’:' 1:1: v0- .nmrwfi

can?! -

..,. «wan.

*
w
m
..,
m

.4. J $73

 

 

:. (1..., v .n... ..~

 

:, .,..,...,. .... ..;,... I. f z 1...;

 
 
 
 
 
 

 

'FHF‘E'S

c/

 

 

got-s
glib/o ‘; “:2 7
This is to certify that the
dissertation entitled
.93
>- E >
a]: (f) 3;: l ELECTRONIC PURCHASES, CROSS-BORDER
(.1: f .. 93 SHOPPOING, AND SALES TAXATION
l: .c: c
.2 3
2 presented by

 

JAIMIN LEE

has been accepted towards fulfillment
of the requirements for the

degree In Economics

 

 

mag/war

Major Professor’ 3 Signature
2 ,Tk) // /®. (13
7

Date

 

MSU is an Affirmative Action/Equal Opportunity Institution

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

 

OCJSZOOIZQPES

 

MAR ’1333 29083

 

SEP 3 6 2008

 

gilt zt 035

 

 

 

 

 

 

 

 

 

 

6/01 c'JCIRC/DatoDuepGS-p. 1 5

 

ELECTRONIC PURCHASES, CROSS-BORDER SHOPPING, AND SALES
TAXATION

By

Jaimin Lee

A DISSERTATION

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

DOCTOR OF PHILOSOPHY
Department of Economics

2003

ABSTRACT

ELECTRONIC PURCHASES, CROSS-BORDER SHOPPING, AND SALES
TAXATION

By

Jaimin Lee

This dissertation reviews a variety of issues regarding online businesses, especially
the relationship between sales taxes and Internet purchases. I investigate the relationship
between them, and expand it to include cross-border shopping. I analyze the roles played
by place of residence and the sales-tax rates in determining the decision of the consumer,
using CPS (Current Population Survey) data for 1997 and 2001

I find that Internet shopping increases with the sales-tax rates. When a consumer
lives in a jurisdiction with a high sales-tax rate, she/he would like to use Internet
shopping. I also estimate the relationship between online purchases and the consumer’s
geographic characteristics. I expect that when she lives in a jurisdiction adjoining another
jurisdiction with a lower sales-tax rate, a consumer would be more likely to travel across
the border and purchase goods in a different jurisdiction, instead of making electronic
purchases. According to my estimation, if the consumer lives in a jurisdiction adjoining a
different jurisdiction with a lower sales-tax rate, an individual would be less likely to
make Internet purchases.

I analyze the welfare effects of a revenue-neutral change in tax policies under which
the sales and use tax is successfiJlly collected from remote sales (online commerce and
mail-order shopping). I compare the welfares of two cases: one is the case in which

consumers can escape the sales and use tax by remote sales (the present real world), and

the other is that in which state and local governments can successfully collect the sales
and use tax from remote sales (an alternative scenario). I use the excess burden to
elucidate the economic inefficiency of the current sales tax system, which is
characterized by tax evasion with mail-order shopping and e-commerce. A representative
consumer can realize a welfare increase when state and local governments can

successfully impose the sales and use tax on remote sales, and decrease sales-tax rates.

This dissertation is dedicated to my family

iv

ACKNOWLEDGMENTS

First of all, I would like to give my sincerest thanks to my advisor, Professor Charles
Ballard, for his technical advice, generous support, and insightful comments. I owe a lot
of things to my mentor, Professor Charles Ballard, and without his special help, my
dissertation would not be made. I would also like to thank Professor Lawrence Martin,
Professor Paul Menchik, and Professor Leslie Papke at the department of economics, and
I also give a thanks to Professor Ed Outslay at the department of accounting. They gave
me very helpful comments for my dissertation and helped me complete it.

I would like to express my special gratitude to my family, parents, parents-in-law,
my brothers, and my brothers-in-law. Most of all, I would like to appreciate the help of
my wife, Yunhee, and my daughter, Seungbin. My wife, Yunhee, has encouraged me to
concentrate on my dissertation. In addition, I owe a lot of things to my daughter,
Seungbin, because I was neglectful in taking care of her. Without their help and
encouragement, I might not make my dissertation done.

I also like to give my gratitude to my Korean friends at Michigan State University.
Especially, the Korean students at the department of economics gave me a lot of useful
comments and supports. The members of the Korea-university alumni at Michigan State
University have always supported and encouraged me when I have been stressed with my
research.

Finally, I dedicate this dissertation to all of those whomiI love and to all of those

who have loved me.

TABLE OF CONTENTS

LIST OF TABLES ............................................................................... viii
Chapter 1: Online Business with Cross-Border Shopping .................................................. 1
1.1 . Introduction .............................................................................................................. 1
1.1.1.Basic Idea .......................................................................................................... 1
1.1.2. Literature Review .............................................................................................. 4
1.2. Theoretical Background ........................................................................................... 6
1.3. Empirical Analysis ................................................................................................. 13
1 .3.1 . Introduction ..................................................................................................... 13
1.3.2. Model Specification and Data ......................................................................... 14
1.3.2.1. Data .......................................................................................................... 14
1.3.2.2. Model Specification ................................................................................. 21
1.3.3. Empirical Analysis .......................................................................................... 23
1.3.3.1. Empirical Analysis Without Population Limits ....................................... 23
1.3.3.2. Empirical Analysis with Population Limits ............................................. 27
1.3.3.3. Probit Analysis ......................................................................................... 36
1.3.3.4. Empirical Analysis with Inflation level ................................................... 41
1.4. Conclusion ............................................................................................................. 48
Chapter 2: Does Sales-Tax Evasion on Remote Sales Reduce Welfare? ......................... 51
2.1 . Introduction ............................................................................................................ 51
2.2. Theoretical Background ......................................................................................... 56
2.2.1. Specification of the Model in the Case of Sales-Tax Evasion and Avoidance.
................................................................................................................................... 59
2.2.2. Specification of the Model for Uniform Sales-Tax Rate. with no Evasion or
Avoidance. ................................................................................................................ 60
2.2.3. Comparison of the current sales-tax system with the new sales-tax economy 62
2.3. Data ........................................................................................................................ 64
2.4. Computation of Welfare Change ........................................................................... 66
2.4.1. Cobb-Douglas Utility Function ....................................................................... 66
2.4.1.1. Computation of the excess burden when tax evasion for remote sales is
widespread ............................................................................................................ 67

vi

2.4.1.2. Computation of welfare when state government can impose sales tax on

remote sales ........................................................................................................... 68
2.4.1.3. Computation of the welfare change in C-D utility function .................... 70
2.4.2. CES Utility Function ....................................................................................... 73

2.4.2.1. Computation of the excess burden when tax evasion for remote sales is
widespread ............................................................................................................ 77

2.4.2.2 Computation of the excess burden when state govemment can impose

sales tax on remote sales ....................................................................................... 78
2.4.2.3 Computation of the welfare change in CBS utility function ..................... 81
2.5. Conclusion ............................................................................................................. 85
CHAPTER 3: CONCLUSION AND POLICY IMPLICATION ..................................... 88
APPENDIX ....................................................................................................................... 90
BIBLIOGRAPHY ............................................................................................................. 98

vii

LIST OF TABLES

Table 1.1 Summary Statistics (county level) .................................................................... 20
Table 1.2 Fixed-Effects Estimation (No Population Limit) .............................................. 25
Table 1.3 The Partial Effects of Group-Dummy Variables .............................................. 29
Table 1.4 Fixed-Effects Estimation (Population Limit=1,000,000) ................................. 33
Table 1.5 Fixed-Effects Estimation (Population Limit=2,000,000) ................................. 34
Table 1.6 Fixed-Effects Estimation (Population Limit=3,000,000) ................................. 35
Table 1.7 Probit Estimation (Population Limit=l ,OO0,000) ............................................. 38
Table 1.8 Probit Estimation (Population Limit=2,000,000) ............................................. 39
Table 1.9 Probit Estimation (Population Limit=3,000,000) ............................................. 40
Table 1.10 Fixed-Effects Estimation (No Population Limit & Pccpi) ............................. 44
Table 1.11 F ixed-Effects Estimation (Population Limit=1,000,000 & Pccpi) ................. 45
Table 1.12 Fixed-Effects Estimation (Population Limit=2,000,000 & Pccpi) ................. 46
Table 1.13 Fixed-Effects Estimation (Population Limit=3,000.000 & Pccpi) ................. 47

Table 2.1 Michigan Sales and Use Tax Revenue Loss from Consumer Remote Sales... 53

Table 2.2 Michigan Sales and Use Tax Revenues, and Tax Loss Estimates .................... 53
Table 2.3 Estimates of the Sales Amount of Each Good (thousands of dollars) .............. 64
Table 2.4 The values of as in OD Utility Function ......................................................... 68
Table 2.5 Computation of Welfare Change in CD Utility Function ............................... 72

Table 2.6 Computation of Welfare Change Using the CBS Utility Function (Tax
Elasticity=2.3) ........................................................................................................... 80

Table 2.7 Computation of Welfare Change Using the CES Utility Function (Tax
Elasticity=3.4) ........................................................................................................... 80

viii

Table 2.8 Computation of Welfare Change Using the CES Utility Function (Tax

Elasticity=4.3) ........................................................................................................... 81
Table 2.9 Computation of Welfare Change Using the CES Utility Function (Population

Limit=2,000,000) ...................................................................................................... 82
Table A] Weighted Average Sales Tax Rates ................................................................. 90
Table A2. Populations of MSAs ...................................................................................... 96

ix

Chapter 1: Online Business with Cross-Border Shopping

1.1. Introduction

1.1.1. Basic Idea

We expect that when a consumer lives in a jurisdiction with a high sales-tax rate,
she/he would be more likely to make online purchases (Goolsbee. 2000). Consumers
have an incentive to buy goods over the Internet, because they can evade sales and use
taxes easily, since state and local governments do not have reliable methods to enforce
their sales and use taxes. In spite of the fact that these purchases are officially subject to
tax in 45 states, there is widespread tax evasion in online purchases.

Subnational governments in the US depend heavily on general sales taxes. The
general sales taxes are roughly one-third of the total tax revenues for state governments,
around 14 percent of local government tax revenues, and one-fourth of total state and
local tax revenues (US. Census Bureau, 2001). However, the recent growth of electronic
commerce generates various problems for state and local governments, such as use-tax
evasion and erosion of the sales-tax base.

Currently, state and local governments cannot require online firms to collect sales or
use taxes if they have no physical presence (nexus) in the originating jurisdiction.‘ For
example, if Dell.com sells a computer to a resident of Texas. the state of Texas can
require Dell to collect Texas sales taxes, because Dell is headquartered in Texas.

However, when Dell.com in Texas sells a computer to a resident of Michigan, the state of

 

' Supreme Court rulings in 1967 and I992 (Fox and Murray. I997) held that states don’t have the power to
collect taxes on out-of-state mail-order companies that sell into their states (These rulings came in the cases
of National Bella Hess v. Department of Revenue of the State of Illinois and Quill C orp. v. North Dakota).

Michigan cannot require Dell computer to collect Michigan sales taxes. When a
Michigan resident buys a computer over the Internet, he is supposed to report that
purchase to the tax authorities of the state of Michigan, and pay the use tax. In practice,
however, this occurs rarely.2

The National Governors Association (NGA) has forecast that there may be more
than $300 billion of electronic commerce by 2002. They estimate that states may lose up
to $20 billion per year in tax revenues. Forrester Research Corporation (Williams et al.,
1999) forecasts $184 billion of electronic commerce in 2004. Bruce and Fox (2000)
estimate that there will be $10.8 billion in sales tax revenue losses from e-commerce
nationwide in 2003. So, the revenue loss is potentially a serious problem to the state and
the local governments,3 even though there is disagreement about the precise extent of the
revenue loss.

Cross-border shopping is a form of tax avoidance, which consumers use for reducing
their sales-tax burden. If a consumer buys goods in his own jurisdiction, it is generally
difficult to evade the sales taxes. When he purchases a good in a neighboring state or
county by crossing the border, it is also usually difficult to evade the sale taxes levied by
neighboring jurisdictions. However, if the foreign local government has a lower sales-tax
rate than his home jurisdiction, and if this tax differential dominates the traveling costs,

the consumer may have an incentive to cross the border.4

 

2 However, in 2003, several large retailers announced that they would begin to collect state sales tax on
their online sales. See Mike Wedeland (2003).

3 Due in part to the erosion created by online purchases, the relative importance of the sales tax has
decreased. From fiscal 1947 to fiscal 1997, general sales taxes produced more revenue than any other state
tax (Mikesell, 2000). However, the general sales taxes were less than the individual income taxes from
1998 to 2001 in state government revenues, and the gap between them rose from $5 billion in 1998 to $28
billion in 2001.

‘ Ferris (2000) analyzes the reasons for the cross-border shopping of Canadian residents to US. markets,
and suggests that the price differentials resulting from the tax differential between Canada and the US.

Residents of each state can purchase private goods in any of three ways. First, they
can buy goods in their own jurisdiction, paying sales taxes if the jurisdiction levies a sales
tax. Second, consumers can travel across a border and purchase products in different
jurisdictions, paying the sales taxes of the bordering jurisdiction as well as the traveling
costs associated with the border crossing. Finally, they might buy over the Internet after
incurring shipping fees and information costs. As mentioned above, cross-border
shopping is a sort of tax avoidance. On the other hand, Internet shopping is encouraged
by the self-reporting feature of the use tax. This is a kind of tax evasion.

There are three information problems with online sales. First is the uncertainty about
whether the credit card will be used improperly. Second is the fact that the consumer does
not actually get to inspect the item at the moment of purchase. When it arrives, it may not
be satisfactory, which would mean it would have to be returned, which is expensive.
Third, there is the uncertainty about whether the item will be delivered in a timely
manner. If the consumer perceives that the benefit of evading sales taxes is greater than
the information costs, he would have an incentive to buy over the lntemet. Although
private goods may have the same qualities in the three markets, their effective consumer

prices will be different, in general.

 

play a major role in Canadian residents’ crossing the border and purchasing liquor and tobacco in US.
markets.

1.1.2. Literature Review

In spite of the increasing recognition of the importance of these issues, few articles
analyze them economically. Fox and Murray (1997) conclude that electronic commerce
should be taxed and that a destination-based tax system is appropriate for the sales-tax,
even though a source-based tax system has clear compliance and administrative cost
advantages over a destination-based tax system.5 Mche (1997) proposes that all
electronic sales should be taxed except for electronic sales to business. Weiner (1997)
explains the possibility of taxing electronic commerce from the point of view of three
features such as nexus, administration, and legislation. Hellerstein (1997) argues that a
uniform taxation approach should be adopted, to cure many of the problems of taxing
electronic commerce. However, their analyses are very normative and these papers are in
the nature of surveys.

Goolsbee and Zittrain (1999) conclude that the potential revenue loss associated with
electronic purchases is exaggerated by the press. They also examine the costs and
benefits of enforcing taxes on online purchases. Goolsbee (2000) performs an empirical
test of the relation between sales-tax rates and lntemet purchases, using data for 1997
from Forrester Research (a market research company in Massachusetts),6 and finds a
positive relation between them. In other words, according to the results of Goolsbee,
consumers who live in high sales-tax-rate locations are more likely to buy over the

Internet, and lntemet purchases are highly sensitive to local taxation.

 

5 The destination-based system is to impose the tax at the site of consumption, and the comprehensive tax
base is domestic consumption. The source-based system is to impose the tax where production takes place,
and the comprehensive tax base is domestic consumption plus net exports. (Fox and Murray, 1997)

6 This is a nationally representative survey of more than 110,000 US. households, and it includes various
demographic characteristics and information on online purchases.

Goolsbee’s paper is a pioneering work in the economic analysis of the electronic
commerce, but there are some limitations. He has restrictive residence information about
consumers. He only knows which metropolitan area a consumer lives in. He does not
have information about the city or county which a consumer lives in. For example, he
assumes that the sales-tax rate in the suburban area of Chicago is the same as that of
Chicago’s downtown. This might lead to measurement error for the tax rates in
Goolsbee’s regression analyses. I use a data set in which information on the county of
residence is available. In order to avoid the measurement error problem, I define the
sales-tax rate at the county level, and use the weighted average sales-tax rates of counties
which consumers live in.

Goolsbee only considers the tax evasion aspect of Internet shopping, and he does not
deal with cross-border shopping in the same framework. However, cross-border shopping
is an important issue, because consumers make substantial use of cross-border shopping,
to elude the sales-tax burden of their local jurisdictions. My studies include cross-border
shopping as well as lntemet shopping in the same framework.

Cross-border shopping has long been popular, and has been studied for a long time.
Cross-border shopping occurs in the European countries because of the differences in
their rates of value-added tax. Gordon and Nielsen (1997) estimate that 3.9 billion Danish
Kroner of value added escaped taxation from Denmark in 1992 because of cross-border
shopping.

In the survey of border crossing between the Republic of Ireland and the United
Kingdom by Fitz Gerald (1989), residents living close to the border were most likely to

buy goods in another country with a lower tax rate.

Lockwood (1993) compares the destination-based tax with the origin-based tax in
small and large countries. He proves that only uniform taxation can guarantee the
equivalence of destination and origin regimes. Ferris (2000) estimates the tax revenues
forgone from the introduction of the Goods and Service Tax (GST) in Canada. He points
out that the dramatic rise and fall of cross-border shopping in Canada between 1989 and
1994 was caused by the substitution of the GST for the Federal Manufacturer’s sales tax.

I investigate whether online business can make a state or local government with a
high sales-tax rate lose its tax base, and whether online business is very sensitive to sales-
tax rates. I extend Goolsbee’s model to consider a consumer’s residence, and examine
how the differences in the relative prices of online shopping and cross-border shopping
affect a consumer’s buying behavior. I set up a theoretical model, which assumes
imperfect mobility on the part of the individual consumer, as well as use-tax evasion on
Internet purchases. I test empirically the effect of local sales-tax rates and the consumer’s
geographic traits on the online business. I compare lntemet purchases with the cross-

border shopping effect.

1.2. Theoretical Background

In this model, I consider only the demand side; producers do not play an active role.
The producer price of each good is fixed, and I use the small-country assumption. In

other words, changes in sales-tax rates do not affect producer prices.

