~ 1.304 0.617 ~ 1.498 0,698 ~ 2.30

 

 

 

 

 

 

 

Note: MCF; denotes the MCF of a tax (tariff) rate increase in the absence of evasion. MCF, denote the
MCF of a tax (tariff) rate increase in the presence of evasion. MCFA denotes the MCF of an audit rate
increase in the presence of evasion. MCF D denotes the MCF of a detection rate increase, given an audit, in

the presence of evasion.

152

 

 

Table 39. Effective Tariff Rates by Category

 

 

 

Classification Import (million 5) Effective Tariff Rate (%)

Consumer Goods 8,896 25.07
Cereal 2,520 45.49
Direct Consumer Goods 2,458 41.86
Durable Consumer Goods 2,800 19.24
Non-durable Consumer Goods 1,089 18.58
Crude Materials & Fuels 50,591 17.95
Fuels 18,165 18.69
Minerals 3,146 11.84
Light-Industrial Crude Materials 3,778 12.58
Oil & Fat 279 18.50
Fibre 2,073 20.90
Chemicals 6,117 18.41
Iron & Steel Products 2,979 17.23
Non-Ferrous Metals 3.317 16.95
Capital Goods 33,795 17.02
Machinery & Precision Equipment 11,227 18.44
Electric & Electronic Machinery 19,961 16.10
Transport Equipment 1,491 18.50
Overall 93,2 82 l 8.70

 

153

Table 40. Demand Elasticities of Imported Goods by Commodity

 

 

Price Change Quantity Change Quantity Change Elasticity
Classification (A) (before adjust) (after adjust, C) (C/A)
Consumer Goods 18.75 -27.79 -21.99 -l.17
Cereal 23.75 -6.80 -l .0 -0.04
Direct Consumer Goods 25.08 -35.20 —29.40 -1.17
Durable Consumer Goods 29.27 -33.07 -27.27 -0.93
Non-durable Consumer Goods -29.93 - l 9.49 -13.69 0.46
Crude Materials & Fuels 18.83 -21.48 -15.68 -0.83
Fuels 12.41 -15.71 -9.91 -0.80
Minerals 21.78 -5.04 0.76 0.03
Light-Industrial Crude Material 23.83 -24.94 -19. 14 -0.80
Oil & Fat 38.41 -13.72 -7.92 -0.21
Fibre 21.90 -25.95 -20. 15 -0.92
Chemicals 19.73 -l7.36 -11.56 -0.59
Iron & Steel Products 36.24 -44.72 -38.92 -l.07
Non-Ferrous Metals 23.59 -l7.57 -l 1.77 -0.50
Capital Goods 13.26 -l9.l9 -13.39 -1.01
Machinery & Precision Equipment 30.95 -44.39 -38.59 -1.25
Electric & Electronic Machinery 3.48 3.54 9.34 2.68
Transport Equipment 40.66 -45.58 -39.78 -0.98
Overall 16.65 -21.51 -15.71 -0.94

 

 

Note: Units for price and quantity changes are percentage (%). Quantity change (before adjust) and (after
adjust) denote changes in the quantity of imports before and after the adjustment of quantity change
accounting for the income decrease, respectively.

154

Table 41. The MCF in the Absence of Tariff Evasion

 

 

Classification Effective Tariff (%) Elasticity MCF;

Consumer Goods 25.07 -1 .17 1.3064
Cereal 45.49 -0.04 1 .0127
Direct Consumer Goods 41.86 -1 .17 1.5273
Durable Consumer Goods 19.24 -0.93 1.1766
Non-durable Consumer Goods 18.58 0.46 0.9328
Crude Materials & Fuels 17.95 -0.83 1.1446
Fuels 18.69 -0.80 1.1441
Minerals 1 1.84 0.03 0.9968
Light—Industrial Crude Materials 12.58 -0.80 1.0982
Oil & Fat 18.50 -0.21 1.0339
Fibre 20.90 -0.92 1.1891
Chemicals 18.41 -0.59 1.1010
Iron & Steel Products 17.23 -1.07 1.1866
Non-Ferrous Metals 16.95 -0.50 1.0781
Capital Goods 17.02 - 1 .01 1.1722
Machinery & Precision Equipment 18.44 -1.25 1.2416
Electric & Electronic Machinery 16.10 2.68 0.7291
Transport Equipment 18.50 -0.98 1.1806
Overall 18.70 -0.94 1.1738

 

 

155

 

Table 42. The MCF: by Category for Different Import Demand Elasticities

 