A resident can buy goods in local retail stores, or in the stores of another local
jurisdiction with lower sales-tax rates,7 or over the lntemet. It is assumed that online
stores do not collect the tax for the home jurisdiction. In this model, I assume that there is
no risk of being caught engaging in use-tax evasion on lntemet purchases. Instead, I will
add costs of noncompliance (costs of tax evasion and tax avoidance) into the consumer
prices when a resident buys goods either in other counties or over the lntemet.

The producer prices are the same for a good regardless of where the good is
purchased, but their consumer prices are different. I assume a representative consumer, so
we do not need to discuss distributional issues. I assume that the sales-tax rate of the
home county, th, may be greater than, equal to, or less than that of another county, ti. In
this model,8 there is no lump-sum tax, and there are K kinds of goods, which can be
taxed, in addition to labor, L, which is not taxed. The price of labor, w, is one, which
means labor is the numeraire. The sales-tax rates in a jurisdiction are allowed to be
different among goods.9 The model can include the case of local jurisdictions that have a
uniform rate, as well as the case of counties that have non-uniform rates. Important
parameters are
x": vector of net purchases made in the home jurisdiction,

x’: vector of net purchases made in the foreign jurisdiction,

 

7 I use the county as a measure of local jurisdiction in this model, because the data show which county and
state a resident lives in. In some counties, the general sales-tax rates are different among cities. I use the
population-weighted average sales-tax rates of each county, based on the sales-tax rates of the various
cities.

' I will follow Lovely’s model (1995), which analyzes cross—border shopping. However, she does not
consider online businesses. I will add online purchases into the model, and change her model to compare
lntemet purchases with the cross-border effect.

9 Some states have constant rates for a large number of goods, but other states have different sales-tax rates
for different goods. Some states exempt food, but others do not.

x': vector of net purchases made on the lntemet,

I pk: producer price of good k,

t": sales-tax rate in the home jurisdiction,

If: sales-tax rate in the foreign jurisdiction,

T: traveling costs to buy goods in the foreign jurisdiction,

1: information costs to search for goods over the lntemet,

C: shipping fees to purchase goods over the lntemet,

qkh=pk(1 H"): consumer price of good k in the home jurisdiction,
q[(=pk(1+t’))+T: consumer price of good k in the foreign jurisdiction,

qki(=pk) +I+Cz consumer price of good k in online stores.

The important variables and parameters are the consumption vectors (x), producer
prices (p), consumer prices (q), and noncompliance costs (T. C, and 0.") The effective
consumer price in the foreign jurisdiction, q/, is composed of the producer price, pk, and
its sales-tax rate, t’, and the consumer also has to spend traveling costs, T, to buy goods in
the foreign jurisdiction; qfk(=p’k(1+t’))+T. The consumer price for the lntemet sales
consists of the producer price, pk, and information cost, I, and the economic agent should
also pay the shipping fee, C, to purchase goods over the lntemet; q’k( =pk) +1 +C .

In this model, it is assumed that consumer prices rise by exactly the amount of the

sales taxes, and that sales taxes are fully shifted to consumers. However, it is arguable

 

'0 Economists ofien identify three kinds of social costs: efficiency, administrative, and compliance costs. I
include another social cost, noncompliance costs. Compliance costs are the value of resources expended by
taxpayers in meeting their tax obligations (Tran-Nam, et al., 2000), and noncompliance costs are the value
of resources spent by agents in escaping their tax obligations.

whether retail prices will rise by less than the full sales tax when residents of the
jurisdiction can either migrate or cross borders to make purchases (Poterba, 1996). There
is lntemet shopping as well as cross-border shopping, which may induce less-than-full
consumer price increases. However, in this section, my focus is not to estimate the degree
of tax-shift. I assume here that the sales taxes are fully shifted to consumers, and that the
consumer prices increase by the exact amount of the sales taxes.

Traveling costs (T) are one important part of the noncompliance costs. These
traveling costs are a function of distance, d, and the vector of net purchases made in the
foreign state, x/. If a resident is going to cross the border and shop in a foreign state or
county, he must pay traveling costs. T is an increasing function of distance, and non-
decreasing in x’k. The relative size of cross-border shopping depends not only on tax
differentials, but also on consumption technologies, which involve transaction and
transportation costs.11 Thus, I include traveling costs in my model, to consider the causes
of cross-border shopping.

If a resident uses online businesses, he must pay a shipping fee, C, which is a
function of the vector of net purchases made over the lntemet, xk'. In this model, the
shipping fee, C, includes not only explicit shipping fees, but also the costs associated
with late delivery and uncertainty about delivery. We assume that C is a non-decreasing
function with respect to xk'. Information costs, I, include the costs of computers and other
instruments necessary to access the lntemet, and the time to search for lower prices for

goods. The consumer’s problem is

 

1' Scharf (1999) shows that the transaction and transportation costs are very important factors in cross-
border sh0pping.

Max U(X) =U(x," +x,f +x;,...,x,’; +x,{ +x;)
x,{,j=h,f,andi:,k=1,2,3,...,K
(1.1)
K h h K K ‘
s.t. wL =Zx.(p. +0. )+2(x{(pi +t.’ )+T(d.x.’ ))+Z(xl(p. +I)+C(xl))
Ic=I k=l k=l

xf,x{,andx,: 20

where U(.) is a twice differentiable, strictly quasi-concave utility function, and L is labor
supply. Let U! be the first partial derivative of utility with respect to xr’i.

The utility function is increasing with respect to the vector of net purchases made,
and traveling costs are a quasi-convex fimction of the vector of net purchases made in the
foreign state or county and the distance between the consumer’s residence and the retail
stores in the foreign jurisdictions. The shipping fee is increasing with respect to the

quantity of lntemet purchases. The Lagrange function for this problem is

L = U(X)+2(W -i((x:(p. +tl') + xttp. +t.”)+ thtx.’ )+ x}. (p. + 1)+ Col») (12)

k=l

I differentiate this function with respect to xk”, x/, and x," to maximize this function. I
use Kuhn-Tucker conditions, because corner solutions can exist in this problem. In other
words, consumers may choose to make zero purchases by one or more of the methods.

The Kuhn-Tucker conditions for the consumer problem are

10

U,f —/l(pk +zf)so,x;' 20,(U: -/l(pk +t:))x,’; :0
U.’ mp. +r.’ +T.)50.x{ 20401—16. +r.’ +T.))x,.’ =0 (1.3)

U; —-/i(pk +1+C')SO,x," 20,(U,: -/l(pk +1+C'))x,: =0

T2 is the partial derivative of traveling cost with respect to x/, and C ' is the first
derivative of the shipping fee with respect to xr'. A is the Lagrange multiplier in this
problem, and is interpreted as the marginal utility of income. If we assume that a
consmner gains the same marginal utility from the good, regardless of whether it is
purchased through his local jurisdiction stores, or from foreign jurisdiction stores, or from
the lntemet, the first partial derivatives of the utility function are the same. That is,
Ukh= Urf= Uk'. Let U], be the first partial derivative of the utility function.

When the consumer makes his decision about where to buy good k, the differences in

effective prices across the different modes of purchase are taken into account. The

consumer would buy good k in his local jurisdiction’s retail stores if (pk + (If) were less
than (pk +t,‘f +T2) and (p, +1 +C’). He would buy it in the foreign jurisdiction if
(p, +t,‘f +T2) were less than (pk + if) and (p, + I + C’). He would buy it on the lntemet

if(p, +1+C')were less than (pk. (If) and (p, +1,f +T2).

If the sales-tax rate were high in the local jurisdiction, a resident would be less likely
to buy good k in home-jurisdiction retail stores. He might buy good k in the foreign
jurisdiction retail stores or over the lntemet. If he is very risk averse, that is, if he hates
revealing his credit-card number, he would be more likely to buy good k in the foreign

jurisdiction than over the lntemet. If the shipping fee were very high, he would also be

11

more likely to buy good k in the foreign jurisdiction rather than over the lntemet. If the
traveling costs were very expensive, he would be more likely to purchase over the
lntemet.

If the consumer lives in a jurisdiction adjoining some other local jurisdictions with
lower sales-tax rates, he is more likely to prefer buying the good k in the foreign state or
county, rather than on the Internet. However, if he does not live in a county adjoining
some other local jurisdictions with lower sales-tax rates, he would be more likely to
purchase goods over the lntemet. Thus, the consumer’s geographic characteristics have

an effect on his purchasing behavior. '2

 

‘2 In the original problem, I include the distance as a geographic variable, but it is very difficult in practice
to use the distance as a geographic variable. The data set only provides very rough information about the
location of residents-which state and county they live in. Thus, distance between the home county and the
foreign county would be measured with considerable error, and I choose not to include it as an explanatory
variable.

12

1.3. Empirical Analysis

1.3.1. Introduction

I analyze the data by county to focus on group effects.l3 Since there are some group
effects, which cannot be explained explicitly, we have to capture the specific
characteristics of counties. For example, some governments depend heavily on general
sales taxes, but others do not. In order to capture the effects of different characteristics of
all jurisdictions, I use group dummies and fixed-effects estimation.

State governments differ in how they enforce the collection of the sales and use tax.
Nevada, South Dakota, Texas, and Washington do not have either a state individual
income tax or a state corporate income tax, so their counties depend relatively heavily on
the sales and use tax for their budgets.” These states may enforce the sales and use tax
strongly, which may give these states’ residents an incentive to rely on online businesses.

Some states rely heavily on the general sales tax, but others do not. Florida, Nevada,
South Dakota, Tennessee, Texas, and Washington have ratios of general sales tax
revenues to total tax revenues of 49% or more. Nevada. South Dakota, Texas, and
Washington do not have state individual and corporate income taxes.” Florida and
Tennessee do not have a state individual income tax. Therefore, these states must rely
heavily on the sales tax for financing their expenditures. Differences in the importance of

the sales tax may lead to differences in the degree of enforcement of the sales tax.

 

'3 The analysis could be carried on at the state level, rather than at the county level. However, this would
make it impossible to study cross-border shopping in the desired way. Virtually all residents of the United
States are close to a county boundary. However, many are relatively far from the nearest state border.

" The average ratio of general sales taxes to total taxes in all state governments is about 30 percent, but the
ratios in Nevada, Texas, and Washington are above 50%.

'5 I get information about state finance structures from http://taxfoundation.org/statefinancehtml.

13

Even counties in the same state have different atmospheres. Some counties have
well-equipped lntemet infrastructures, and other do not. Generally, the counties in large
metropolitan areas have well-equipped lntemet background, and their residents would be
more likely to use lntemet shopping, all else equal. Macomb County, Michigan, is a
suburban region of the Detroit metropolitan area, and its residents may be more likely to
use lntemet shopping than those in other counties in the state, such as Schoolcraft
County, which is geographically more distant from major metropolitan areas.

I include group effects in the regression analysis to capture the effects mentioned
above. I estimate the effects of sales-tax rates and consumers’ geographic characteristics

on lntemet purchases at the county level.

1.3.2. Model Specification and Data

1.3.2.]. Data

I use CPS (Current Population Survey) data for 1997 and 2001.16 The Current
Population Survey (CPS) is a monthly survey of about 50,000 households conducted by
the Bureau of the Census for the Bureau of Labor Statistics. The data contain information
for each individual, such as household income, gender, race, education level, age, and
marital status. The data set provides information on the geographic location of each
individual, as well as whether she or he purchases goods over the lntemet. The complete
sample includes 266,160 observations, but many of these are not usable for the

econometric analysis.

 

'6 http://www.bls.census.gov/cps/cpsmain.htm

14

I select only those individuals who have access to the lntemet, because, if I were to
use the whole data set, the estimates could suffer from serious sample-selection bias. That
is, for some people who do not purchase goods over the lntemet, the reason is that they
do not have computers in their homes, or do not have access to the lntemet by
themselves. These people cannot purchase goods over the lntemet because they do not
have infrastructure to access the lntemet. They have to buy goods at their “brick-and-
mortar” stores, even if the sales-tax rates are very high. The factor that influences their
consumption behavior is not a sales-tax rate or any demographic and geographic variable,
but the infrastructure of lntemet access. Therefore, I choose the individuals who have
lntemet access, and analyze them at the county level. Since I exclude those who do not
have lntemet access, I lose many observations. The sample that remains after excluding
those without lntemet access consists of 91,227 observations.

In addition, the CPS data do not report the county names for all individuals. I
cannot identify the exact county names of some individuals, so I cannot find the exact
sales-tax rates of those people. I also exclude those for whom county of residence is not
available. After excluding those without lntemet access as well as those for whom the
county could not be identified, the sample is reduced to 35,667 observations.

Education level and marital status are important explanatory variables. However,
these variables are not available for everyone in the sample. The CPS reports the
education level and marital status only for those who are 15 years old or older. Therefore,
I also exclude those who are less than 15 years old. After excluding those without
lntemet access, those for whom the county could not be identified, and those younger

than 15 years, the sample is reduced to 30,911 observations.

15

The dependent variable is a binary variable, measuring whether the individual
engages in online purchases. In other words, if the consumer buys goods online, the
dependent variable is one; if not, it is zero.17 I use household income, gender, race,
marital status, education level, age, and the number of computers in a household as
explanatory variables.

The key explanatory variables are the sales-tax rate and the geographic variables.
The sales-tax rate at the county level is a weighted average sales-tax rate, based on the
population ratios of cities and towns to the county. Sales-tax rates are measured in
percentage points.

The states fall into four categories, with respect to the way in which they impose a
retail sales tax.18 First, Delaware, Montana, New Hampshire, and Oregon'9 do not impose
sales tax on purchasing goods. Second, some states impose a uniform sales-tax rate.20 In
these cases, I do not need to calculate a weighted average sales-tax rate.

Third, there are some states with variations of sales-tax rates across counties, but no
variation within each county (Florida, Georgia, Nevada, North Carolina, Pennsylvania,
South Carolina, Wisconsin, and Wyoming). For these states, I do not need to calculate a

weighted average sales-tax rate at the county level.

 

'7 This variable has some weaknesses-it would be better if we knew not just whether they use the lntemet
for purchases, but how much they spend in online purchases. In this data set, a person who spends $10 per
year via lntemet is treated as identical to one who spends $ 1000 per year.

" Some 45 states and Washington D.C., impose general sales taxes, and 34 states allow their local
governments to impose their own sales taxes (Comia er al., 2000)

9 Alaska does not have a general sales tax, but its municipalities and counties impose their own sales taxes.
2° They are Hawaii (4%), Virginia (4.5%), Indiana (5%), Maine (5%), Maryland (5%). Massachusetts (5%),
Vermont (5%), the District of Columbia (5.75%), Connecticut (6%), Kentucky (6%), Michigan (6%), West
Virginia (6%), Mississippi (7%), and Rhode Island (7%). Mississippi has a uniform sales-tax rate (7%), but
only the city of Tupelo has its own general retail sales tax of 0.25%, which makes its sales-tax rate 7.25%.

16

Fourth, in some states, there are variations in sales-tax rates, both across counties and
within counties. That is, cities within the same county have different sales-tax rates.21 In
this case, I calculate a weighted average sales-tax rate at the county level. I calculate the
population ratio of each city to the county, and then multiply the population ratio of each
city by the city’s sales-tax rate. The weighted sum of the cities’ sales-tax rates is a
weighted average sales-tax rate for the county. In other words, the weighted average

sales-tax rate for county j is defined as

t}. =25, 4' (1.4)
h p}

where

t 1 :weighted average sales - tax rate of county j,
t 1,, : sales - tax rate for city h in county j,

p}, : population of city h in county j, and

p j :population of county j.

The other important explanatory variables are geographic variables, such as the tax
differentials between the home jurisdiction and adjoining jurisdictions, and the sales-tax
ratios of the home jurisdiction to the adjoining jurisdictions.

The tax differential is equal to the sales-tax rate of the home jurisdiction, minus the
minimum sales-tax rate of all adjoining local jurisdictions. The variable is measured in
percentage points, and it can be positive, zero, or negative if the sales-tax rate of a

consumer’s own jurisdiction is above, equal, or below the sales-tax rates of neighboring

 

2' They are Alabama, Alaska, Arizona, Arkansas, California, Colorado, Idaho, Illinois, Iowa, Kansas,
Louisiana, Minnesota, Missouri, Nebraska, New Jersey, New Mexico, New York, North Dakota, Ohio,
Oklahoma, South Dakota, Tennessee, Texas, Utah, and Washington.

17

jurisdictions. The sales-tax ratio of the home jurisdiction to the adjoining jurisdiction is
(1+ the sales-tax rate of home jurisdiction)/(l+the minimum sales-tax rate of adjoining
local jurisdictions).22 I include these geographic variables in separate equations because
they are closely related to each other. If I were to include them all together in the same
regression, the estimates could suffer from multicollinearity.

These geographic variables are included in an attempt to learn about the patterns of
cross-border shopping. The theoretical model presented earlier suggests that the distance
between jurisdictions could have an important effect on cross-border shopping behavior.
The geographic variables are only a rough proxy for distance. These variables are
constructed by comparing the tax rate in the consumer’s home jurisdiction with the
lowest tax rate in any of the adjoining jurisdictions. Thus, it is possible that this
specification misses certain aspects of cross-border shopping. For instance, it is possible
that a consumer may engage in cross-border shopping by going two counties away, rather
than by going to an adjoining county. Also, we treat everyone in a given county the
same, regardless of whether they live just next to the county line, or miles away. This
may be especially important in some western states, where the counties are very large.
Therefore, it should be understood that the geographic variables are not perfect.
However, it can certainly be argued that these variables allow us to study cross-border
shopping in a way that is not possible in the model used by Goolsbee (2000). This is
because Goolsbee assumed that everyone in a given metropolitan area faces the sales-tax

rate that is used in the central city of that metropolitan area.

 

22If I were to use (home jurisdiction sales-tax rate/adjoining jurisdiction sales-tax rate) as a tax ratio, the tax
ratio might be undefined in some cases, because some states do not have general sales taxes.

18

Table 1.1 presents summary statistics for all variables at the county level. The sales-
tax rate at the county level (taxrate) is a weighted average sales-tax rate based on the
sales-tax rates of the various cities and towns in the county, with weights determined by
population shares. Every sales-tax rate is measured in percentage points. For the sample
as a whole, the average sales-tax rate is above 6 percent. Tax differential (taxdifi) is the
difference between the sales-tax rates in the consumer’s home jurisdiction and
neighboring jurisdictions. I subtract the minimum sales—tax rate of all neighboring
counties from the sales-tax rate of the consumer’s home county. For example, Allegheny
County, Pennsylvania has a 7% sales-tax rate in 2001, but adjoining counties in
Pennsylvania have a 6% sales-tax rate. Therefore, the tax differential is 1 for a resident of
Allegheny County.

The sales-tax ratio is (l+t;,)/(l+tf), where I, is the sales-tax rate of the home
jurisdiction and t; is the minimum sales-tax rate among all neighboring jurisdictions. The
residents of Washington, DC, could buy goods at retail stores in counties of Maryland
and Virginia, as well as in Washington, DC. The sales-tax rates in DC, Maryland, and
Virginia are 5.75%, 5%, and 4.5%, so the residents in Washington, DC, would have an
incentive to buy goods in Virginia. The sales-tax ratio (taxratio) for a DC. resident is
(l+0.0575)/(1+0.045), which equals 1.0120. Consumers who live in local jurisdictions
with higher sales-tax rates have sales-tax ratios which are greater than 1. Those who live
in a county with the same sales-tax rate as the sales-tax rates of neighboring counties
have sales-tax ratios which are equal to 1. Those who live in a county with lower sales-
tax rates have sales-tax ratios which are less than 1. Summary information key variable is

given in Table 1.1, for those consumers who are included in the sample.

19

Table 1.1 Summary Statistics (county level)

 

 

All individuals Buyers Non-buyers
N
24,804 10,500 14,304
6.2815 6.2237 6.3239
mm“ (0%) (1.9361) (1.9742) (1.9066)
. .6958 .7255 .6741
Tw‘d’fl (1.5142) (1.5532) (1.4845)
Taxmfi 1.0066 1.0069 1.0064
0 (.0148) (.0152) (.0145)
Household income 58,950.42 61,990.83 56,718.58
(22,930.92) (21,916.30) (23,398.66)
Fem [e .4973 .4986 .4964
a (.5000) (.5000) (.5000)
White .8465 .8637 .8339
(3605) (3431) (3722)
Married .5833 .6160 .5594
(4930) (4864) (4965)
Education 14.52 15.03 14.14
(2.8576) (2.7626) (2.8676)
# ofcom mm 1.4701 1.5844 1.3863
P (.6939) (.7483) (.6382)
38.51 39.16 38.03
Age
(14.48) (13.11) (15.40)

 

The numbers in parentheses are standard deviations. The household income is measured
in dollars. Education and age are measured in years.

1.3.2.2. Model Specification

I use fixed-effects estimation to capture the group-specific effects of sales-tax rates

and geographic features. The main model is

yij=fl0 + )Bltaxraleg' + flzgeo), + ,8 31inc0mea- + flfemaley- + fl5whitey- + fl6marriedjj
+ fizfemaredy- + ,nghmaredg- + flgfewhiteij + flmfewhmary + ,6, ,highgrady-
+ ,Bncolgradij + ,Bz3progradg- + ,614nchg- + fll5age15y- + fl16089200+fll7age300 +

fllgage50ij + ,Blgage60,-) + fl20d01+ a ,-+ ug- (1.5)

where

femared=female X married
whmared=white X married
fewhite =female X white

fewhmar =female X white X married

The i subscripts refer to the local jurisdiction, that is, county. The j subscripts refer to
the individual consumer in county i. The dependent variable. y. is whether a consumer
purchases goods over the lntemet. taxrate is the weighted average sales—tax rate in the
county in which a consumer lives. geo includes the geographic variables for the
consumer. It includes taxdifl and taxratio at the county level. taxdifl is the tax differential

between the sales-tax rate of the consumer’s home county and the minimum sales-tax rate

21

of his neighboring counties. taxratio is the sales-tax ratio of the individual’s home county
to the adjoining county with the lowest sales-tax rate.

[income is the log of the resident’s household income. This variable is included for
several reasons. First, [income may be viewed as a proxy for aspects of human capital
that are not captured by other variables, such as education and age. Second, [income may
be a proxy for the opportunity cost of time. In this respect, it would be valuable to use
data on wage rates, rather than income data. However, for a substantial portion of the
sample, wage rate data are not available.

female, white, and married are all binary variables to explain the individual’s
demographic features.23 femared is an interaction term of female and married, and
whmared is also an interaction term of white and married. fewhite is an interaction of
female and white, and fewhmar is an interaction term of female, white, and married.
highgrad, colgrad, and prograd are all education-dummy variables. If an individual has
graduated from high school, highgrad is 1, or 0 if not. If a consumer has graduated from
college, colgrad is 1. If an individual has received an MA, Ph.D, or professional degree,
prograd is 1. nch is the number of computers in the household, which represents the
lntemet infrastructure of a consumer. agel5, age20, age30. age50, and age6024 are age-
group-dummy variables, and consumers in their 405 are a base group. d0] is a yearly
dummy variable, and is one when an observation is taken in 2001. If it is taken in 1997, it

is zero. a,- is the group-specific effect of a county and Uy' is an error term.

 

23 If a consumer is female, female is 1, otherwise 0. If she is married and white, married and white are equal
to l.

2’ agel5 is I if a consumer’s age is from 15 to 19, age20 is I if her age is from 20 to 29, and age60 is I if
her age is greater than or equal to 60.

22

As mentioned earlier, I use fixed-effects estimation to capture the unobserved group-
specific effects of counties. Fixed-effects estimation can consider the specific and inborn
effects of counties, and capture the characteristics within a group as well as across
groups.25

We shall see that the results for the tax variables are somewhat sensitive with respect
to the size of the metropolitan areas from the observations are chosen. This may be
because large metropolitan areas have better developed transportation systems, which
facilitate cross-border shopping. The large metropolitan areas may also have different
lntemet infi'astructure or different attitudes toward tax evasion, which may lead to
different degrees of responsiveness to tax rates. Thus, to begin with, I report the
estimation results for the entire sample. After that, I restrict the sample to those who live
in metropolitan areas whose p0pulations are above one, two, and three million, and
estimate them separately. Finally, I will compare the estimation results of metropolitan

areas with population limits with that of the whole sample.

1.3.3. Empirical Analysis

1.3.3.1. Empirical Analysis Without Population Limits

In this subsection, I analyze the effects of sales-tax rates and geographic variables on
online purchases when I use all observations. Table 1.2 shows the effects of demographic

and geographic variables on online purchases if I use all observations. Table 1.2

 

2’ The explanatory variables should be related to group effects. Therefore, fixed-effects estimation is more
appropriate for this model than random-effects estimation. In order to check this, I use Hausman
specification test.

23

describes the results of fixed-effects estimation with robust standard errors at the county
level. The first column is the estimation result when I include the sales-tax differential
(taxdifi') as a geographic variable, and the second column is the result with the sales-tax
ratio (taxratio). I report the robust standard errors adjusted for cluster sampling
(Wooldridge, 2001), allowing that the observations are independent across counties
(clusters), but not necessarily independent within counties (clusters). In this analysis, the
sales-tax rate (taxrate) coefficient is positive, but is not significant in either specification.
A consruner who lives in a county with a high sales-tax rate would be more likely to
make online purchases, all else equal. When the sales-tax rate increases by one
percentage point, all other variables held constant, the probability that a consumer would
make online purchases will increase by 0.41% if I include taxdifl in the model. When
taxratio is used as a geographic variable, the probability will increase by 0.42% if taxrate
increases by one percentage point.

The coefficients on the geographic variables are all positive, which is contrary to my
expectation, and taxdifi’ and taxratio are not statistically significant in separate
regressions. When I use all observations, the coefficient estimates of the geographic
variables do not give strong results.

The coefficient estimate of lincome is positive and statistically significant in both
regressions. When a consumer’s income increases by 1%, the probability that she would

make online purchases will increase by .0003 (03%).26

 

26

fl = Ay
100 %Aincome

 

, which means that y is increasing by ,6/100 when income increases by 1%.

24

Table 1.2 Fixed-Effects Estimation (No Population Limit)

 

Taxrate
Taxdifl
Taxratio
Lincome
Female
White
Married
F emared
Whmared
F ewhite
F ewhm ar
Highgrad
Colgrad
Prograd
Nch
Age I 5
Age20
A ge3 0
Age50
A ge60
DO]

N

# of groups
R2

Tax Elasticity

Dependent Variable: whether she or he makes online purchases

0.0041
(0.0313)
0.0059
(0.0448)

0.0326'”
(0.0064)
-0.0496“‘
(0.0169)
0.0736‘”
(0.0209)
-0.0320
(0.0235)
0.0166
(0.0225)
0.0029
(0.0245)
0.0261
(0.0204)
0.0183
(0.0283
0.0783 "
(0.0132)
0.1007‘“
(0.0078)
0.0127
(0.01 17)
0.0837‘“
(0.0053)
-0.1 161‘”
(0.0160)
0.0100
(0.0105)
0.0453‘“
(0.0089)
-0.0251”‘
(0.0093)
-0.1 196‘”
(0.0137)
0.3760‘”
(0.01 19)

24,804
215
0.1740
0.0612

0.0042
(0.0306)

0.6503
(4.4428)
0.0321‘”
(0.0064)
-0.0496”’
(0.0170)
0.0739’"
(0.0209)
00318
(0.0235)
0.0166
(0.0226)
0.0027
(0.0245)
0.0259
(0.0205)
0.0185
(0.0283
0.0776 ”
(0.0132)
0.1003‘”
(0.0078)
0.0129
(0.0117
0.0835 ‘°
(0.0053)
-0.1166"‘
(0.0160)
0.0101
(0.0105)
0.0460'”
(0.0089)
.0.0260‘”
(0.0093
-0.1194 “
(0.0138)
0.3761‘“
(0.0119)
24,804

215
0.1740
0.0627

 

The numbers in parentheses are robust standard errors, adjusted for cluster sampling. w, W, and I denote
significance at the 1%, 5%, and 10% levels respectively. # of groups is the number of counties represented

in the sample.

25

The partial effect of female is -.0496 + .0166 X married + .0261 X white +
.0183 X white X married if a consumer is female. In the case of white, its partial effect is
roughly .0736 + .0029Xmarried + .0261 X female + .0183 X femaleXmarried. The
partial effect of married is -.0320 + .0166 X female + .0029thite +

.0183 X female X white.

This estimation ”equation implies that education-dummy variables have positive
effects on online purchases. These coefficients for highgrad and colgrad are statistically
significant. The coefficient estimate of nch is positive and statistically significant in both
regressions. If a consumer owns one more computer at her household, the probability of
online purchases is predicted to increase by about 8%. The age dummy variables have
different signs and coefficient estimates. Compared to consumers in their 405, consumers
from 15 to 19 and those in their 50s and 60s would be less likely to make online
purchases. Teenage people generally do not have credit cards which are necessary to
online purchases, and old people prefer general retail shopping to online shopping. Old
people may have greater concern about revealing their credit card numbers over the
lntemet, and a strong preference for inspecting goods before purchasing. Consumer in
their 208 and 305 prefer e-commerce, compared to people in their 40s. The age-group
dummy variables are statistically significant, except for age 20.

The yearly dummy variable, d0], is positive and statistically significant, which

explains the current trend. Year to year, more consumers engage in e-commerce, and the

26

number of transactions in e-commerce is increasing. The tax elasticities27 in the model
are .0612 and .0627 respectively. If the tax elasticity were very large, the state or local

governments would lose their sales-tax revenues when they increase their sales-tax rates.

1.3.3.2. Empirical Analysis with Population Limits

In this subsection, I analyze the effects of sales-tax rates and geographic variables on
online purchases when I place some restrictions on the sample. I choose observations for
those in metropolitan areas with populations of at least one million, two million, and
three million. Table 1.4, Table 1.5, and Table 1.6 show the effects of demographic and
geographic variables on online purchases when I use different population limits. I use
fixed-effects estimation at the county level,28 and choose metropolitan areas whose
populations are more than one, two, and three million. The first column in each table is
the estimation result when I include the sales-tax differential (taxdifi) as a geographic
variable, and the second column is the result with the sales-tax ratio (taxratio).

Unlike the estimation results without a population limit, these estimation results with
population limits provide expected signs and significant coefficients for the sales-tax rate

(taxrate), the sales-tax differential (taxdifi’), and the sales-tax ratio (taxratio). The

 

6y taxrate

ataxrate y

that I use a binary variable for y. Ideally, y should be how much a consumer purchases online, to measure
the tax elasticity correctly. However, this information is not available in my data set, and I use this binary
data for the tax elasticity.

2' Fixed-effects estimation assumes that the unobserved effect (a,) is correlated with each explanatory
variable. Random-effects estimation assumes that the group specific effect ((1,) is uncorrelated with each
explanatory variable. We can use the Hausman test to determine which estimation method is appropriate.
According to the Hausman test, random-effects estimation is inconsistent, but fixed-effects estimation is
consistent. Therefore, fixed-effects estimation is more appropriate for this model than random-effects
estimation (Wooldridge, 2001).

 

27 The tax elasticity is defined as . The weakness of my measure of the tax elasticity is

27

coefficient estimates of taxrate in all population limit cases have positive signs, and they
are significant, marginally significant, and insignificant in the three million, two million,
and one million case respectively. When the population limits increase, the coefficient
estimates and t-statistics of taxrate increase together. All coefficient estimates of taxdiff
are negative, and they are statistically significant in the cases of one and two-million
population limits. The coefficient estimates of taxratio are negative in all cases, and they
are statistically significant except for the three-million case. Thus, for the large
metropolitan areas, the sales-tax rate and geographic variables are influential factors in
determining consumers’ shopping patterns.

Let me explain the effect of each variable in the case of the two-million population
limit (Table 1.5). The coefficient estimate of taxrate is about .0581, which means that the
probability that a consumer makes online purchases is predicted to increase by 5.81%
when the sales-tax rate increases by one percentage point. The tables show that the
coefficient estimates of taxrate are increasing when the population limit increases. These
results indicate that, in large metropolitan areas, the sales-tax rate is expected to be an
important factor in consumers’ purchasing patterns. The effects become more
pronounced when we consider those in very large metropolitan areas.

Table 1.5 (two-million case) shows that the coefficient estimate of taxdifl is -.0930,
and is statistically significant at the 1% level. If the tax rate of a consumer’s home
jurisdiction is above that of her neighbor’s jurisdiction by one percentage point, the
probability that she makes online purchases is predicted to decrease by 9.30%. I interpret
this as meaning that a consumer would prefer cross-border shopping to lntemet shopping

when the tax-rate differential increases. The coefficient estimate of taxratio is negative

28

and statistically significant at the 1% level in Table 1.5 (two-million case). It explains
that the sales-tax ratio as a geographic variable plays a significant role in e-commerce.
The interpretation of taxratio is similar to the interpretation of taxdifl.

The coefficient estimates of lincome are positive and statistically significant in all
regressions. Table 1.5 (two-million case) implies that the probability that she would make
online purchases will increase by about .000291 (.0291%) when a consumer’s income
increases by 1%.

The partial effect of female is -.0396 + .0166Xmarried + .0182thite +
.0291 thiteXmarried. In the case of white, its partial effect is roughly .0583 +
.0128Xmarried + .0182 Xfemale + .0291XfemaleXmarried. The partial effect of

married is -.0603 + .0166 X female + .0128 X white +. 0291 X female >< white.

Table 1.3 The Partial Effects of Group-Dummy Variables

 

Base Group: the unmarried, nonwhite, and male

 

the unmarried, nonwhite, and female 003%
the unmarried, white, and male 0.0583
the married, nonwhite, and male -0.0603
the unmarried, white, and female 0.0162
the married, nonwhite, and female -0.0833
the married, white, and male 0.0108
the married, white, and female 0.0351

 

29

For the purpose of comparisons of some dummy variables and interactions terms, I
call the unmarried, nonwhite, and male a base group. The unmarried, nonwhite, and
female would be less likely to make online purchases than the base group by 3.96%, the
unmarried, white and male would be more engaged in lntemet shopping by 5.83%, and
the married, nonwhite, and male would be less likely to use lntemet shopping by 6.03%.
If I interpret the interaction terms, the unmarried, white, and female would be more
engaged in e-commerce than the base group by 1.62%, the married, nonwhite, and female
would be less likely to make online purchases by 8.33%, and the married, white, and
male would be more engaged in lntemet shopping by 1.08%. The married, white, and
female would be more likely to make online purchases by 3.51%. The results are shown
in Table 1.3.

Table 1.5 (two million case) implies that all education-dummy variables have
positive effects on online purchases. The coefficient estimate of nch is positive and
statistically significant in all regressions. Not surprisingly, if a consumer owns one more
computer at her household, she would be more likely to make online purchases. The age
dummy variables indicate that the likelihood of lntemet shopping will increase with age
until the consumer is in her 303. After that, increasing age is associated with declining
use of the lntemet for purchases. Compared to consumers in their 40s, consumers from
age 15 to age 19 and those in their 508 and 605 would be less likely to make online
purchases. Teenage people generally do not have credit cards which are necessary for
online purchases, and old people prefer general retail shopping to online shopping. Old
people may have greater concern about revealing their credit card numbers over the

lntemet, and a strong preference for inspecting goods before purchasing. Consumers in

30

their 208 and 308 prefer e-commerce, compared to people in their 405. Except for the age-
group dummy for the 205, the age-group dummy variables are statistically significant.

The yearly dummy variable, d0], is positive and statistically significant, which
explains the current trend. Recently, the number of consumers who engage in e-
commerce has increased dramatically.

If I compare the rich ($80,000), white, unmarried, male who graduates from a
college with the poor ($8,000), black, unmarried, female who drops out of high school,
the former person would be more likely to make online purchases than the latter person
by 33.85%.

The tax elasticities in the model are increasing when I increase the population limit.
When I choose metropolitan areas whose population is above three million, the tax
elasticity is above 4 in both regressions. This means that consumers would be more likely
to depend on Internet shopping when their local governments try to increase their tax
rates. In large metropolitan areas, state and local governments would lose their sales-tax
revenues severely by way of online commerce.

In conclusion, when I restrict the sample and choose only those observations from
metropolitan areas whose populations are more than one, two, and three million, the
sales-tax rate (taxrate) is positive and statistically significant. If I increase the population
limit, the estimated coefficient increases. Consumers who live in metropolitan areas with
a lot of population are very sensitive to the sales-tax rate in making online purchases.
They also consider geographic variables such as the sales tax-rate differential (taxdifl)
and the sales-tax ratio (taxratio), when deciding whether to make online purchases. When

her own jurisdiction has a higher sales-tax rate than adjoining jurisdictions, a consumer

31

would be less likely to engage online businesses. I interpret this as evidence that the
consumer would prefer cross-border shopping to online shopping, because of the sales
tax-rate gap between her own jurisdiction and the neighboring jurisdiction. In other
words, when this tax-rate gap dominates transportation costs which are engaged in cross-
border shopping, consumers would have an incentive to purchase goods in retail markets

in neighboring jurisdictions, instead of making online purchases.