Category

Import Demand Elasticites

 

 

 

-0.59 -0.70 -0.83 -0.94 -1.01 - l .17 -l .25

Consumer Goods 1.1341 1.1632 1.1996 1.2322 1.2538 1.3064v 1.3343
Materials & Fuels 1.0986 1.1192 1.1446. 1.1669 1.1816 1.2166 1.2349
Capital Goods 1.0939 1.1 134 1.1373 1.1584 1.1722. 1.2051 1.2222
Overall 1.1025 1.1239 1.1504 1.1738' 1.1892 1.2260 1.2452

 

 

 

 

 

 

 

 

Note: "‘ denotes our central case of elasticities and MCF 1‘ by category.

156

 

Table 43. The MCF ,1 for Different Parameter Values

 

Import Demand Elasticities

 

 

 

Tarifir Rate -0.20 -0.70 -1.30 -2.00 «3.00
5% 1.0096 1.0345 1.0660 1.1053 1.1667
8% 1.0150 1.0547 1.1066 1.1739 1.2857
12% 1.0219 1.0811 1.1618 1.2727 1.4737
18% 1.0315 1.1195 1.2474 . 1.4390 1.8438

25% 1.0417 1.1628 1.3514 1.6667 2.5
40% 1.0606 1.250 1.5909 2.3333 7.0

 

 

 

 

 

 

157

 

Table 44. Estimated Elasticities of Tariff Evasion by Category

 

 

 

 

Classification Tariff Elasticity Audit Elasticity Detection Elasticity
Consumer Goods 0.102 -O.416 -0.081
Crude Materials & Fuels 0.852 -O.164 -0.083
Capital Goods 1.026 -0.186 -0.050
Overall 0.1 14 -O.232 -0.072

 

 

 

 

 

Note: Tariff Elasticity, Audit Elasticity, and Detection Elasticity denote the elasticities of tariff evasion with
respect to the tariff rate, the audit rate, and the detection rate, respectively. Only for Consumer Goods are
the elasticities of evasion with respect to the audit rate significantly different from zero. The elasticities of
tariff evasion with respect to the detection rate in all commodity categories are not significantly different

from zero, since the coefficients of the detection rate are insignificant.

158

A sM-fl_
,

 

Table 45. Concealing Costs per One Million Korean Won of Imports

 

 

Classification c. ( W) c2 ( W)
Consumer Goods 1,053 527
Crude Materials & Fuels 333 167
Capital Goods 261 13 1

Overall 405 203

 

 

 

 

 

159

 

Table 46. Expected Tariff Rates by Category

 

 

 

 

Classification [(1-0t) + AD(1+0)0L] Expected Tarifl(%) True Tariff (%)
Consumer Goods 0.9941 24.924 25.072
Crude Materials & Fuels 0.9977 17.913 17.954
Capital Goods 0.9981 16.983 17.015

Overall 0.9972 18.648 18.70

 

 

 

 

160

 

Table 47. The MCF, in the Presence of Tariff Evasion with em = 0.05 and 6'2

 

 

 

 

Classification 0 Lower Bound Point Estimate Upper Bound
Consumer Goods 1.3043 1.3043 1.3044 1.3044
Crude Materials & Fuels 1.1443 1.1443 1.1443 1.1444
Capital Goods 1.1718 1.1718 1.1719 1.1719

Overall 1.1733 1.1733 1.1733 1.1733

 

 

 

 

 

161

 

Table 48. Sensitivity Analysis of the MCF, with Different Parameter Values

 

 

 

 

 

MCF,
Category em = 0.1 em = 0.05 em = 0.01
c. c; Cl C; c. c:
Consumer Goods 1.3043 1.3044 1.3042 1.3044 1.3042 1.3043
Crude Materials 1.1444 1.1444 1.1443 1.1443 1.1443 1.1443
CapitalsGoods 1.1720 1.1719 1.1719 1.1719 1.1718 1.1718
Overall 1.1733 1.1733 1.1733 1.1733 1.1733 1.1733

 

 

 

 

 

 

 

 

 

Note: For the sensitive analysis, the partial effect of the tariff rate on evasion in each of the commodity
categories is set equal to the point estimate, so that em = 0.102 for Consumer Goods, em = 0.852 for Raw
Materials & Fuels, ea, = 1.026 for Capital Goods, ea, = 0.114 for Overall, as given in Table 43.