32

Table 1.4 F ixed-Effects Estimation (Population Limit=1,000,000)

 

T axrate
Taxdiflr
Taxratio
Lincome
Female
White
Married
F emared
Whmared
F ewhite
F ewhm ar
Highgrad
Colgrad
Prograd
Nch
Age] 5

A ge20
Age3 0
Age50
Age60
DO]

N

# of groups
R’
Tax Elasticity

Dependent Variable: whether she or he makes online purchases

0. 0282
(0. 0400)”
-0 0705
(0.0231)

0.0364‘”
(0.0078)
-0. 0412“
(0. 0189)”
0 0673
(0.0239)
-0.0455‘
(0.0260)
0.0100
(0.0263)
0.0008
(0.0248)
0.0165
(0.0234)
0. 0357
(0. 0322)”
0. 0771
(0. 0163)
0 1043
(0.0097)
0.0147
(0.0145)
0.0855‘”
(0.0065)
-0.1276‘“
(0.0200)
-0. 0005
(o 0122)
0 0483
(0.011 1)
.0. 0407‘“
(0. 0106)”
-0. 1155
(0. 0174)
0. 3918
(0.0156)
16,014

107
0.1887
0.4553

0.0285
(0.0383)

-9 8426‘”
(2 2541)
0 0364
(0.0078)
-0. 0410"
(0. 0189)
0.0676
(0.0240)
-0.0453’
(0.0260)
0.0096
(0.0263)
0.0006
(0.0248)
0.0164
(0.0234)
0 0361
(0 0322)
0. 0770
(0 0163)
0 1043
(0.0097)
0.0147
(0.0146)
0.0855'”
(0.0065)
-0.1278’”
(0.0199)
-0 0007
(0. 0122)"
0 0482
(0.0111)
-0. 0407‘”
(0 0105)
-0. 1154
(0. 0174)
0. 391 1
(0.0155)
16,014

107
0.1888
0.4602

 

The numbers in parentheses are robust standard errors, adjusted for cluster sampling. "r, ", and ' denote
significance at the 1%, 5%, and 10% levels respectively. # of groups is the number of counties represented

in the sample.

33

Table 1.5 Fixed-Effects Estimation (Population Limit=2,000.0()O)

 

Taxrate
Taxdtflr
Taxratio
Lincome
Female
White
Married
F emared
Whmared
F ewhite
F ewhm ar
H ighgrad
Colgrad
Prograd
Nch

A ge15
Age20
Age3 0
Age50

A ge60
DO]

N

# of groups
R’
Tax Elasticity

Dependent Variable: whether she or he makes online purchases

0.0581
(0. 0492)"
-0. 0930
(0.0206)

0.0291‘”
(0.0091)
4). 0396‘
(0. 0206)
0. 0583
(0.0266)
-0.0603“
(0.0275)
0.0166
(0.0309)
0.0128
(0.0258)
0.0182
(0.0264)
0.0291
(0.0384)
0. 0800‘"
(0. 0167)
0. 1136
(0. 0106)
0. 0325
(0. 0150)
0 0888
(0.0076)
-0.1099“‘
(0.0245)
0. 0023
(0 0149)
0. 0504
(0.0130)
-0.0439”‘
(0.0125)
-0.1212“‘
(0.0195)
0.3947‘“
(0.0207)

1 1,341
69
0.1898
0.9208

0.0570
(0.0495)

.9.5537‘”
(2.0855)
0.0291‘”
(0.0091)
-0. 0396‘
(0. 0206)
0 0583
(0.0266)
0.0604“
(0.0275)
0.0166
(0.0309)
0.0128
(0.0258)
0.0182
(0.0264)
0. 0291
(0 0384)
0.0800
(0 0167)
0 1136
(0.0106)
0.0325
(0. 0150)”
0 0888
(0.0076)
-0.1099‘“
(0.0245)
0. 0023
(0. 0149)”
0.0504
(0.0130)
-0. 0439‘”
(0 0125)
.01212
(0.0195)
0.3946‘“
(0.0207)
1 1,341

69
0.1898
0.9034

 

The numbers in parentheses are robust standard errors, adjusted for cluster sampling. “r, ", and ' denote
significance at the 1%, 5%, and 10% levels respectively. # of groups is the number of counties represented

in the sample.

34

Table 1.6 Fixed-Effects Estimation (Population Limit=3,000,000)

 

Dependent Variable: whether she or he makes online purchases

 

Taxrate 0.2849" 0.2851"
(0.1187) (0.1185)
Taxdifl -00243
(0.0394)
Taxratio -2.5238
(4.0614)
Lincome 0.0293" 0.0293"
(0.0118) (0.0118)
Female -0.0420‘ -0.0420‘
(0.0225) (0.0225)
White 0. 0298 0.0298
(0. 0267) (0 0267)“
Married -0. 0767 -0 0767
(0.0296) (0.0296)
F emared 0.0329 0.0329
(0.0381) (0.0381)
Whmared 0.0283 0.0283
(0.0325) (0.0325)
Fewhite 0.0235 0.0235
(0.0363) (0.0363)
Fewhmar 0.0163 0.0163
(0.0548) (0.0548)
Highgrad 0.0777‘” 0. 0777‘“
(0.0199)” (0 0199)
Colgrad 0.1171 0.1171
(0.0134) (0.0134)
Prograd 0. 0408” 0.0408”
(0. 0207) (0.0207)”
Nch 0 0825 0.0825
(0.0100) (0.0100)
Agel5 -01098‘“ -01098‘”
(0.0331) (0.0331)
Age20 0.0021 0.0021
(0. 0182) (0 0182)”
Age30 0. 0638 0.0638
(0.0165) (0.0165)
Age50 -0. 0402“ 00402"
(0.0159) .. (0.0159).
Age60 -0. 1196 -0.1 196 ‘
(0.0285) (0.0285)
001 0.4268‘” 0.4269‘”
(0.0352) (0.0351)
N 6,087 6,087
# of groups 31 31
R’ 0.1973 0.1973
Tax Elasticity 4.4416 4.4447

 

The numbers in parentheses are robust standard errors, adjusted for cluster sampling. "'7‘, and ' denote
significance at the 1%, 5%, and 10% levels respectively. # of groups is the number of counties represented

in the sample.

35

1.3.3.3. Probit Analysis

The dependent variable is a binary variable in this model, so I can use Probit
estimation to grasp the characteristics of the dependent variable. In addition, I have to
include the county-dummy variables in the Probit estimation, to find the innate trait of
each county. Table 1.7, Table 1.8, and Table 1.9 Show}9 the results of the Probit
estimation at the county level, for consumers in metropolitan areas of various sizes.

The estimation results are very similar to those in the fixed-effects analysis. Like the
estimation results in the fixed-effects analysis, the sales-tax rate (taxrate) is positive in all
estimation results, and insignificant, marginally significant, and significant in the cases of
the one-million, two-million, and three-million population limits respectively. When the
population limits increase, the marginal effects and t-statistics of taxrate increase
together. All marginal effects of taxdifir and taxratio are negative, and they are
statistically significant in the cases of one- and two-million population limits. Thus, the
sales-tax rate and geographic variables are influential factors in determining consumers’
shopping patterns in the Probit analysis for large metropolitan areas. just as they were in
the fixed-effects analysis. Other variables such as lincome, female, white, married, edu,
nch, d0] etc., have the same signs as in the fixed-effects analysis.

When I change my regression method from the fixed-effects estimation to the Probit
estimation, the signs of the sales-tax rate (taxrate) and the geographic variables (taxdiff
and taxratio) do not change, and their t-statistics do not change very much, either. These

results support the results of the last subsection that taxrate, taxdifl, and taxratio play

 

29 The estimates in Table 1.7, Table 1.8, and Table 1.9 are the marginal effects of explanatory variables.

36

major roles in determining consumers’ purchasing patterns. As suggested by the results in
the last subsection, a consumer would be more likely to make online purchases when she
lives in a jurisdiction with a high sales-tax rate. In addition, even if she lives in a
jurisdiction with a high sales-tax rate, a consumer would be less likely to make online
purchases, as long as her jurisdiction adjoins other jurisdictions with lower sales-tax
rates. These results suggest that residents of large metropolitan areas are reducing their
sales-tax liabilities, both by evading taxes through lntemet purchases, and by engaging in

cross-border shopping.

37

Table 1.7 Probit Estimation (Population Limit=l,OO0,000)

 

Dependent Variable: whether she or he makes online purchases

 

Taxrate 0.0448 0.0479
(0.0529) (0.0499)
Taxdrfl -o.0934”
(0.0404)
Taxratio -145611’“
(4.8225)
Lincome 0.0417'“ 0.0417’"
(0.0095) (0.0095)
Female -0. 0414' -0. 0413.
(0.0221) (0.0221)
White 0 0787 0 0789
(0.0263) (0.0264)
Married -0.0517‘ -0.0516‘
(0.0303) (0.0303)
F emared 0.0061 0.0057
(0.0308) (0.0308)
Whmared -0.0008 -0.0009
(0.0290) (0.0291)
Fewhite 0.0117 0.0116
(0.0280) (0.0280)
Fewhmar 0.0474 0. 0478
(00384)“ (0. 0385)
Highgrad 0.0917 0. 0917
(0.0189)” (0 0189)"
Colgrad 0.1161 0.1161
(0.0111) (0.0111)
Prograd 0.0189 0. 0188
(0.0172)” (0. 0172)
Nch 0. 0987 o 0988
(0. 0072) (0.0072)"
Agel5 -0. 1584 -0.1585
(0.0212) (0.0212)
Age20 -0. 0023 -0.0025
(0 0138)” (0.0138).
Age30 0. 0556 0.0555 ’
(0.0131) (0.0131)
Age50 -0.0481”‘ .0.0481‘”
(0.0119) (0.01 19)
Age60 -0.1265”’ -01264‘"
(00180)” (0. 0180)”
DO] 0.4030 0. 4025
(0.0121) (0.0120)
N 16,013 16,013
Pseudo R’ 0.1564 0.1565

 

The partial effects of the dummy variables 1n Probit are for discrete changes from 0 to I. The numbers in
parentheses are robust standard errors, adjusted for cluster sampling. M, ,and denote significance at the
1%, 5%, and 10% levels respectively.

38

Table 1.8 Probit Estimation (Population Limit=2,000,000)

 

Dependent Variable: whether she or he makes online purchases

 

Taxrate 0.0659 0.0646
(0.0606) (0.0609)
Taxdifl -0.1 199'“
(0.0418)
Taxratio -12. 3473‘”
(4. 2741)
Lincome 0.0340 0. 0340
(0.01 14) (0.0114)
Female -00406‘ -00407‘
(0.0238) (0.0238)
White 00695" 0.0695"
(0.0296) (0.0296)
Married -0.0705” -0.0705”
(0.0323) (0.0323)
Femared 0.0144 0.0144
(0.0357) (0.0357)
Whmared 0.0138 0.0138
(0.0304) (0.0304)
Fewhite 0.0147 0.0147
(0.0310) (0.0310)
Fewhmar 0.0395 0 0395
(0.0447)" (0. 0447)
Highgrad 0.0950 0. 0950
(0.0200)“ (0 0200)
Colgrad 0.1266 0 1266
(0.0122) (0.0122)
Prograd 0.0392” 0.0392“
(0.0183) (0.0183)
Nch 0.1030‘" 0 1030‘“
(0. 0086) (0. 0086)”
Agel5 -0. 1394 -0 1394
(0.0272) (0.0272)
Age20 -0.0001 -0. 0001
(00168)” (0. 0168)
Age30 0.0582 0 0582
(0.0155) (0.0155)
Age50 -0.0526“‘ -0. 0526‘”
(0.0143)" (0. 0143)
Age60 -0.1338 -0. 1338
(0. 0204) (o. 0204)”
001 0. 4081 0.4081
(0.0160) (0.0160)
N 11,341 1 1,341
Pseudo R’ 0.1561 0.1560

 

The partial effects of the dummy variables in Probit are for discrete changes from 0 to I. The numbers in
parentheses are robust standard errors, adjusted for cluster sampling. .. , ,and denote significance at the
1%, 5%, and 10% levels respectively

39

Table 1.9 Probit Estimation (Population Limit=3.()O0,0()O)

 

Dependent Variable: whether she or he makes online purchases

 

Taxrate 0.2971‘ 0.2976
(0.1567) (0.1564)
Taxdifl -0.0843
(0.0779)
Taxratio -8.7982
(8.0032)
Lincome 0.0346" 0.0346“
(0.0148) (0.0148)
Female -0.0449’ -0.0449’
(0.0251) (0.0251)
White 0.0359 0. 0359
(0.0306)” (0. 0306)
Married -0.0949 -0. 0949
(0.0345) (0.0345)
F emared 0.0363 0.0363
(0.0431) (0.0431)
Whmared 0.0350 0.0350
(0.0394) (0.0394)
Fewhite 0.0195 0.0194
(0.0419) (0.0419)
Fewhmar 0.0256 0.0256
(0.0627) (0.0627)
Highgrad 0.0894‘" 0 0894‘"
(00231)” (0.0231)
Colgrad 0.1313 0.1313
(0.0157) (0.0157)
Prograd 0. 0501‘ 0.0501‘
(0 0263) (0.0263).
Nch 0 0974 0.0974 ‘
(0. 0114)" (00114)”
Agel5 -0. 1429 -01429
(0.0381) (0.0381)
Age20 -0 0003 -00003
(0 0207)” (0.0207)"
Age30 0. 0735 0.0735
(0.0199) (0.0199)
Age50 -00502‘" -00502‘”
(0.0188) (0.0188)
Age60 -0.1293'” -0 1293‘"
(00294)” (0. 0294)
001 0.4301 0.4302
(0.0274) (0.0274)
N 6,087 6,087
Fseudok’ 0.1612 0.1612

 

The partial effects of the dummy variables in Probit are for discrete changes from 0 to l. The numbers in
parentheses are robust standard errors, adjusted for cluster sampling. "H, a.nd denote significance at the
1%, 5%, and 10% levels respectively.

40

1.3.3.4. Empirical Analysis with Inflation level

In the regressions reported so far, there is no correction for price level, or for the
inflation rate. Price-level data are available for certain metropolitan areas from ACCRA.
However, these data are normalized within each year. Therefore. they do not allow a
meaningful comparison across years, either for different metropolitan areas or for the
same area. Consequently, it has not been possible to use price-level data. As an
alternative, I include a measure of the rate of change of the price level. I include the
percentage change in the Consumer Price Index (PCCPI) as an explanatory variable in
this subsection. When the price levels in retail stores in a region are increasing very
much, consumers have an incentive to find other shopping methods, such as mail-order
shopping and lntemet shopping. Therefore, I expect that inflation rates would have a
positive effect on lntemet shopping. Thus, if inflation rates are not included, the
estimated coefficients may suffer from omitted-variable problems. In order to compare
the inflation rates in different regions, I use percentage changes in the Consumer Price
Index (pccpi) as a proxy variable.30

In the preceding paragraph, I argued that price-level data might be superior to
inflation data. However, there is a theoretical argument that could be advanced in favor
of the pccpi variable. If we adopt a model with habit formation, it is possible that a
sudden jump in prices could cause consumers to become more sensitive to economic

variables than they had been previously. In a model such as this. the inflation rate might

 

3° pccpi=(cpi,—cpi,_,)x 100/cpi,-,

41

be more appropriate than the price level. In any event, the decision to use pccpi is driven
largely by data availability.

The subsection 1.3.3.1 and 1.3.3.2 show that the sales-tax rate (taxrate) and the
geographic variables (taxdijf and taxratio) have expected signs and significant results
when I restrict the sample and choose only those observations from big metropolitan
areas. The Bureau of Labor Statistics“ reports the Consumer Price Index (CPI) for 26
metropolitan areas annually. There are large metropolitan areas, such as Chicago, Detroit,
Los Angeles, New York, Philadelphia, and San Francisco. Thus, if I include the
percentage change in the Consumer Price Index (pccpi) as an explanatory variable,
automatically 1 choose observations for those in big metropolitan areas. After that, I can
compare the results of this model specification with the results of population limits in the
previous subsections.

I use the fixed-effects estimation at the county level. At first, I do not place a
population-limit restriction on the sample, and then I restrict the sample to metropolitan
areas with population limits such as one, two, and three million. Table 1.10 shows the
estimation results without population limits when I include pccpi as an explanatory
variable. Table 1.11, Table 1.12, and Table 1.13 show the estimation results with pccpi
when I place one-, two-, and three-million population limits on the sample.

Table 1.10 (No Population limit) shows that the coefficient estimate of the sales-tax
rate (taxrate) has a positive sign and a significant t-statistic and that the coefficient
estimates of the geographic variables (taxdijf and taxratio) also have expected signs
(negative signs) and significant t-statistics. The estimation results support the results of

the previous subsections, in which population limits are placed on the sample without

 

3' CPIs for metropolitan areas are available from http://www.bls.gov/cpi/home.htm.

42

pccpi. The coefficient estimate of the percentage change in the Consumer Price Index
(pccpi) has a positive sign, and is statistically significant. It implies that the consumers
who live in the metropolitan areas with high inflation rates would be likely to engage in
online commerce. Other variables have similar results as in the previous subsections.
When I use population limits such as one, two, and three million, Table 1.1 1, Table
1.12, and Table 1.13 show that the sales-tax rate (taxrate) has a positive sign and its
coefficients are statistically significant. The geographic variables (taxdiff and taxratio)
are negative, and their coefficient estimates are significant except for the three million
case. pccpi are positive, and its coefficient estimates are significant except for the three
million case too. All estimation results strengthen the results of the previous subsections,

and support the evidence that the inflation rate is an influential factor in e-commerce.

43

Table 1.10 F ixed-Effects Estimation (No Population Limit & Peep!)