162

 

Table 49. Expected Tariff Rates with Higher Evasion Ratio (CL = 0.1)

 

 

Classification [( 1-(1) + AD(1+0)0L] Expected Tar{[f(%) True Tarifl(%)
Consumer Goods 0.9033 22.648 25.072
Crude Materials & Fuels 0.9012 16.180 17.954
Capital Goods 0.9013 15.336 17.015

Overall 0.9016 16.860 18.70

 

 

 

 

 

 

163

 

EO—

 

Table 50. The MCF, in the Presence of Tariff Evasion with or = 0.1

 

 

 

 

Classification 0 Lower Bound Point Estimate Upper Bound
Consumer Goods 1.2766 1.2766 1.2766 1.2766
Crude Materials & Fuels 1.1309 1.1309 1.1309 1.1310
Capital Goods 1.1553 1.1553 1.1554 1.1554

Overall 1.1572 1.1572 1.1572 1.1572

 

 

 

 

 

Note: 1 use the same parameter values as in Table 46, such that em = 0.05 and CZ.

164

 

 

 

Table 51. The MCFg in the Presence of Tariff Evasion with Sue = -0.01

 

 

 

 

 

 

Category cm 0.1 ecu = 0.05 em = 0.01
c, c; c. c2 c, c;
Consumer Goods 1.0069 1.1366 1.1374 1.2152 1.2688 1.2864
Crude Materials 0.6295 0.8122 0.8122 0.9501 1.0580 1.0995
Capitals Goods 0.6449 0.8319 0.8320 0.9731 1.0836 1.1260
Overall 0.7166 0.8898 0.890 1.0122 1.1035 1.1373

 

 

 

 

 

 

 

 

165

 

 

_._..1!’

Table 52. The MCFg in the Presence of Tariff Evasion with Sue = -0.05

 

Category

 

 

 

em = 0.1 em = 0.05 a“, = 0.01
0. c2 6. 62 Ci 6'2
Consumer Goods 0.5251 0.7488 0.7492 0.9518 1.1374 1.2152
Crude Materials 0.2248 0.3757 0.3758 0.5658 0.8122 0.9501
Capitals Goods 0.2303 0.3850 0.3850 0.5796 0.8320 0.9731
Overall 0.2801 0.4523 0.4523 0.6530 0.890 1.0122

 

 

 

 

 

 

 

 

 

 

166

 

fly

- a '7'
[PH

 

Table 53. The MCFg in the Presence of Tariff Evasion with sag = -0.1

 

Category

 

 

 

 

 

 

8w: 0.1 eca= 0.05 ecu: 0.01
c1 c; c. 02 Ci 62
Consumer Goods 0.3286 0.5250 0.5251 0.7488 1.0069 1.1366
Crude Materials 0.1246 0.2248 0.2248 0.3757 0.6295 0.8122
Capitals Goods 0.1277 0.2303 0.2303 0.3850 0.6449 0.8319
Overall 0.1591 0.2801 0.2801 0.4523 0.7166 0.8898

 

 

 

 

 

 

167

 

Table 54. The MCFA in the Presence of Tariff Evasion with em = 0.05 and c;

 

 

 

 

Classification Lower Bound Point Estimate 0 Upper Bound
Consumer Goods 0.8624 1.0183 1.3043 1.2431
Crude Materials & Fuels 0.6198 0.8769 1.1443 1.4980
Capital Goods 0.6170 0.8709 1.1718 1.4797

Overall 0.7455 0.8781 1.1733 1.0666

 

 

 

 

 

168

 

Table 55. Sensitivity Analysis of the MCFA with Different Parameter Values

 

 

 

 

 

categc’ry ecu: 0.1 am: 0.05 em= 0.01
c. c: c. C; c. c;
Consumer Goods 0.6142 0.8351 0.8351 1.0183 1.1725 1.2350
Crude Materials 0.5155 0.7108 0.7107 0.8769 1.0198 1.0785
Capitals Goods 0.4919 0.6929 0.6929 0.8709 1 .0295 1.0960
Overall 0.5003 0.7015 0.7015 0.8781 1.0342 1.0994

 

 

 

 

 

 

 

 

 

Note: For the sensitive analysis, the partial effect of the audit rate on evasion in each of the commodity
categories is set equal to the point estimate, so that 8M = -0.416 for Consumer Goods, 8M = -0.164 for
Raw Materials & Fuels, em = -0.186 for Capital Goods, 8a,; = -0.232 for Overall, as given in Table 43.