 

Dependent Variable: whether she or he makes online purchases

 

Taxrate 0.2080" 0.1933"
(0.0946) (0.0927)
Taxdifl -00756‘”
(0.0269)
Taxratio -7.8194'"
(2.7479)
Pccpi 0.0436” 0. 0419‘
(00214)" (0. 0219)
Lincome 0.0355 0.0354
(0.0083) (0.0083)
Female -0. 0460’” -0 046 1
(0. 0178) (0. 0178)”
White 0.0713 0.0714
(0.0242) (0.0243)
Married -0.0357 -0.0356
(0.0256) (0.0256)
Femared 0.0106 0.0105
(0.0232) (0.0233)
Whmared 0.004 1 0.004 1
(0.0272) (0.0272)
F ewhite 0.0246 0.0246
(0.0225) (0.0225)
Fewhmar 0.0237 0.0239
(0. 0308) (0.0308).
Highgrad 0. 0749 0.0748 ‘
(0.0165) (0.0165)
Colgrad 0.0961‘” 0.0961
(0.0093) (0.0093)
Prograd 0. 0247‘ o. 0248'
(00132) (0. 0132)
Nch 0. 0822 0 0822
(0. 0065)” (0.0065)
Age15 -o. 1065 -0.1065 ”
(0.0202) (0.0202)
Age20 0.0116 0.0117
(0. 0127) (0. 0127)"
Age30 0.0438 0.0438
(0.0115) (0.0115)
Age50 -0.0417”‘ 0.0416“
(0.0105) (0.0105)
Age60 -0.1147’” -0.1 146‘”
(0. 0168) (0.0168)
001 0 3594 0.3598’“
(0.0241) (0.0243)
N 16,188 16,188
# of groups 106 106
R’ 0.1819 0.1818

 

The numbers in parentheses are robust standard errors, adjusted for cluster sampling. m, ", and ' denote
significance at the 1%, 5%, and 10% levels respectively. # of groups is the number of counties represented

in the sample.

44

Table 1.11 Fixed-Effects Estimation (Population Limit=1,000,000 & PCCDI)

 

Dependent Variable: whether she or he makes online purchases

 

Taxrate 0.2268" 0.21 19"
(0.1017) (0.0999)
Taxdlfi“ -0.0676”
(0.0271)
Taxratio -7.8001"
(3.0693)
Pccpi 0.0462” 0 043 7‘
(00227)” (0. 0234)
Lincome 0.0372 0.0371
(0.0089) (0.0089)
Female -0.0391. -0. 0390.
(0.0201 ). (0 0201)
White 0.0648 0 0650
(0.0256) (0.0257)
Married -0.0418 -0.0417
(0.0279) (0.0279)
F emared 0.0096 0.0094
(0.0278) (0.0279)
Whmared -0.0037 -0.0038
(0.0266) (0.0266)
Fewhite 0.0117 0.0116
(0.0257) (0.0257)
Fewhmar 0.0388 0.0390
(0. 0349) (0.0349).
Highgrad 0. 0708 0.0708 ’
(0.0177) (0.0177)
Colgrad 0.103 1 0.1030‘”
(0.0099) (0.0099)
Prograd 0.0208 0.0209
(0. 0148) (0.0148)
Nch 0. 0853 0.0853‘”
(0.0072) (0.0072)
Agel5 -0.1156‘” 0.1157’"
(0.0222) (0.0222)
Age20 0.0037 0.0037
(0. 0129) (0. 0129)”
Age30 0. 0487 0 0487
(0.0127) (0.0127)
Age50 -0.0450‘” 0.0449”
(0.0116) (0.0116)
Age60 -0. 1 116‘” -0. 1 1 15‘“
(0. 0176) (0. 0176).
001 0. 3636 0. 3648
(0.0263) (0.0265)
N 13,233 13,233
# of groups 84 84
R’ 0.1934 0.1935

 

The numbers in parentheses are robust standard errors, adjusted for cluster sampling. m, and ‘ denote
significance at the 1%, 5%, and 10% levels respectively. # of groups is the number of counties represented
in the sample.

45

Table 1.12 F ixed-Effects Estimation (Population Limit=2,000,000 & Pccpi)

 

Dependent Variable: whether she or he makes online purchases

 

Taxrate 0.2904“ 0.2903"
(0.1165) (0.1164)
Taxdifl 00544“
(0.0268)
Taxratio -5.6301"
(2.7799)
Pccpi 0.0602" 0 0603“
(0.0280)" (0 0280)
Lincome 0.0303 0. 0303
(0.0094) (0.0094)
Female .0. 0395’ -0. 0395‘
(0.0204) (0. 0204).
White 0. 0599 0.0599
(0.0270) (0.0270)
Married -0.0573“ 0.0573”
(0.0282) (0.0282)
Femared 0.0144 0.0144
(0.0306) (0.0306)
Whmared 0.0056 0.0056
(0.0265) (0.0265)
Fewhite 0.0160 0.0160
(0.0265) (0.0265)
Fewhmar 0 0356 0.0356
(0 0381) (0 0381)
Highgrad 0. 0716 0.0716
(0. 0159) (0. 0159)"
Colgrad 0. 1133 0. 1133
(0.0112) (0 0112)
Prograd 0.0307 0.0307
(0 0154) (0 0154)”
Nch 0. 0878 0.0878
(0.0080) (0.0080)
Agel5 -0.1164’” -0.1 164‘“
(0.0244) (0.0244)
Age20 0. 0006 0.0006
(0. 0156) (0. 0157)
Age30 0.0487 0.0487
(0.0138) (0.0138)
Age50 -0.0410”‘ -0.0410”‘
(0.0127) (0.0127)
Age60 -0.1155”‘ .0.1155‘“
(0. 0197) (0.0197).
001 0. 3425 0.3424 ‘
(0.0336) (0.0336)
N 10,785 10,785
# of groups 67 67
R’ 0.1901 0.1901

 

The numbers in parentheses are robust standard errors, adjusted for cluster sampling. m, ", and ' denote
significance at the 1%, 5%, and 10% levels respectively. # of groups is the number of counties represented
in the sample.

46

Table 1.13 Fixed-Effects Estimation (Population Limit=3,000,000 & Pccpr)

 

Dependent Variable: whether she or he makes online purchases

 

Taxrate 0.3481' 0.3481'
(0.1822) (0.1820)
Taxdifi' -0.01 10
(0.0436)
Taxratio -1 . 1539
(4.5269)
Pccpi 0. 0459 0.0459
(0. 0479) (00479)”
Lincome 0. 031 1 0.031 1
(0.0119) (0.0119)
Female -0.0438’ -0.0438‘
(0.0229) (0.0229)
White 0.0292 0.0292
(0.0276) (0.0276)
Married 00738“ -00738“
(0.0310) (0.0310)
Femared 0.0325 0.0325
(0.0383) (0.0383)
Whmared 0.0216 0.0216
(0.0340) (0.0340)
F ewhite 0.0244 0.0244
(0.0376) (0.0376)
Fewhmar 0.0213 0.0213
(0. 0552) (0. 0552)"
Highgrad 0. 0659 0. 0659
(00181) (0.0181)
Colgrad 0. 1141 0.1141‘"
(0.0138) (0.0138)
Prograd 0.0345‘ 0.0345‘
(0.0209) (0.0209)
Nch 0.0773‘" 0.0773‘"
(0.0101) (0.0101)
Agel5 -0.1167“‘ -0.1167”’
(0.0332) (0.0332)
Age20 00014 -00014
(0.019) (0.019)
Age30 0.0644’” 0.0644‘”
(0.0175) (0.0175)
Age50 -0. 0356“ 00356”
(0. 0167)" (00167)”
Age60 -0. 1066 -0.1066
(0. 0275) (00275)”
001 0. 3766 0.3766
(0.0597) (0.0597)
N 5,690 5,690
# ofgroups 29 29
R’ 0.2017 0.2017

 

The numbers in parentheses are robust standard errors, adjusted for cluster sampling. and I denote
significance at the 1%, 5%, and 10% levels respectively. # of groups is the number of counties represented

in the sample.

47

1.4. Conclusion

In the previous section, I analyze empirically the effects of sales-tax rates and
geographic variables on Internet shopping. If I place no restrictions on the sample, and do
not use an inflation variable, the estimates for the sales-tax rate without population limits
give the expected signs, but the estimates are not statistically significant. The estimates
for the geographic variables do not give the expected signs.

If I focus on metropolitan areas with population above certain limits, the estimates
for the sales-tax rate and the geographic variables are of the expected signs, and they are
statistically significant. When I choose the metropolitan areas whose populations are
above one, two, and three million, the sales-tax rate has a positive effect on online
purchases. In addition, an individual would be less likely to buy goods over the lntemet
when she resides in a region adjoining regions with lower sales-tax rates.

Although I get the expected results in the sample of large population metropolitan
areas, I do not get the expected signs of the geographic variables and the statistical
significance of the sales-tax rate without population limits. My interpretation of this
result is that a person who lives in a small town would not have sufficient lntemet facility
and an incentive to cross the border to shop goods at the retail markets of adjoining
jurisdictions. Thus, she would not be likely to use lntemet shopping if her own
jurisdiction has a high sales-tax rate. In addition, she would not be likely to use cross-
border shopping, because there might not exist adjoining jurisdictions whose sales-tax
rates are lower than that of her home jurisdiction. No-population-limit data are not

especially well-suited to the task of estimating the effects of sales-tax differences on

48

cross-border shopping and sales-tax rates on lntemet shopping. On the other hand,
metropolitan-area data with population limits are much better suited to the task, because
the life-styles and physical Internet infrastructures of metropolitan areas are suited to
lntemet shopping, and metropolitan areas may have convenient highway systems that are
necessary to cross-border shopping.

Demographic variables have predictable signs and are statistically significant. I also
observe that lntemet infrastructure, such as the number of computers in the household,
has an influence on online purchases. I also find that the amount of lntemet shopping
increases dramatically during the period studied.

According to my estimates, online commerce is more sensitive to the sales-tax rate if
I increase the population limit. The tax elasticities with a population limit of three million
are above 4. Thus, state or local governments with high sales-tax rates that include
metropolitan areas with large populations would suffer from erosion of the tax base.
Since lntemet shopping is very sensitive to sales-tax rates in big metropolitan areas, it
appears that state and local governments with high sales-tax rates suffer from severe tax
evasion because of lntemet shopping, all else equal.

On the other hand, the geographic variables such as the sales tax-rate differentials
and the sales-tax ratios appear to have negative effects on lntemet shopping. Even if a
consumer lives in a local jurisdiction which has a high sales-tax rate, he would be less
likely to use online commerce when his local county adjoins other counties with lower
sales-tax rates, all else equal.

My interpretation is that a consumer crosses the border and buys goods in

jurisdictions with lower sales-tax rates (i.e., the consumer engages in cross-border

49

sh0pping). Thus, if a county has high sales—tax rates, it appears that the county will suffer
from erosion of its tax base, although the mechanism by which this occurs will be
different in different cases. If the surrounding counties have similar tax rates, then the
local consumers may turn to lntemet shopping. If the surrounding counties have lower
tax rates, then the local consumers will not be so likely to engage in lntemet shopping,

but this may mean that they are merely engaging in cross-border shopping, instead.

50

Chapter 2: Does Sales-Tax Evasion on Remote Sales Reduce

Welfare?

2.1. Introduction

Online shopping and mail-order purchases are developing very rapidly. Consumer
mail-order sales are estimated to be $208,787 million in calendar year 2000 and $216,720
million in calendar year 2001. They are expected to increase to $264.7 billion by 200532.
E-commerce is expected to increase to $164.2 billion by 2005. The growth rate of e-
commerce is substantially larger than that of mail-order sales.

Generally, it is relatively easy to escape sales taxes by engaging in remote sales.33
Tax evasion is one reason for the relatively rapid growth of remote sales. Sales-tax
evasion is a growing issue with e-commerce and mail-order purchases. Sellers have an
obligation to collect sales taxes from consumers. However, state and local governments
cannot require sellers to collect sales tax from consumers if sellers do not have “nexus”
(physical presence) in the jurisdiction where goods are purchased“. The buyer is
supposed to report that purchase and fill out a use-tax form. However, state and local
governments do not have effective methods of enforcing the use tax, so use-tax evasion is
widespread. As a result, use-tax revenues are relatively small, compared to the amount

that legally should be collected.

 

32 National Mail Order Association and Michigan Department of Treasury.

33 “Remote sales” refers to both e-commerce and mail-order purchases.

3’ The Supreme Court issued these rulings in 1992 in the case of Quill v. North Dakota and in 1967 in the
case of National Bella Hess v. Department of Revenue of the State of Illinois.

51

The loss of sales and use-tax revenue from remote sales is probably severe, and
taxation of mail-order purchases and lntemet sales is a serious issue for state and local
governments. However, it is difficult to obtain precise estimates of the revenue loss. It is
especially difficult to measure the total revenue loss for the nation as a whole exactly,
because the sales-tax rates of states governments are different, and some state
governments allow their local governments to impose local sales taxes.

In spite of the difficulties of estimating the revenue losses, some attempts have been
made to estimate them. Table 2.1 and Table 2.2 show Michigan use tax revenue loss from
mail-order shopping and E-commerce, and the sales tax loss from not imposing the sales
tax on food, drugs, and services respectively. The state of Michigan is estimated to lose
$210.1 million of sales and use tax revenues from mail-order purchases and Internet
shopping in fiscal year (FY) 2001.35 Mail-order purchases explain $156.2 million of the
revenue loss, and Internet shopping explains the other $53.9 million. By FY 2005, the
revenue loss from remote sales is forecast to grow to $349 million, and mail-order
purchases and lntemet shopping are expected to account for $188.4 million and $160.3
million, respectively. Since e-commerce is increasing very rapidly, the expected revenue
loss in Michigan is projected to increase.

In order to prohibit discriminatory taxes on the lntemet, Congress passed the lntemet
Tax Freedom Act (ITFA) in 1998. The ITFA prohibits any state and local taxes on
lntemet access and any discriminatory taxes on the lntemet. However, sales and use taxes
can still be imposed on goods sold by remote sales. The ITFA does not allow consumers

to avoid the sales and use tax for remote sales. On the other hand, state and local

 

3’ Michigan Department Treasury, Office of Revenue and Tax Analysis. Tax Revenue Loss Estimates for
Consumer Remote Sales. Lansing, MI, 2001.

52

Table 2.1 Michigan Sales and Use Tax Revenue Loss from Consumer

Remote Sales

 

Fiscal Year
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005

Mail-Order Purchases E-Commerce

$104.4
$112.8
$123.0
$131.7
$138.8
$149.3
$156.2
$163.6
$172.0
$180.2
$188.4

All amounts are million of dollars.
Source: Tax revenue Loss Estimates for Consumer Remote Sales. Lansing. MI. 200 I.

NA
NA
$7.2
$14.0
$25.1
$37.3
$53.9
$76.0
$101.5
$128.9
$160.3

Total Remote Sales
$104.4
$1 12.8
$130.2
$145.7
$163.9
$186.7
$210.1
$239.6
$273.4
$309.1
$348.7

Table 2.2 Michigan Sales and Use Tax Revenues, and Tax Loss Estimates

 

 

Year 2000 2001

Sales and Use Tax Revenues $7,632,900 $7,685,900
Loss of Use Tax from Mail-Order Purchases $149,300 $156,000
Loss of Use Tax from E-Commerce $37,300 $54,000
Sales Tax Loss from Exempting Food $888,000 $900,000
Sales Tax Loss from Exempting Drugs $199,000 $269,000
Sales Tax Loss from Exempting Services $4,260,000 $4,584,000
Per Capita Personal Income $29,516 $29,788

 

All values except for per-capita personal income are thousands of dollars. Per-capita personal income is

measured in dollars.

Source: Michigan Department of Treasury and the Bureau of Economic Analysis.

53

governments do not have reliable methods to enforce the sales and use tax from lntemet
shopping and mail-order purchases.

Some articles in the literature address the welfare effect and the excess burden when
there is commodity-tax avoidance in the form of cross-border shopping. Lovely (1994)
compares a no-tax-avoidance equilibrium with an equilibrium characterized by cross-
border shopping. She finds that, under certain conditions, the consumer’s welfare
decreases when there is cross-border shopping among small jurisdictions, and the
government faces a revenue requirement. The reason she suggests is that resources
wasted by tax avoidance exceed the benefit of consumption changes by cross-border
shopping. She shows that a revenue-neutral tax reform to deter cross-border shopping
could increase the consumer’s welfare.

Trandel (1992) uses a spatial competition model of firms to analyze the welfare
effects of cross-border shopping by consumers. According to his analysis, the existence
of use-tax evasion in cross-border shopping can lead firms to charge lower prices than if

36 economic

use-tax evasion were impossible. As a result, use-tax evasion could raise
welfare even when a higher tax rate is needed, if it reduces the seller’s market power. In
his model, the seller is assumed to have market power, but I do not assume that there is
seller’s market power.

Unlike the case of tax avoidance through cross-border shopping, however, there is

not crurently any literature to assess the welfare change or the excess burden associated

with e-commerce and mail—order shopping.

 

3‘ Trandel’s result is contrary to the general view that tax evasion leads to welfare losses. Yitzhaki (1987)
shows that tax evasion is related to uncertainty, which leads to a utility loss. In Yitzhaki’s model, as a result
of tax evasion, the excess burden of a distortionary tax becomes larger.

54

In this study, I analyze the welfare change that would occur if the sales and use tax
were collected successfully from remote sales. That is, 1 will compare the consumer’s
welfare in two cases; one in which consumers can escape the sales and use tax by
engaging in remote sales, and the other in which state and local governments can impose
the sales and use tax on remote sales successfully, and use the resulting revenue to
decrease sales-tax rates. The first case describes the present reality. The second case
describes an ideal world, in which state and local governments can collect the sales and
use tax from e-commerce and mail-order shopping. The welfare of the tax-evaded
economy will be compared with that of the economy with no tax evasion in this study.

In order to do so, I will derive a benchmark case, an economy with lump-sum
taxation. The excess burden of each sales-tax system is a deadweight loss that would
occur if the tax system were not lump-sum taxation. If we compare their excess burdens,
we can conclude which one is preferable to consumers. I will also compare the tax-
evasion economy with the no-tax-evasion economy directly. I will also calculate the
excess burden of the current sales-tax system. If we assume that the current sales-tax
system is changed marginally in the direction of the new sales-tax system, we could
calculate the excess burden of one dollar of revenue in the current sales-tax system.

In section 2.2, I will show a basic framework for this study, and I will summarize
data in section 2.3. In section 2.4, I will analyze the welfare effect of the new revenue-
neutral tax policy using the Cobb-Douglas and CBS utility function. The last section will

summarize the results.