169

Table 56. The MCFA in the Presence of Tariff Evasion with or = 0.1

 

 

 

 

Classification Lower Bound Point Estimate 0 Upper Bound
Consumer Goods 1.2387 1.2555 1.2766 1.2728
Crude Materials & Fuels 1.1095 1.1231 1.1309 1.1370
Capital Goods 1.1362 1.1479 1.1553 1.1598

Overall 1.1391 1.1465 1.1572 1.1540

 

 

 

 

 

Note: 1 use the same parameter values as in Table 53, such that em = 0.05 and oz.

170

 

Table 57. The MCFD in the Presence of Tariff Evasion with em = 0.05 and C2

 

 

 

 

Classification Lower Bound Point Estimate 0 Upper Bound
Consumer Goods 1.0562 1.2367 1.3043 1.4929
Crude Materials & Fuels 0.7165 0.9913 1.1443 1.6034
Capital Goods 0.6983 1.0722 1 . 1718 2.2998

Overall 0.8762 1.0624 1 . 1733 1.3471

 

 

 

 

 

171

 

Table 58. Sensitivity Analysis of the MCFD with Different Parameter Values

 

 

 

 

 

 

 

 

 

 

 

 

 

 

MCF D
Category em = 0.1 em = 0.05 em = 0.01
c. c; c. c; c. c;

Consumer Goods 1.0701 1.1757 1.1756 1.2367 1.2763 1.2902
Crude Materials 0.7075 0.8744 0.8743 0.9913 1.0777 1.110
Capitals Goods 0.8543 0.9882 0.9882 1.0722 1.1298 1.1504

Overall 0.8278 0.9707 0.9707 1.0624 1.1263 1.1493
Note: For the sensitive analysis, the partial effect of the detection rate on evasmn in each of the commodity

categories is set equal to the point estimate, so that sup = -0.081 for Consumer Goods, sap = -0.083 for
Raw Materials & Fuels, Sap = -0.050 for Capital Goods, Sap = -0.072 for Overall, as given in Table 43.

172

Table 59. The MCFD in the Presence of Tariff Evasion with on = 0.1

 

 

 

 

Classification 0 Lower Bound Point Estimate Upper Bound
Consumer Goods 1.2766 1.2589 1.2724 1.2863
Crude Materials & Fuels 1.1309 1.1157 1.1269 1.1383
Capital Goods 1.1553 1.1408 1.1533 1.1660

Overall 1.1572 1.1464 1.1538 1.1613

 

 

 

 

 

Note: 1 use the same parameter values as in Table 56. such that em = 0.05 and c2.

173

 

CHAPTER VIII

POLICY IMPLICATIONS

Our central case of the MCF,"s in the absence of evasion and the MCF,’s associated
with different policy instruments in the presence of tariff evasion are summarized in
Table 60.

First, if we consider tariff evasion in measuring the MCF of a tariff rate, despite
concealing costs incurred by the importers, the MCF, with evasion is smaller than the
corresponding value without tariff evasion in all commodity categories, although the
differences are usually small. This is mainly because the domestic price of the imported
goods in the presence of evasion is shown to be lower than the price without evasion
(tariff-inclusive world price). Thus, considering tariff evasion in calculating the MCFs,
we should not a priori conclude that benefits from the government project financed by an
increase in tariff rates with evasion must be greater than those which do not consider
tariff evasion. For example, in the presence of evasion, when we increase a tariff rate on
Consumer goods by one percent, the MCF per additional won of tariff revenue is in the
range of 1.3043 and 1.3044. This implies that a public project must produce marginal
benefits of at least W 1.3044 per won of cost if it is to be justified. However, without
considering evasion, the marginal benefits from a public spending should be at least W
1.3064 per won of its cost.

Second, the government can finance additional dollars of public spending by
increasing tax rates, by increasing penalty rates, by increasing the probability of audit, or

by increasing the probability of detection. These government policy tools cause

174

distortions in the individual‘s behavior through various elasticities and through the
concealing costs of evaders. Thus, the government should compare the MCF of each
policy instrument when it chooses one policy tool to raise additional revenues for a
certain public project. As discussed in chapter VI and summarized in Table 60, the
values of the MCFs associated with the penalty rate, the probability of audit, and the
probability of detection are usually lower than those of the MCFs of tariff rates, with and
without evasion. For some commodities, these values are sometimes less than one.