55

2.2. Theoretical Background

I use partial-equilibrium analysis to weigh the welfare effect of imposing the sales
tax successfully on remote sales. I use differential-incidence analysis37 instead of
balanced-budget incidence analysis. This means that I hold the total sales and use-tax
revenues fixed. When the state government collects the sales tax from remote sales, the
total sales-tax revenues are assumed to be equal to those that were collected when
consumers could evade the sales tax from remote sales. In order to hold tax revenues
constant, it is necessary to adjust the tax rate when the tax base is broadened. Since the
comparison is between the current sales-tax system and a new equal—revenue tax system,
the income effect washes out.

I model the economy in terms of a representative consumer. Therefore, there is no
' distributional issue in this economy. I consider the case in which the representative
consumer has a Cobb-Douglas (C-D) utility function as well as the more general case of a
CBS utility function. The representative consumer is assumed to consume three kinds of
goods: taxed goods, untaxed food and services,38 and untaxed remote sales goods. The
model does not include foreign trade or international capital flows. In this model, there
are no extemalities, and the only distortion is the sales and use tax. This model does not
divide government expenditure into parts such as exhaustive projects and transfer

a ent ro ams.39 It is assumed that ovemment sales-tax revenues are fixed, and that
P ym P gr g

 

37 Rosen, Harvey 8., Public Finance, 5"h edition. Irwin/McGraw-Hill.

3‘ Many state governments do not impose their sales taxes on food and services, and the forgone revenues
from those items are huge. 1n the US. as a whole, services represented 45.4% of GDP in 1980, and 65.9%
in 2000 (Slemrod, 2002).

39 I do not cover the calculations of marginal cost of public funds (MCF) and marginal efficiency cost of
redistribution (MECR) in this study (see Ballard (1988) for a discussion of the MECR, and Ballard (I991)

56

the composition of the sales-tax revenues would be changed after the new-tax policy: the
revenue requirement is exogenous. I use the real sales and use tax revenues of the state of
Michigan.40

The representative consumer purchases goods to maximize his utility function,

subject to his budget constraint. His problem is,

Max U=f(x1 axzaxs)

x,,x2,x3

3
st. W =Zq,x, (2.1)

1-1

where x) represents taxed retail goods, x2 represents untaxed food and services, and x3
represents untaxed remote sales. W is the representative consumer’s income, and q, is the
consumer price for each good. q,- consists of the producer price and the sales-tax rate
(p)+t). In this model, it is assumed that the producer price (p,) is fixed, and the sales-tax
rate (1) is changeable. Therefore, any change of the consumer price must result from a
change in the sale-tax rate. Suppose that the utility function, U, is twice differentiable and

strictly quasi-concave. The Lagrange function for this problem is,

3
L = f(x19x2 9x3 )+ 1(W -2q1xi (2-2)

I=l

 

for a discussion of the MCF), because this study is not concerned with changes in government expenditure

grograms.
http://wwwmichigan.gov/treasury.

57

Differentiating the Lagrange function with respect to x1, x2, and x3, I get the first-

order conditions for this problem.

f1(x1’x2’x3)"MI =0
f2(x1’x2:x3)‘M2 =0
f3(x. .x2.x_~,)—/lq, =0 (2.3)

W—Zajq. ,=0

1-1

where fl is the partial derivative of the utility function with respect to x,-. If I manipulate
the first-order conditions and budget constraint, 1 can derive the ordinary demand
function for each good. When I put the demand function for each good into the
consumer’s utility function, I get the indirect utility function, V(q, W), which depends on
prices and income as well as on the parameters of the utility function.

The theoretical section of the earlier chapter had Kuhn-Tucker conditions, because
the analysis was at the level of individual, and the individual may experience corner
solutions. Here, we assume interior solutions throughout, because the analysis is being
conducted at the level of an aggregated, representative consumer.

In order to compare the tax-evasion economy (the current sales-tax system) with the
no-tax-evasion economy (new equal-revenue sales-tax system), I derive the Excess
Burden that would be calculated if the lump-tax economy with equal revenue were
replaced with each tax system. The excess burden from the current sales-tax system will
be compared with one from the alternative sales-tax system. It can tell us which sales-tax

system is more desirable for consumers.

58

2.2.1. Specification of the Model in the Case of Sales-Tax Evasion and Avoidance.

Because of tax evasion, the consumer price of remote sales, q 3, should be equal to
the producer price, p3. In general, a consumer price is composed of the producer price and
the sales tax, that is, q,-=(I +t)p,-. However, the consumer price of remote sales is equal to
the producer price, p 3, because the representative consumer escapes the sales-tax burden.

In addition, the consumer price for food and services“ is also equal to the producer
price, because state and local governments do not collect sales and use tax from them:
q2=p2. However, the consumer price of general retail goods is different from their
producer price. In other words, the consumer price of general retail goods, q 1, consists of
their producer price times (1+ sales-tax rate), that is, q1=p1(1 +1). I use the existing sales-
tax rate, t, to calculate the consmner price in this setup. Some states have a constant sales-
tax rate across the state, but other states have different sales-tax rates in different parts of
the state, because counties and municipalities can impose their own sales taxes. I use data
for Michigan in my simulations. The state of Michigan has a 6% constant sales-tax rate
across the state.

Let the superscript, e, refer to the sales-tax evasion economy (the current sales-tax
economy). Thus, q" and t" are the consumer price and sales-tax rate vector with the sales-

tax evasion economy, and p is the producer price vector. According to Kay (1980),

 

“ Generally, the sales and use tax is not imposed on services. However, state and local governments
impose the sales tax on some service activities such as telephone, vehicle lease payments, and hotel and
motel charges. Several states impose sales tax on food, but Michigan (like most other states) does not.
Some states that tax food have an income-tax credit for estimated sales-tax payments on food, in an attempt
to offset regressivity.

59

Pazner and Sadka (1980), and Triest (1990), the excess burden in the sales-tax evasion

economy is,

EB((1‘.17,W)=W-E(P,V(q”,W))-T(qerP,W) (2-4)

EB(q‘,p, M0 is the excess burden from the current sales-tax system which is characterized
by the tax evasion, E(p, V(q‘, "0) is the expenditure function to achieve the utility level,
V(q‘, 119, under the lump-sum tax economy, and T(q",p, W) is the sales-tax revenue under
the current sales-tax system, T (q‘,p, M=f Xx)(q‘, m. EB(q",p, ll? is the additional burden
that the representative consumer would bear if the distortionary current sales-tax
economy were substituted for lump-sum taxation, while keeping the consumer welfare at

his post-tax level, v(q", ”0( See Pazner and Sadka, 1980).

2.2.2. Specification of the Model for Uniform Sales-Tax Rate, with no Evasion or

Avoidance.

The consumer price for remote sales is different from the producer price when
governments can impose sales tax successfully on e-commerce and mail-order purchases.
Its consumer price, q 3, is composed of the producer price and the sales-tax rate, p 3(1 +1).
The consumer price for general retail goods, q), is also composed of the producer price
and the sales-tax rate, p)(1+t). However, I assume that state and local governments do not

change their policy toward food and services. In other words, state and local governments

60

do not impose the sales and use tax on food and services in this setup. Thus, the
consumer price for food and services, q 2, is the same as the producer price, p2.

In this section, sales and use tax is collected from remote sales, as well as from sales
of goods that were previously taxed. The new sales-tax rate, which can make the sales-tax
revenues equal to the sales-tax revenues of the previous section, is smaller than the
existing sales-tax rate. The reason is that the tax base is broadened in this section.

Let superscript 11 represent the new economy with no tax evasion. q" and f' are the
new consumer price and tax rate vector corresponding to the new sales-tax. system (in the

no tax-evasion economy). The excess burden of the new equal-revenue economy is,

EB(q",p.W)=W-E(p,V(q",W))-T(q",p.W) (25)

EB (q”, p, M is the additional burden that the representative consumer should bear if
lump-sum taxation were replaced with the new sales-tax system, while keeping the
1 consumer welfare at his post-tax level, v(q", M(Pazner and Sadka, 1980). E60, V(q", W)) is
the expenditure function to achieve the utility level, V(q", W), under the lump-sum tax

economy, and T (q", p, "0 is the sales-tax revenue under the new sales-tax system,

77‘1"»10 m=fthfanm+flXx3(anM-

61

2.2.3. Comparison of the current sales-tax system with the new sales-tax economy

I can compare the current sales-tax system (with sales-tax evasion on remote sales)
with the new sales-tax economy (with no tax evasion), for the purpose of calculating the
excess burden. According to Pazner and Sadka (1980), the comparisons of the excess
burdens can give rise to the correct utility ranking of alternative tax schemes, which yield
the same amount of tax revenue. When EB(q",p,llO is larger than EB(q",p, 80, we can say
that the new sales—tax system is superior to the current tax-evasion economy, and vice
versa. The former case explains the inefficiency of tax evasion on e-commerce and mail-
order shopping, and shows policy implications regarding the imposition of sales tax on
remote sales.

I also investigate the excess burden of the current sales-tax system using the
equivalent variation (EV) and compensating variation (CV)42 if we assume that the
current sales-tax system is replaced with the new equal-revenue tax economy. Excess
burden is derived, to calculate the inefficiency of the current sales-tax system. It is the
differential welfare change per dollar of revenue (Ballard, 1990, and Ballard and

Fullerton, 1992).43

 

‘2 In calculating the change in consumer’s welfare, Ballard, Shoven, and Whalley (1985) use the equivalent
variation (EV). EV and CV are equivalent at the margin (See Mayshar, 1990). Therefore, I will use EV and
CV. Note that Stuart (1984) uses the compensating surplus (CS). See Fullerton (1991) and Stuart (1984).

‘3 Originally, this is used to compare a particular type of distortionary tax system with an equal-revenue
lump-sum tax. This should be used in differential incidence analysis. In this paper, I use it to calculate the
excess burden of the current sales-tax system.

62

Excess Burden=21 W/AT
AW: Change in Consumer Welfare

AT: Amount of Tax Revenue replaced with the equal-revenue tax system (2.6)

For the numerator, I use the equivalent variation (EV) or compensating variation
(CV) when the current sales-tax system is replaced with the new equal-revenue sales-tax
system. When I calculate the denominator, the amount of tax revenue from remote sales

after the tax-policy change will be used.

EV(q‘.V")=W-E(q".V(q".W))
CV(q",V‘)=E(q".V(q".W))-W

(2.7)

AT =t" I"x3(q",W) (2.8)

The excess burden measures the additional welfare costs per dollar of tax revenue of the
current sales-tax system in Michigan due to the sales-tax evasion. Therefore, the excess

burden could be decreased if the state of Michigan selected the new tax policy.

63

2.3. Data

I use data for the state of Michigan in fiscal year (FY) 2000 and 2001.44 1 get the
important values from the Department of Treasury in Michigan and its tax reports.45 For
example, these include the general sales and use tax revenues for the state of Michigan in
FY 2000 and 2001. I also get the estimates of the loss of use-tax revenues from remote
sales in FY 2000 and 2001. In addition, the state of Michigan does not impose the sales
and use tax on food, drugs, and services, so I obtain estimates of these revenue losses

from the Department of Treasury of Michigan and its tax reports.

Table 2.3 Estimates of the Sales Amount of Each Good (thousands of
dollars)

 

 

Year 2000 2001

General Retail Goods $127,215,000 $128,098,333
Mail-Order Purchases $2,488,333 $2,600,000
E-Commerce $621 ,667 $900,000
Food $14,800,000 $15,000,000
Drugs $3,316,667 $4,483,333
Services $71,000,000 $76,400,000

 

These amounts are net of the sales tax.
Source: Michigan Department of Treasury and the Bureau of Economic Analysis.

 

44 The fiscal year of the state of Michigan runs from October 1 to September 30.
‘5 Michigan Department of Treasury, Office of Revenue and Tax Analysis. Michigan Sales and Use Tax.
Lansing, MI, 2001 and Tax Revenue Loss Estimates for Consumer Remote Sales. Lansing, M l, 2001.

64

146 from the Bureau of

I get per-capita personal income for Michigan in 2000 and 200
Economic Analysis. The per-capita personal incomes of Michigan in 2000 and 2001 are
$29,516 and $29,788 respectively. The sales and use tax rate in Michigan has been 6%
since May 1, 1994.47 The data are summarized in Table 2.2 and Table 2.3.

The sales-tax revenue of Michigan in FY 2001 is $6,352.3 million, an increase of
$74.8 million (1.2%) from FY 2000. The use-tax revenue is $1,333.6 million in FY 2001,
a decrease of $21.8 million (-1.6%) from FY 2000.48 The total loss of use-tax revenue
due to tax evasion from remote sales is estimated to be $210 million and $187 million in
FY 2001 and 2000, respectively. The total revenue losses from the sales-tax exemption
for food, drugs, and services are estimated to be $5,753 million and $5,347 million in FY
2001 and 2000 respectively.

I can estimate the sales amount of each good using the above information and the
sales-tax rate, 6%. Table 2.3 shows the sales amount of each good. According to Table
2.3, e-commerce is a very small part of consumer expenditure, but it is increasing rapidly.

The growth rate of e-commerce is 50% during that period, and that of mail-order

purchases is 4.7%.

 

‘6 http://www.bea.doc.gov/bea/regionaI/spi/

‘7 Previously, the sales and use tax rate in Michigan was 4%.

‘8 The use tax is generally paid by business sectors. The telecommunication sector and the automotive
sector are the main sources for the use tax in Michigan. Most payments in the telecommunication sector are
collected from interstate and intrastate telephone calls, and leasing and private sales of motor vehicles are
the main sources in the automotive sector (Michigan Department of Treasury, 2001). Business services,
hotels and motels, and transportation manufacturing are the other sources for use-tax revenue in Michigan.

65

2.4. Computation of Welfare Change

2.4.1. Cobb-Douglas Utility Function

I use a Cobb-Douglas (C-D) utility function in this section, to find a welfare change.
The Cobb-Douglas utility function is quite restrictive, since it implies unitary own-price
elasticities for all goods and unitary elasticities of substitution among goods. However, it
is a useful starting point for the analysis. The consumer’s utility function and budget

constraint are,

Max U = xf'xgzx;'“"“’
x, ,x2,x3
3
s.t. W = Ear, (2.9)

I=I

x) represents taxed retail goods, x2 represents untaxed food. drugs, and services, and x3
represents untaxed remote sale goods. If we solve the representative consumer’s problem

to maximize the Lagrange function, we get the demand function for each good.

W W l- - W
x.=9"——, x2412 .x.=( 0“ “2) (2.10)
‘11 ‘12 ‘13

 

 

In order to derive the indirect utility function, I insert the demand functions (equation

(2.10)) into the ordinary utility function (equation (2.9)).

66

 

V = (M)al (fly!) ((1 _al ‘02 )W)I-a,—a,
‘11 ‘12 (13 (2 11)
a1 a a2 a (l—al '02) l-a _a ,
=(—) '(—) ’(————) . 2W
ql q2 q3

2.4.1.1. Computation of the excess burden when tax evasion for remote sales is

widespread

Originally, the consumer price is only different from the producer price for general
retail sales. All other goods, including remote sale goods, have identical consumer and

producer prices. That is, (2.11) is written as

V SW = i... “I310: (l-al_a2)
(q ) (mam) (12.” p,

)'""'“"’ W (2.11)’
In order to calculate the indirect utility, I begin by inserting per-capita personal
income into W in equation (2.11)’, and I assume that all producer prices are unitary. The
sales-tax rate in the state of Michigan is 6%, so I use 0.06 for t in equation (2.11)’. In
order to find values for a 1 and a2, 1 use the fact that each a is an expenditure share when
we employ the C-D utility function. Table 2.4 shows the 01 values in FY 2000 and 2001.

The excess burden of the current sales-tax economy due to tax evasion in remote

sales, EB(q‘,p,W) = W -E(p,V(qe,W))—T(q",p,W), is $11.72 and $11.97 per

67

Table 2.4 The values of as in C-D Utility Function

 

 

Year 2000 2001
a, .5938 .5774
a; .3924 .4077
1a,. a; .0137 .0149

 

capita in 2000 and 2001 respectively. Thus, the analysis with the Cobb-Douglas utility
function suggests that the amount by which the excess burden could decrease if the
current sales-tax system were replaced with the lump-sum taxation is about $12 per
person. In other words, the excess burden which a consumer bears under the current

sales-tax system is almost $12.

2.4.1.2. Computation of welfare when state government can impose sales tax on

remote sales

Suppose that the state government of Michigan extends the sales tax to remote sales,
but that the state government still does not impose the sales tax on food and services.
This broadening of the tax base would allow the state to collect the same total revenue,
with a lower tax rate.

The consumer price of e-commerce and mail-order purchases changes from the unit
price, 1, to (1+ new sales-tax rate). The consumer price of general retail goods also
changes, because the state government imposes the new sales-tax rate on them. However,
in this experiment, food, drugs, and services are still not the subject of the sales tax, even

when there is a tax-rate change.

68

In order to calculate the new equal-revenue sales-tax rate. it is necessary to consider
the changes in demands that will occur when consumer prices change.49 When the
representative consumer faces the new consumer prices, he changes his quantity
demanded based upon the new consumer prices. In calculating the new equal-revenue
sales-tax rate, I have to use this property of each demand function.

I introduce the new definitions, “Actual Revenue” and “Required Revenue” in

calculating the new equal-revenue sales-tax rate.

Actual Revenue = t"x,° + t"x;

q . (2.12)
Required Revenue = Nominal Base Tax Revenue(+"’“—)

q old

0 . t t
t" rs the new sales-tax rate under no tax evasron, and x) and x 3 are the demands based
upon the constrained maximization. Nominal Base Tax Revenue is the original sales-tax

revenue, collected from the representative consumer under the current sales-tax system.
21",, is the new composite price index, based on the new sales tax rate, and q d is the

. . . o . S
orrgrnal composrte prrce 1ndex, based on the current sales-tax rate.~0

 

’9 Slemrod (2002) uses a simple method to calculate the new sales-tax rate, t, that solves the equation,
t°(current sales tax revenues+use tax losses from remote sales)=0.06'current sales tax revenues. However,
this does not consider the demand changes that would be caused by the new sales-tax rates.

5° This composite price index is derived from the property of the Cobb-Douglas utility function (Ballard et
al., 1985).