However, although they seem to be attractive policy tools to raise government
revenues, the penalty rate as well as the probability of audit and detection cannot be
increased by a huge amount. The penalty rate for tariff evasion (legally disguised
smuggling) should be compatible with the penalty rates for other crimes, such as tax
evasion.56 Thus, the government cannot resort solely to penalty rates in raising additional
revenue for a public project. In addition, if the actual probability of detection is not
significantly high, then the effect of higher penalty rates on tariff evasion may be small.
In this case, an increase in penalty rates to raise additional government revenues may not
be so effective. Furthermore, if an increase in penalty rates has no effect on evasion
behavior, then the penalty rate is not a superior policy instrument compared with the tariff
rate. As shown in Table 60, with 80,9 = 0, the MCF’S of an increase in penalty rates are
actually the same as the MCF’ s of an increase in tariff rates across commodities.

The audit rate and the detection rate cannot be increased dramatically, and we should

be careful when we use these enforcement variables as one of policy tools. For our

 

56 For example, we cannot raise the time of imprisonment for W 1,000 tariff evasion to 50 years,
or increase a fine to W 1,000,000. According to IRS, in 1997, the average time of imprisonment
for tax violations was 2 years.

175

empirical analysis, the probability of audit and the probability of detection are assumed to
be exogenously determined, since 1 use lagged values of these variables. Thus, we do not
have to consider any government expenditures on enforcement to measure the MCFA and
the MCFD. However, the government would have to spend its own direct resources to
raise audit rates or detection rates, as explained in chapter III. In addition, more
importantly, higher rates of audit on goods at the time of imports would impose a great
burden on firms in terms of money and waiting costs. Considering the industrial
structure of Korea, which heavily depends on foreign trade, it would damage the Korean
economy as a whole.57

Third, as Ballard, Shoven, and Whalley (1985) point out, the MCPs of different taxes
may have substantially different values. Likewise, although I only consider the MCF’s
with tariff evasion, the MCF’s in the presence of other types of tax evasion, such as
income-tax evasion, may have a different value. Higher tariff rates and higher income-
tax rates do not have the same effect on evasion behavior or on distortions. The penalty
rate and the inspection rate may also have different effects on evasion, and thus different
values of the MCF’s, in the case of income-tax evasion. Thus, they may have totally
different values. In addition, if we include income-tax evasion as well as tariff evasion in
the same framework to calculate the MCF’s, we can reasonably expect to find different
values of the MCFs. Therefore, the government should also consider other taxes, which

have smaller MCF’s in the presence of evasion, as policy tools to raise additional

 

’7 In order to achieve two contradicting goals, maintaining speedy clearance and detecting illegal
trade, the post-audit is conducted after import declarations are accepted to check commodity
classification and value determination, as well as legitimacy of impost declaration. In this case,
the government and importers would also incur direct resource costs.

176

revenues. In other words, we can choose those taxes or tariffs which induce less evasion,

other things being equal.

177

Table 60. The MCFs in Our Central Case

 

 

 

 

Category MCF,‘ MCF, MCFe MCFA MCFD
Consumer Goods 1.3064 1.3043 ~ 1.3044 1.2152 ~ 1.3043 0.8624 ~ 1.3043 1.0562 ~ 1.4929
Raw Materials 1.1446 1.1443 ~ 1.1444 0.9501 ~ 1.1443 0.6198 ~ 1.4980 0.7165 ~ 1.6034
Capital Goods 1.1722 1.1718 ~ 1.1719 0.9731 ~ 1.1718 0.6170 ~ 1.4797 0.6983 ~ 2.2998

Overall 1.1738 1.1733 1.0122 ~ 1.1733 0.7455 ~ 1.1733 0.8762 ~ 1.3471

 

 

 

 

 

 

Note: The values of the MCFX, the MCF,, the MCFA, and the MCF D come from Table 40, Table 46, Table
53, and Table 56, respectively. The values of the MCFg are derived with parameter values of em = 0.05,

8&9 = -0.01 and O, and CZ.

178

 

CHAPTER IX

CONCLUSION

In this paper, I examine the major determinants of tariff evasion and calculate the
MCF’s associated with various policy tools, using detailed micro data for 1998 in Korea.
I show that, if we incorporate tariff evasion and concealing costs incurred by firms in
calculating the MCF” 8, different values are measured than those derived without
considering evasion, although the differences are usually small.

I find that tariff evasion increases with tariff rates in all commodity categories. That
is, estimates of the elasticity of tariff evasion with respect to a tariff rate are positive and
generally significant, in the range of 0.102 and 1.026. For Consumer goods, an increase
in the inspection rate has deterrent effects on evasion, with an elasticity of -0.416.
However, the detection rate is foundito have no significant effects on evasion in any of
the commodity categories, although the detection-rate coefficients have the expected
signs.

1 also calculate the MCFS with and without evasion. Without evasion, the MCF’s
associated with tariff rates vary from 1.1446 to 1.3064, depending on commodity
categories. Surprisingly, with lower values of sea (the elasticity of concealing costs), the
MCF/s in the presence of tariff evasion are found to be smaller than those without
evasion. The MCF’s of tariff rates in the presence of evasion are in the range of 1.1443
and 1.3044 for three commodity categories. Overall, an increase in the probability of
audit, an increase in the penalty rate, and an increase in the probability of detection are

found to be more efficient policy tools for raising additional revenues than tariff rates, if

179

these enforcement variables have deterrent effects on evasion (with negative values of the
partial effects of enforcement variables). The MCFS of the penalty rate are in the range
of 0.9501 and 1.3043, with our base case parameter values. The MCF’s of the audit rates
vary from 0.6170 to 1.4980, depending on commodity categories. The MCF’s of the
detection rates are in the range between 0.6983 to 2.2998, depending on commodity
categories.

As novel results, the main contributions of this paper are an estimation of the major
determinants of tariff-evasion behavior and customs fraud, and the provision of nmnerical
values of the MCF’s associated with different government policy tools. I am not aware
of any other empirical work on the effects of tariff rates and government enforcement
activities on tariff evasion or customs fraud. Furthermore, there is no theoretical or
empirical study at all on the MCF’s associated with different policy instruments in the
presence of tariff evasion.

However, there are some limitations in my analysis, although I use the most detailed
and reliable data. First, among other things, in estimating tariff evasion and thus
calculating the MCF’ s, I do not account for all types of legally disguised smuggling, such
as the importation of undeclared items, the undervaluation, the underdeclaration of a
quantity or weight, etc. Except for the declaration of false items, I can calculate an
evasion ratio or propensity to evade only when the actual amount of the undervaluation or
the underdeclaration are identified from the descriptions of actions taken by the KCS
after detecting customs fraud. Thus, accounting for other types of customs fraud, 1 also
estimate behaviors of importers that are associated with committing customs fraud, using

a binary dependent variable model in a separate chapter.

180

A second limitation of my work is that I do not consider enforcement costs by the
government and compliance costs by importers related to inspection on imports. I
assume that the audit rate and the detection rate could be raised at no cost to the
government, to be consistent with our empirical studies. In addition, importers are
assumed not to incur any compliance cost, in response to an increase in each of the policy
variables. This is because these costs can be neither directly observed nor estimated. If
we include these costs, the MCF” 5 may have different values.

Third, unlike Ballard, Shoven, and Whalley (1985), I do not account for interactions
among different taxes. That is, I only focus on the calculation of the MCF’s associated
with tariffs, ignoring the interaction of tariffs with other taxes, such as the income tax.
Thus, this paper does not provide any evidence regarding which kinds of taxes (including
tariffs) are the most efficient way to raise revenues for a certain public project.

Last, our estimates of the MCFs are sometimes very sensitive with respect to the
elasticity of the concealing cost with respect to tariff evasion, 8“,. Although the
elasticities are different from commodity to commodity, depending on their nature and
characteristics, I use somewhat arbitrary values of the elasticities for the calculation of
the MCF’s, since there is no other way to estimate 8901. For some agricultural products,
due to their nature and characteristics, the elasticity of the concealing cost with respect to
tariff evasion would be relatively elastic. For other goods, due to the characteristics of
methods by which the importers evade tariff payments, the concealing costs would be
relatively inelastic with respect to the degree of evasion. Thus, it would be interesting for
further study to find ways to acquire greater certainty about the value of this parameter.

Despite some limitations mentioned above, our numerical measure of the MCF’s in

181

the presence of tariff evasion can still provide valuable information in deciding the most
appropriate policy tool for a certain public project. In addition, the Tobit estimates of
tariff-evasion behavior and the Probit estimates of customs fraud are the first empirical
studies in this area, to my knowledge. These results can also provide the basis for future

research in these fields.