69

- 1+1" 0 1 a 1+t” _.,,_a
q,....=(——-) '(—-) ’(——--—-——)I ' ’

al a2 l—al —a2 (2.13)
— 1.06 a l a 1
61.1.1 =(—) '(-—) ’(

a a —

1-a.—a2
1 2 1“ a1 2 )

I find the new equal-revenue sales-tax rate that makes Actual Revenue equal to Required
Revenue. In order to do so, I guess the first new sales-tax rate, insert it into equations
(2.12) and (2.13), and compare Actual Revenue with Required Revenue. If they are not
equal, I change the sales tax rate. I continue with this iterative procedure until I get the
sales-tax rate that makes Actual Revenue equal to Required Revenue. According to this

method, the equal-revenue sales-tax rates are 5.8559% in FY 2000 and 5.8403% in 2001.
The excess burden, EB(q" , p,W) = W — E(p,V(q" ,W)) — T(q" , p,W) , if lump-sum
taxation were replaced with the new sales-tax system, would be $11.05 and $11.25 in

2000 and 2001 respectively. In other words, the excess burden could decrease, if the new

equal-revenue sales-tax system were replaced with lump-sum taxation, by about $11.
2.4.1.3. Computation of the welfare change in GB utility function

In order to compare the tax-evasion economy with the no tax-evasion case, their
excess burdens are an important yardstick used to evaluate the deviation from the optimal
economy. As I show in the previous sections, the excess burden of the current sales-tax
system is bigger than that of the new sale-tax system. In particular, the per-capita excess
burden of the tax-evasion economy is bigger by about 70 cents than that of the no-tax-

evasion economy. This means that the tax-evasion economy is less efficient than the no-

70

tax-evasion economy, and that the no-tax-evasion economy is superior to the tax-evasion
economy.

I attain the conclusion that the new sales-tax system is better than the current sales-
tax system according to the efficiency measure. I investigate how inefficient the current
sales-tax system is using the excess burden, equation (2.6). To measure the excess
burden, let us assume that the current sales-tax system is replaced with the new sales-tax
system. I am able to get the exact values of E V(q", V"), C V(q", V"), and AT corresponding
to the tax-policy change, and calculate its marginal excess burden.

Table 2.5 shows the result of the welfare analysis in the OD utility function.
E V(q‘, V") and C V(q", V) per capita in FY 2000 are almost $07123, which is very small.
E V(q‘, V") and C V(q", V) per capita in FY 2001 are about $07742, and this is also very
small. Even if I interpret EV and CV as a percentage change of income, they are very
small numbers. EV and CV in FY 2000 are approximately 0.0024% of the representative
income, and they are almost 0.0026% of his income in FY 2001. The excess burden per
dollar is roughly .0317 in 2000 and .0316 in 2001.

Thus, the excess burden which is derived from the sales-tax evasion is about 3 cents
per dollar of revenue. Therefore, the representative consumer would prefer the new
equal-revenue sales-tax system to the crurent sales-tax system. These results are
consistent with the traditional optimal taxation theory, that the uniform taxationSI is

optimal under some conditions.52

 

5' In the current setting, in which it is assumed that the sales tax is the only distortion, second-best
considerations are minimal. Consequently, it is not surprising that this movement toward the optimal tax
system is associated with a welfare improvement.

5 The utility function should be homothetic, and the utilities of consumption goods should be separable
from leisure. In other words, labor, negative leisure, should be inelastically supplied with respect to all
consumer goods prices. In this model, 1 do not cover leisure and labor explicitly, so the model satisfies the
conditions for optimality of uniform taxes.

71

State and local governments could lower the sales-tax rate by broadening the sales-
tax base, to deter sales-tax evasion. By doing so, state and local governments could
increase the welfare of their residents. In this model, I estimate the equal-revenue sales-
tax rate to be about 5.8%, which is a little bit lower than the existing sales-tax rate, 6%.
Since e-commerce is increasing rapidly and mail-order purchasing is also increasing very
stably, the equal-revenue sales-tax rate might be much lower than the existing sales-tax
rate in future years. Thus, the excess burden from sales and use tax evasion could grow

over the time.

Table 2.5 Computation of Welfare Change in C-D Utility Function

 

 

Year 2000 2001
EV .71228 .7742]
E V/W .00002 .00003

E V/AT 0.03168345 0.031647993
CV .71226 .77419
C V/W .00002 .00003
CV/AT 0.031682686 0.031647171

 

EV is the equivalent variation and CV is the compensating variation. W is per-capita personal income in
Michigan. AT is the amount of sales-tax revenues from the new tax base (remote sales) after the policy
change. E V/AT and C V/AT are the excess burden per dollar of revenue.

72

2.4.2. CES Utility Function

Next, I use the constant elasticity of substitution (CES) utility function to calculate
the welfare change due to the policy change. I include general retail goods and remote
sale goods (mail-order purchased goods and online purchased goods) in the CES utility
function. I remove food and services in this setup, because I would need the elasticity of
substitution between e-commerce goods, and food and services, to specify a CBS utility
function that includes all three categories of goods. However, there is no empirical work
that finds their elasticity of substitution. Therefore, I consider only taxed general retail
goods and remote-sale goods in this section.

The representative consumer’s problem is,53

l a—l 1 0-1 0
Max. U=(,B"x," +(l—fl)"x2" )""
(2.14)
s1. W =q,xI +q2x2

where x) is taxed general retail goods and x2 is remote—sale goods. q, and q) are the
consumer prices of x) and x2, and W is the income of the representative consumer. (I is the
elasticity of substitution between x) and x2, and ,8 is a weighting parameter. Constrained

maximization leads to the first-order conditions,

 

’3 1 follow Ballard (1988 and 1990) in his CES utility function.

73

(—a—)(fl5x.°';+(1-fl)"xi" )5-'<5‘—'1—)/35x{° —Aq. =0
0—1 0'
0—1 1 0-1 1 l I

1- __ . _ , _ _
t—‘i—xmx." +0-0er: >a-'("—‘->tl—/3>ax.a 4% =0 (2.15)
0-1 0'

W‘qrxr "qzxz :0

where 3. is Lagrange multiplier in the constrained maximization. If I manipulate the first

two equations of (2.14), I get the relationship between x) and x3,

£2 =(flty M (2.16)
x, ‘12 ,3

If I manipulate equation (2.14), I obtain the demand functions for x, and x 2,

 

 

x _ W
I- 1_fl ql a

1+ "”7

q (M 0 X612)

(2.17)

x _ W
2— “E- 1.2 a

42+Q1(1_fl)(ql)

In order to calculate the real values of the indirect utility function and expenditure
function, I have to specify the elasticity of substitution, 0, and the weighting parameter, ,6.
The technique of calibration is employed here, to obtain the values of the elasticity of
substitution and the weighting parameter. Differentiating the demand function for x; with

respect to q ), yields,

74

_ ___§_ 22. a

6x2 x2(0' 1)(1_fl)(q‘)
53:: fl q2 a (218)

(INLEXZII) +42

 

Using equations (2.18) and (2.16) leads to,

eta—1x95!)
2 = x2 (2.18)’

3— x
‘“ q.(—‘)+qi
x2

 

In this model, a consumer price, qr, consists of a producer price, p,, and the sales-tax rate,
t (q,-=p,-. (1+t)). I assume that each producer price is unity, and that only a change in the
sales-tax rate can lead to a change in a consumer price. Thus, the change in consumer
price is equal to the change in the sales-tax rate (dq,=dt). If 1 use this relationship
between a consumer price and the sales-tax rate, I get the tax elasticity of remote sale

goods, 8,

x
1 t —1 '
£___§_xl(1+i)=3,_fl(1+i):(“(CT )(x,)

(2.19)54
at x2 aq1 x2 q1

 

x
q1( ‘1')‘i'q2
x

2

 

5‘ Goolsbee (2000) selects (l+sales tax rate) as an explanatory variable in his model. Thus, his tax
elasticity is

75

Goolsbee (2000) obtains tax elasticities of online purchased goods that are 2.3, 3.4,
and 4.3.55 However, the tax elasticity in his study is calculated from the binary variable,
namely, whether the consumer would be likely to make online purchases when the sales-
tax rate increases by .01. This is not how much she or he increases online purchases when
a sales-tax rate increases by .01, which is the original idea of the tax elasticity.

In order to calibrate some parameters and calculate the excess burden correctly, it is
most appropriate to use the compensated elasticity that corresponds to the original tax
elasticity, because this study uses the differential incidence approach. However, 1 cannot
obtain the data about the amount of online goods purchased by the consumer, and there is
no empirical work that finds the original tax elasticity. Therefore, I use the tax
elasticities of Goolsbee’s work in this model.

When I manipulate equation (2.19), I get equation (2.19)’ and then insert real values
into equation (2.19)’. Afier the calibration, the elasticities of substitution between general
retail goods and remote sales goods, 0, are about 3.4, 4.5, and 5.4.56 These elasticities of
substitution are larger than that of the C-D utility function (=1). When sales-tax rates

change, a consumer with this CES utility function would make much larger changes to

 

6x2 (1 +1)
6(1 + t) x2
equation (2.19).

 

, and the change in (1+0 is equal to the change in t itself. His tax elasticity is equal to

5’ Goolsbee (2000) uses three groups of his sample in his analysis. To begin with, he uses all individuals
and gets the tax elasticity, 2.3. Second, he selects the residents who live in states with constant sales-tax
rates. With this sample, the calculated tax elasticity is 4.3. The last group that he analyzes is the individuals
who live in metropolitan areas. He finds that the tax elasticity is 3.4 when this sample is used.

’6 The exact values of the elasticities of substitution and weighting parameters are shown in Table 2.6,
Table 2.7, and Table 2.8.

76

his consumption combination between general retail goods and remote sales goods than if

the consumer’s behavior were governed by a C-D utility function.

EMA-EH 11
0' = 1+ 2 (2.19),
(1+t)(~’“)
x2

 

I calculate the value for the weighting parameter, ,6, using equation (2.16) and the
elasticities of substitution which are derived before. The weighting parameters, B, are
about .98 in each case.

In order to calculate the welfare change, I have to use these values to derive the
excess burden of the current sales-tax system. The indirect utility function is derived if I
use each demand function for x) and x2 and the utility function. The indirect utility

function is,

Vera) = mam, +qztlixflrfi +(1-fl)”[qz +q.<—/—’-)<12—)“1 ; 107 (2.20)
:6 qz l‘rB ‘11

2.4.2.1. Computation of the excess burden when tax evasion for remote sales is

widespread

Legally, the residents in state of Michigan have to pay the use tax when they
purchase goods over online or by mail-order. However, a consumer can evade paying the

use tax very easily in remote sales, so the consumer price of remote sale goods is the

77

same as its producer price. The consumer price of general retail goods is different from
its producer price.

In order to calculate the excess burden of a representative consumer, EB(qe,p, W), I
use 1.06 and 1 for q 1 and q2, and per-capita personal income in Michigan for W. For ,8 and
o, I use the values which I obtain in the calibration process. The per-capita excess burden
of the current economy (with sales-tax evasion), when compared with the case of lump-
sum taxation, is $3.42, $4.48, and $5.31 in 2000, and $3.84, $5.02, and $5.96 in 2001,

with the different values corresponding to the different values of the tax elasticity.

2.4.2.2 Computation of the excess burden when state government can impose sales

tax on remote sales

Suppose that the state government of Michigan can collect the sales and use tax
successfully from remote sales, and use the revenues to decrease its sales-tax rate. As in
the case of the C-D utility function, this analysis is a differential-incidence analysis.
Therefore, the total sales-tax revenues are constant, even when government policies
change.

In order to obtain the equal-revenue sales-tax rate, I compare the Required Revenue
with the Actual Revenue, and find the sales-tax rate that makes them equal. This sales-tax
rate will guarantee equal sales-tax revenues after the policy change. Equation (2.12) gives

the idea for the Required Revenue and Actual Revenue. However, I have to use different

78

composite price indices in the Required Revenue from those in the C-D utility function.

The old and new composite price indices in the CES utility function are57

1
E}... = (011;; +(1-fl)ql;ir)':f (2.21)
a... = (£an +(1—B)q;:.:’..)"-°

where qlord and qgold are 1.06 and 1 respectively, and qlm. and qgnew are the new
consumer prices for X] and xz. When I follow the same iteration procedure as in the CD
utility function, I obtain the equal-revenue sales-tax rate in the CES utility function too.
The new sales-tax rates that make the sales-tax revenues constant are shown in Table 2.6,
Table 2.7, and Table 2.8. They are in the range of 5.84% through 5.86%.

When I use the new consumer prices, I am able to obtain the excess burden of the
new sales-tax system (with no tax evasion), EB(q",p, W), which is caused by non-lump-
sum taxation. The excess burdens of the new sales-tax system are zero in all cases. The
new sales-tax system is uniform taxation and is optimal from the point of view of
efficiency. In my model, there is no labor-leisure choice, and the CES utility function is
homothetic. Therefore, uniform taxation is optimal. My analysis to measure the excess
burden of the new sales-tax system demonstrates the traditional optimal taxation theory

numerically.

 

’7 The composite price index is based upon the CES utility function (Ballard, I990).

79

Table 2.6 Computation of Welfare Change Using the CES Utility Function
(Tax Elasticity=2.3)

 

 

2000 2001
Elasticity of Substitution (o) 3.353158658 3.359285234
Weighing Parameter (fl) 0.980247004 0.978028166
New Sales Tax Rate (t) .05855712220587 0.0583952853071
EV 3.845931996 4.313453974
E V/W 0.000130300 0.000144805
EV/AT 0.119246515 0.119450405
CV 3.845430937 4.312829454
CV/W 0.000130283 0.000144784
CV/AT 0.119230980 0.119433110

 

EV is the Equivalent Variation and CV is the Compensating Variation. W is per-capita personal income in
Michigan. AT is an amount of sales-tax revenues from the new tax base (remote sales) after the policy
change. E V/A T and C V/A T are the excess burden.

Table 2.7 Computation of Welfare Change Using the CES Utility Function
(Tax Elasticity=3.4)

 

 

2000 2001
Elasticity of Substitution (o) 4.478582363 4.487639041
Weighing Parameter (,8) 0.981477568 0.979397498
New Sales Tax Rate (t) 0.05855462817169 0.05839251893086
EV 5.033726425 5.647012309
E V/W 0.000170542 0.0001 895 73
E V/A T 0.166450899 0.166781154
CV 5.032868108 5.645941989
CV/W 0.000170513 0.000189537
CV/AT 0.166422517 0.166749543

 

EV is the Equivalent Variation and CV is the Compensating Variation. W is per-capita personal income in
Michigan. AT is an amount of sales-tax revenues from the new tax base (remote sales) after the policy

change. E V/A T and C V/A T are the excess burden.

80

Table 2.8 Computation of Welfare Change Using the CES Utility Function
(Tax Elasticity=4.3)

 

 

2000 2001
Elasticity of Substitution (a) 5.3993 83577 5.410837611
Weighing Parameter (B) 0.982428179 0.980455414
New Sales Tax Rate (t) 0.05855266297583 0.05839033902004
EV 5.969729613 6.697960976
E V/W 0.000202254 0.000224854
E V/AT 0.208087602 0.208535339
CV 5.968522456 6.696455249
C V/W 0.000202213 0.000224804
CV/A T 0208045524 0208488459

 

EV is the Equivalent Variation and CV is the Compensating Variation. W is per-capita personal income in
Michigan. AT is an amount of sales-tax revenues from the new tax base (remote sales) after the policy
change. EV/ATand CV/ATare the excess burden.

2.4.2.3 Computation of the welfare change in CES utility function

I compare the tax-evasion economy (the current sales-tax system) with the no tax-
evasion economy (the new equal-revenue sales-tax system) using their excess burdens.
From the earlier sections, I attain the important conclusion that the new sales-tax system
which is characterized by uniform taxation is as efficient as a system with lump-sum
taxation. Therefore, the new sales-tax economy is better than the current sales-tax
economy, and the consumer prefers the no tax-evasion economy to the tax-evasion
economy. The excess burden if the current sales-tax system is replaced with the new
sales-tax system can be calculated. The E V(qe, V") and C V(q", V") can be calculated if we
assume that the current sales-tax system is changed into the new sales-tax system. AT (=
x2(q", M't”), is the new sales-tax revenue from the remote sales.

The EV and CV of the representative consumer in FY 2000 are $3.85, $5.03, and

$5.97, and his EV and CV in FY 2001 are $4.31, $5.65, and $6.70, corresponding to each

81

tax elasticity. The values of EV and CV in the CES utility function are bigger than those
in the OD utility function. EV and CV are increasing with the tax elasticity and year
when we look at Table 2.6, Table 2.7, and Table 2.8.

If I interpret EV and CV as a percentage change of income, they are small numbers.
EV and CV in FY 2000 and FY 2001 are about 0.01% through 0.02% of the
representative consumer’s income. A consumer is willing to spend 0.01% or 0.02% of her
income for the new tax policy. When I use the excess burden per dollar of revenue as the
measure of welfare change, they are 0.12, 0.17, and 0.21, depending on the tax elasticity.
After the tax-policy change, the excess burden per-dollar revenue generated by sales-tax

evasion decreases by 12 cents through 2 1 cents.

Table 2.9 Computation of Welfare Change Using the CES Utility Function

(Population Limit=2,000,000)

 

 

2000 2001
Elasticity of Substitution ( o ) 15.59081 141 15.62879938
Weighting Parameter ( B ) 0.990219868 0.989128809
New Sales Tax Rate (t) 0.058555879 0.058397028
EV 13.90196433 15.1153939
E V/W 0.000470998 0.000507432
E V/AT 0.870597142 0.845977606
CV 13.89541962 15.84444598
C V/W 0.000470776 0.000531907
CV/AT 0.870187286 0.886781156

 

EV is the Equivalent Variation and CV is the Compensating Variation. W is per-capita personal income in
Michigan. AT is an amount of sales-tax revenues from the new tax base (remote sales) after the policy

change. E V/A T and CV/A T are the excess burden.

82

It is appropriate to conduct sensitivity analysis with respect to the elasticity of
substitution. The results presented above are based on the elasticities that are derived
from Goolsbee’s estimates (2000). To give a sense of the welfare effects associated with
various values of the elasticity of substitution, I use an elasticity derived from my earlier
chapter (two-million population limit). The simulation results are shown in Table 2.9.
When I use my results (two-million case), the elasticity of substitution is very large, and
EV and CV are relatively large compared to those in Goolsbee’s case. Thus, the excess
burden in my case is also large such as 85 cents through 89 cents.