182

APPENDIX

183

APPENDIX

THE MCF AND THE OPTIMAL PROVISION OF PUBLIC GOODS

First, the Lagrangian of the individual problem is

i = (100 + V(6)+ ulL—PX].

where P equals to P = (1 + c + 1.9)}; in the presence of evasion and PA = (1 + t)P’ in its

absence, as with the previous section. The first-order condition is as follows.

max = Ux - up: 0, (A1)

where UX (=6U/0X) denotes the derivative of the individual’s utility with respect to
private consumption of imported goods, X. From above equations, we can implicitly
derive optimal values of X = X(t,0,d), given I, 0, and d. Substituting the optimal choice of
X into the individual’s budget constraint and differentiating it with respect to t, 0, and d,

we get,

(op/at) X + PX, = 0 (A2)
(op/00) X + PXe = 0 (A3)
(aP/ad)X+ PXd= 0, (A4)

184

where X,, X0, and Xd denote the derivatives of the individual’s consumption of imported

goods with respect to the tariff rate, the penalty rate, and enforcement expenditures,

respectively.

Then, the Lagrangian of the government problem is

£ = U(X(t, 0, d)) + V(G) + Moe - d)P‘X- PGG/H].

The first-order conditions are as follows.

(7106!: UXX, + 1[(af/ar)P‘X+ (f - d) P‘X,] = 0 (A5)
dis/60 = (1ng + A[(ar/ae)P‘X + (1" - d)P’Xe] = 0 (A6)
arc/ad = UXXd+ mare/ad - 1)P‘X + (f - d)P‘Xd] = 0 (A7)
61060 = VG - APO/H = 0, (A8)

where VG (=aV/6G) denotes the derivative of the individual’s utility with respect to the
level of public goods, G. Using the first order condition of individual problem, equation
(A1), equation (A8) can be written as,

HVG/ Ux = (it/p) x PG/P. (A8)'

From (A8)', we can derive the condition of the optimal provision of public goods,

ZMRS = MCF,- x MRT,

185

where ZMRS = HVG/UX is the sum of the marginal rates of substitution, MCF,- = 70,711 is
the marginal cost of public funds associated with each policy variable, and MRT = P(;/P
is the marginal rate of transformation.

Third, using the first-order condition of the individual problem and the derivatives of
the individual budget constraint, we can derive MCFs of each of the government policy
tools from equations (A5), (A6), (A7), and (A8)'. It should be noted that derivatives of
[(te - d)P'X - PUG/H] with respect to t, 0, and d, are just equal to the marginal tariff
revenue from an increase in t, 0, and d, respectively. Thus, the term of [ - ] in each
equation (A5), (A6), and (A7) equals to MT R,, MT R9, and MT Rd, respectively. The term,
U 1X,- in each of these equation is equal to -p.(6P/6i)X, where i = t, 0, and d, since UX = uP
from equation (A1), and , PX,- = -(aP/6i)X from equations (A2), (A3), and (A4). From
equations (18) through (21), the term, (6P/6i)X, is just the marginal utility loss (MUL.)
from each of the government policy tools. Thus, the MCF,- = Ar/u is shown to be equal to

MUL/MTR), where i = t, 0, and d.

186

BIBLIOGRAPHY

187

BIBLIOGRAPHY

Allingham, MG, and A. Sandmo (1972), “Income Tax Evasion: A Theoretical
Analysis.” Journal of Public Economics 1, 323-338.

Andreoni, J ., B. Erard, and J. Feinstein (1998), “Tax Compliance.” Journal of Economic
Literature 36, 818-860.

Atkinson, A.B., and NH. Stern (1974), “Pigou, Taxation and Public Goods.” Review of
Economic Studies 41, 119-128.

Ballard, CL. (1997), “Marginal Efficiency Cost Calculations for Different Types of
Government Expenditure: A Review.” in J .G Head and R. Krever (eds.), Taxation
Towards 2000, Melbourne: Australian Tax Research Foundation, 537-563.

Ballard, CL. (1990), “Marginal Welfare Cost Calculations: Differential Analysis vs.
Balanced-Budget Analysis.” Journal of Public Economics 41, 263-276.

Ballard, CL, and D. Fullerton (1992), “Distortionary Taxes and the Provision of Public
Goods.” Journal of Economic Perspectives 6, 117-131.

Ballard, C.L., J .B. Shoven, and J. Whalley (1985), “General Equilibrium Computation of
the Marginal Welfare Costs of Taxes in the United States.” American Economic
Review 75, 128-138.

Clarete, R. L., and J. Whalley (1987), “Comparing the Marginal Welfare Costs of
Commodity and Trade Taxes.” Journal of Public Economics 33, 357-362.

’_ Clotfelter, CT. (1983), “Tax Evasion and Tax Rates: An Analysis of Individual
Returns.” The Review of Economics and Statistics 65, 363-373.

Cremer, H., and F. Gahvari (1993), “Tax Evasion and Optimal Commodity Taxation.”
Journal of Public Economics 50, 261 -275.

1 " Cremer, H., and F. Gahvari (1999), “Excise Tax Evasion, Tax Revenue, and Welfare.”

Public Finance Review 27, 77-95.

Cullis, J ., and P. Jones (1998), Public Finance and Public Choice. New York, NY:
Oxford University Press.

Davidson, R., and J. MacKinnon (1993), Estimation and Inference in Econometrics. New
York: Oxford University Press.

188

Dubin, J. A., and L. L. Wilde (1988), “An Empirical Analysis of Federal Income Tax
Auditing and Compliance.” National Tax Journal 41, 61-74.

Erard, B. (1997), “Self-selection with Measurement Errors: A Microeconometric
Analysis of the Decision to Seek Tax Assistance and its Implications for Tax
Compliance.” Journal of Econometrics 81 , 319-356.

Falkinger, J. (1991), “On Optimal Public Good Provision with Tax Evasion.” Journal of
Public Economics 45, 127-133.

F einstein, J. S. (1991 ), “An Econometric Analysis of Income Tax Evasion and its
Detection.” Rand Journal of Economics 22, 14-35.

Fortin, B., and (1 Lacroix (1994), “Labor Supply, Tax Evasion and the Marginal Cost of
Public Funds: An Empirical Investigation.” Journal of Public Economics 55, 407-
431.

Giorgianni, L., and G. M. Milesi-Ferretti (1997), “Determinants of Korean Trade Flows
and their Geographical Destination.” Working Paper 54, International Monetary
Fund.

Kaplow, L. (1990), “Optimal Taxation with Costly Enforcement and Evasion.” Journal
of Public Economics 43, 221-236.

Korea Customs Service (1999), Overview on Korea Customs Service.

Mah, J. S. (1993), “Structural Change in Import Demand Behavior: The Korean
Experience.” Journal of Policy Modeling 15, 223-227.

Martin, L., and A. Panagariya (1984), “Smuggling, Trade, and Price Disparity: A Crime-
Theoretic Approach.” Journal of International Economics 17, 201-217.

Mayshar, J. (1990), “On Measure of Excess Burden and Their Application.” Journal of
Public Economics 43, 263-289.

Mayshar, J. (1991), “On Measuring the Marginal Cost of Funds Analytically.” American
Economic Review 81, 1329-1335.

Myles, G. D. (1995), Public Economics. New York, NY: Cambridge University Press.

Schulze, G G (1994), “Misinvoicing Imports: The Interdependence of Tax and Tariff
Evasion.” Public Finance Quarterly 22, 335-365.

Senhadji, A. (1998), “Time-Series Estimation of Structural Import Demand Equations: A
Cross-Country Analysis.” Stafir Papers 45, International Monetary Fund, 236-268.

189

V Slemrod, J ., and S. Yitzhaki (1996), “The Costs of Taxation and the Marginal Efficiency
Cost of Funds.” Stafl Papers 43, International Monetary Fund, 172-198.

Srinivasan, T. N. (1973), “Tax Evasion: A Model.” Journal of Public Economics 2, 339-
346.

Tanzi, V., and P. Shome (1993), “A Primer on Tax Evasion.” Staff Papers 40,
International Monetary Fund, 807-828.

Thursby, M., R. Jensen, and J. Thursby (1991), “Smuggling, Camouflaging, and Market
Structure.” Quarterly Journal of Economics 106, 789-814.

Usher, D. (1984), “An Instructive Derivation of the Expression for the Marginal Cost of
Public Funds.” Public Finance 39, 406-411.

Usher, D. (1986), “Tax Evasion and the Marginal Cost of Public Funds.” Economic
Inquiry 24, 563-586.

Wildasin, D. E. (1984), “On Public Good Provision with Distortionary Taxation.”
Economic Inquiry 32, 227-243.

Wooldridge, J. M. (2001), Econometrics Analysis of Cross Section and Panel Data.
Forthcoming, Cambridge, MA: MIT Press.

Wooldridge, J. M. (2000), Introductory Econometrics: A Modern Approach. Cincinnati,
OH: South-Westem College Publishing.

Yitzhaki, S. (1987), “On the Excess Burden of Tax Evasion.” Public Finance Quarterly
15, 123-137.

190

IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII

111111111111111llljllljlllllllflllll