In the case of the OD utility function, the elasticity of substitution between taxed
general retail goods and untaxed remote-sale goods is 1. The elasticity of substitution of
the CES utility function is bigger than that of the OD utility function. This means that
the representative consumer would be more likely to substitute general retail goods for
remote-sales goods in the case of the CES utility function if the sales tax were imposed
on remote-sale goods. After the tax-policy change, the proportion of the remote sales
would be smaller in the consumer’s consumption bundle, and the sales-tax revenues from
remote sales would be smaller, if the CES utility function is the correct description of the
consumer’s preferences, rather than the C-D utility function. Therefore, the excess burden
associated with the CES utility function is relatively large compared to that associated
with the OD utility function.

I can conclude from my analysis that the representative consumer prefers the new
equal-revenue sales-tax system to the current sales-tax system, and that the representative

consumer in the CES utility function would respond more sensitively to the tax-policy

83

change than one in the C-D utility function would. Thus, the representative consumer
with a CES utility function would be better off than one with a C -D utility function, if the

sales tax were imposed on remote-sales goods.

84

2.5. Conclusion

In the previous sections, 1 estimated the welfare of the representative consumer in
two cases: one represents the present case, in which sales-tax evasion on remote sales is
widespread. The other is the policy scenario in which state and local governments can
succeed in collecting the sales tax from remote sales.

The welfare of the representative consumer would increase as a result of the policy
change. That is, if state and local governments collected the sales tax successfirlly from e-
commerce and mail-order shopping and decreased the sales-tax rate, the consumer would
prefer the no-tax-evasion economy to the tax-evasion economy, because the excess
burden of the current sales-tax system is larger than that of the new equal-revenue sales-
tax system. In addition, since the elasticity of substitution of the CES utility function is
larger than that of the C-D utility function, the excess burden of the current sales tax
system would be larger with the CES utility function than with the C-D utility function.

The motto of public finance is to broaden the tax base and to reduce the tax rate.58
When state and local governments can collect sales tax successfully from online and
mail-order purchases, they could reduce their sales-tax rates. When commodity-tax rates
are the same across all goods, uniform taxation would make the economy optimal under
some conditions (Sandmo, 1976 and 1974, and Atkinson and Stiglitz, 1972). In this
model, I do not consider the labor-leisure choice in the problem of the representative
consumer. The consumption Choice of the representative consumer in my model is

invariant with respect to the labor-leisure choice. This property corresponds to the

 

’3 Slemrod (2002) called it the first folk theorem.

85

conditions in which uniform taxation is optimal according to Sandmo’s model (1974).
That is, “if ...the utility function is weakly separable between labor and consumption
goods and homogenous in consumption goods, the proportional taxation is optimal.”

We have two goods (general retail goods and remote sales goods) in the CES utility
function. After the tax-policy change, uniform taxation would be imposed, and we would
have the uniform sales tax for all goods. As a result, the welfare of the representative
consumer would increase. In the case of the C-D utility function, there are three goods:
general retail goods, food and services, and remote sales goods. After the tax-policy
change, general retail goods and remote sales goods would be taxed at the same rate, but
food and services would not be taxed. Therefore in this model, it is not the case that all
goods are taxed at the same rate, even after the policy change.59 However, since the tax
distortions are the only distortions in this model, any movement in the direction of
uniformity would improve the welfare of the representative consumer, even if we don’t
get all the way to uniformity.

If state and local governments have ways to enforce sales and use tax from remote
sales and decrease the sales and use tax rate, the economic efficiency would be improved.
However, state and local governments do not have reliable ways to enforce the sale and
use tax for remote sales. If I focus on this problem differently, the resource costs devoted
to enforcement could be very large. Specifically, the probability of detecting tax evasion
may be so low that state and local governments cannot deter sales-tax evasion by those

who are engaged in online shopping and mail-order shopping. If I were to include the

 

’9 I investigate the excess burdens of the current sales-tax system in two-good case, and its excess burdens
are about more than 3 cents. Like the CES utility function in two-good case, the uniformity would improve
the welfare of the representative consumer under the C-D utility function in two-good case. Its excess
burdens are smaller than those of the CES utility function.

86

resource costs to detect the tax evasion to my model, I could get a different result about
welfare changes."0

The welfare simulations reported here are based on a model in which the sales tax is
the only distortion. It would be possible to construct a model in which there is a labor-
leisure choice, and in which income taxes and payroll taxes distort the labor-supply
decision. In such a model, the sales tax would be only a small part of the overall wedge
between the marginal revenue product of labor and the marginal utility of consumption.
This has the potential to have a substantial effect on the welfare estimates. Building such

a model is a promising avenue for future research.

 

6° In a model of cross-border shopping, Lovely (1994) compares a no-enforcement equilibrium with costly
evasion to an enforcement equilibrium with costly evasion, and proves that enforcement reduces welfare if
administrative and uncertainty costs outweigh the benefit of consumption changes.

87

CHAPTER 3: CONCLUSION AND POLICY

IMPLICATION

We find that the sales-tax rate is an important factor in online shopping according to
Goolsbee (2000). In my model, I show that the geographic characteristics of consumers
as well as the sales-tax rate are essential factors in determining the purchasing patterns of
consumers. The sales-tax rate and geographic variables are not significant and do not give
interesting implications when we use the whole sample. However, the empirical results
with population limits tell us that the sales-tax rate is an influential factor in electronic
purchases, and the geographic variables such as the sales-tax ratios and the sales-tax rate
differentials affect the purchasing patterns of consumers in large metropolitan consumers.

These empirical results tell us that a consumer considers the sales-tax rate of his own
county in making online purchases. The empirical results can lead us to the speculation
that consumers also consider the sales-tax rates of their neighboring counties and
compare the relative prices of online shopping and cross-border shopping. Another
important conclusion is that consumers living in regions with high inflation rates would
be more likely to make online purchases. In order to evade the increasing price levels of
his or her jurisdiction, he or she could use online shopping.

In the second chapter, I estimate the welfare change that could be derived from the
tax-policy change; under which state and local governments can successfully collect the
sales and use tax from remote sales. A consumer’s welfare can increase in the cases of
both the CD and CES utility functions if the new tax policy is adapted. The excess

burden that is induced by the current sales-tax evasion is about 3 cents per dollar if we

88

assume the OD utility function. In the case of the CES utility function, the excess
burdens are substantially larger than those of the OD utility function.

The existence of tax evasion in remote sales distorts the decision-making of
consumers and causes the additional excess burden of the sales tax. Therefore, the ideal
sales-tax system is to collect sales tax from online shopping and mail-order shopping.
However, it is very hard to determine which state government owns the sales-tax
revenues from remote sales, if state and local governments can collect the sales tax from
remote sales. It is also difficult to decide which state or local governments make online
stores to collect the sales tax from consumers. According to the destination base, the state
government of consumers (online shoppers) should have the authority to impose the sales
tax. Therefore, online stores may collect the sales tax from consumers at the rate of the

consumers’ home jurisdiction, and pay it to the state or local government of consumer.

89

APPENDIX

Table A.1 Weighted Average Sales Tax Rates

 

 

 

 

 

 

State County I 997 I 997
Alabama 4 4
Calhoun 5.8769 5.8769
Jefferson 7.2091 7.2091
Madison 8.0009 8.0009
Tuscaloosa 5.3177 5.3177
Alaska 0 0
Anchorag: 0 0
Arizona 5 5
Maricopa 6.9733 6.9733
Pirna 6.2647 6.2647
Pinal 7.0333 7.0333
Yavapai 6.6825 6.6825
Yuma 7.1727 7.1727
California 6 6
Alameda 8.25 8.25
Butte 7.25 7.25
Contra Costa 8.25 8.25
El Dorado 7.25 7.25
Kern 7.25 7.25
Los Angeles 8.25 8.25
Marin 7.25 7.25
Merced 7.25 7.25
Monterey 7.25 7.25
Orange 7.75 7.75
Placer 7.25 7 .25
Sacramento 7.75 7.75
San Diego 7.75 7.75
San Francisco 8.5 8.5
San Joaquin 7.75 7.75
San Luis Obispo 7.25 7.25
San Mateo 8.25 8.25
Santa Barbara 7.75 7.75
Santa Clara 8.25 8.25
Sonoma 7.5 7.5
Stanislaus 7.375 7.375
Tulare 7.75 7.75
Ventura 7.25 7.25
Yolo 7.25 7.25
Colorado 3 3
Arapahoe 6.2716 6.2716
Boulder 7.2155 7.2155
Denver 7.3 7.3
ElPaso 5.6218 5.5151
Jefferson 6.4877 6.4399
Larimer 5.4156 5.8947
Pueblo 6.7223 6.4265
Weld 5.2355 5.2431

 

90

Table A.1 (cont’d)

 

State

County

1997

N
O
O

 

Delaware

Kent
New Castle
Sussex

 

District of Columbia
Florida

Alachua
Bay
Brevard
Broward
Charlotte
Clay
Collier
Dade
Hernando
Hillsborough
Lake

Lee
Manatee
Marion
Okaloosa
Orange
Osceola
Palm Beach
Pasco
Pinellas
Polk
Sarasota
Seminole

O\ f"
GQNONQ-GGQCOOO
LI!

05
{11

05
'Jr

LII

o‘g‘qqqooqaagoooo
MM Lit

6.7

 

Georgia

Hawaii

Illinois

Indiana

Clayton
Cobb
Dekalb
Fuhon
Gwinnett

Hawaii

Lasalle
Macon

Hamilton
Lake
Laporte
Porter
St.Joseph

91

6.25
6.25
7.1458

MMMMM

ka‘xlflMO\-bQNON\IO\O\QO\O\O\O\O\\J

6.25
6.5
7.1421

MMMMM

Table A.1 (cont'd)

 

 

 

 

 

 

 

 

 

 

State County 1 997 200 1
Iowa 5 5
Black Hawk 6 7
Linn 5.0014 5.0045
Scott 6 7
Kansas 4.9 4.9
Shawnee 5.6436 6.5330
Kentucky 6 6
Kenton 6 6
Louisiana 4 4
E.Baton Rouge 8 9.016705
Calcasieu 8.6489 8.6523
Jefferson 8.75 8.75
Ouachita 8.2645 8.5230
Maine 6 5
Kennebec 6 5
Maryland 5 5
Baltimore 5 5
Carroll 5 5
Frederick 5 5
Harford 5 5
Howard 5 5
Montgomery 5 5
Prince George's 5 5
Washinflrn 5 5
Michigan 6 6
Berrien 6 6
Genesee 6 6
Jackson 6 6
Macomb 6 6
Monroe 6 6
Washtenaw 6 6
Minnesota 6.5 6.5
Anoka 6.5 6.5
Dakota 6.5 6.5
Hennepin 6.6687 6.6714
Ramsey 6.7659 6.7810
St.Louis 6.9408 6.9591
Washington 6.5 6.5
Missouri 4.225 4.225
Jefferson 6.0348 6.0843
St.Louis 6.3262 6.5656

 

92

Table A.1 (cont'd)

 

 

 

 

 

 

 

State County 1997 2001
Nebraska 5 5
Lancaster 6.3599 6.3519
Nevada 6.5 6.5
Clark 7 7 .25
Washoe 7 7.25
New Jersey 6 6
Bergen 6 6
Burlington 5.9137 5.9154
Camden 5.4989 5.5290
Cumberland 3.8458 3.8307
Essex 4.1927 4.2662
Hudson 3.681 1 3.7225
Hunterdon 6 6
Mercer 5.2273 5.2696
Middlesex 5.7409 5.7280
Monmouth 5.7683 5.7647
Morris 6 6
Ocean 5.7757 5.7882
Passaic 4.6947 4.6683
Somerset 6 6
Union 5.0553 5.0332
New Mexico 5 5
Dona Ana 6.1 107 6.1069
New York 4 4
Bronx 8.25 8.25
Chautauqua 7 7
Dutchess 7.25 7.25
Kings 8.25 8.25
' Monroe 8 8
Nassau 8.5 8.5
New York 8.25 8.25
Orange 7.25 7.25
Oswego 7 7
Queens 8.25 8.25
Richmond 8.25 8.25
St.Lawrence 7 7
Suffolk 8.25 8.5
Ulster 7.75 7.75
Westchester 7.3494 7.3543
North Carolina 4 4
Cumberland 6 6
Forsythe 6 6
Mecklenburg 6 6.5
New Hanover 6 6
Pitt 6 6
Robeson 6 6
Wake 6 6

 

93

Table A.1 (cont’d)

 

State
North Dakota

County

Cass

5

1997

.8624

 

Ohio

Oklahoma

Oregon

Pennsylvania

Clerrnont
Columbiana
Cuyahoga
Hamilton
Lake

Lorain
Medina

Tulsa

Jackson
Lane

Allegheny
Beaver
Berks

Bucks
Butler
Chester
Delaware
Erie

Fayette
Lancaster
Montgomery
Philadelphia
Washington
Westrnoreland
York

\lQOL/t

05

5.75
5.75
5.5
4 5

7.7433

 

S.Carolina

S.Dakota

Tennessee

Horry
Lexington
Richland
York
Minnehaha

Montgomery

94

LII

\1
O5ghmmmmmGOGQQOQ@QOGOO\O\\IO\OOO

\O
00

.°°
L11

L11

'oo
awhammmmo~o~o~qo~o~o~o~ao~o~o~o~o~<o~ooo

J}.
N

.°°
Ln

Table A. 1 (cont‘d)

 

 

 

 

 

 

State County I 997 200 1
Texas 6.25 6.25
Brazoria 7.5980 7.6450
Cameron 7.7986 7.7295
El Paso 8.1173 8.0555
Fort Bend 7.6306 7.4083
Galveston 7.8235 7.8134
Hidalgo 7.6305 7.5826
Lubbock 7.7490 7.7491
Midland 7.5833 7.7756
Tarrant 7.9484 7.9998
Webb 7.9571 8.1686
Utah 4.75 4.75
Utah 6.0040 6.2335
Virginia 3.5 3.5
Chesterfield 4.5 4.5
Fairfax 4.5 4.5
Henrico 4.5 4.5
Prince William 4.5 4.5
Alexandria City 4.5 4.5
Hampton City 4.5 4.5
Newport News City 4.5 4.5
Norfolk City 4.5 45'
Virginia Beach City 4.5 4.5
Washington 6.5 6.5
Clark 7.6017 7.7126
Pierce 8.2544 8.3552
Spokane 7.9500 7.9539
Thurston 8 8
Whatcom 7.6615 7 .7665
Wisconsin 5 5
Brown 5 5.5
Dane 5.5 5.5
Racine 5.1 5. I

 

95

Table A.2 Populations of MSAs

 

 

MSA 1997 2001
Los Angeles, CA PMSA 9,126,131 9,519,338
New York, NY PMSA 8,650,056 9,314,235
Philadelphia, PA-NJ PMSA 4,942,723 5,100,931
Washington, DC-MD-VA-WV PMSA 4,601,800 4,923,153
Detroit, MI PMSA 4,441,551 4,467,658
Houston, TX PMSA 3,841,934 4,177,646
Atlanta, GA MSA 3,632,206 4,1 12,238
Phoenix-Mesa, AZ MSA 2,841,395 3,251,876
Minneapolis-St. Paul, MN-WI MSA 2,795,024 2,968,806
Orange County, CA PMSA 2,760,948 2,846,289
San Diego, CA MSA 2,722,060 2,813,833
Nassau-Suffolk, NY PMSA 2,659,447 2,753,913
St. Louis, MO-IL MSA 2,557,804 2,603,607
Baltimore, MD PMSA 2,474,658 2,552,994
Tampa-St. Petersburg-Clearwater, FL MSA 2,225,002 2,395,997
Oakland, CA PMSA 2,271,696 2,392,557
Pittsburgh, PA MSA 2,358,695 2,359,388
Miami, FL PMSA 2. 132,1 12 2,253,362
Cleveland-Lorain-Elyria, OH PMSA 2,229,246 2,250,871
Denver, CO PMSA 1,901,102 2,109,282
Newark, NJ PMSA 1,942,813 2,032,989
Portland-Vancouver, OR-WA PMSA 1,790,675 1,918,009
San Francisco, CA MSA 1,671,200 1,731,183
Fort Worth-Arlington, TX PMSA 1,554,162 1,702,625
San Jose, CA PMSA 1,621,660 1,682,585
Cincinnati, OH-KY-TN PMSA 1,606,655 1,646,395
Orlando, FL MSA 1,464,182 1,644,561
Sacramento, CA PMSA 1,527,825 1,628,197
Fort Lauderdale, FL PMSA 1,476,860 1,623,018
Indianapolis, IN MSA 1,504,939 1,607,486
Norfolk-Virginia Beach-Newport News, VA-NC MSA 1,549,527 1,569,541
Las Vegas, NV-AZ MSA 1,260,307 1,563,282
Charlotte-Gastonia-Rock Hill, NC-SC MSA 1,350,679 1,499,293
Bergen-Passaic, NJ PMSA 1.333.504 1,373,167

 

The source of this table is the US. Census Bureau (hwgsm 19311506.. 3L1) ). The populations of MSAs in
1997 are populations estimates and the numbers in 2001 are extracted from the United States Census 2000.

MSA and PMSA mean Metropolitan Statistical Area and Primary Metropolitan Statistical Area.

96

Table A.2 (cont’d)

 

 

MSA 1997 2001
New Orleans, LA MSA 1,307,184 1,337,726
Greensboro-Winston Salem-High Point, NC MSA 1,153,179 1,251,509
Raleigh-Durham-Chapel Hill, NC MSA 1,050,697 1,187,941
Middlesex-Somerset-Hunterdon, NJ PMSA 1,104,576 1,169,641
West Palm Beach-Boca Raton, FL MSA 1,014,725 1,131,184
Monmouth-Ocean, NJ PMSA 1,079,186 1,126,217
Jacksonville, FL MSA 1,029,416 1,100,491
Rochester, NY MSA 1,083,427 1,098,201

 

97

BIBLIOGRAPHY

Atkinson, A, B., and J .E.Stig1itz (April, 1972), “The Structure of Indirect Taxation and
Economic Efficiency.” Journal of Public Economics v1, n1: 97-119.

Ballard, Charles, L., and Don Fullerton (Summer. 1992), “Distortionary Taxes and the
Provision of Public Goods.” Journal of Economic Perspectives V6, n3: 117-131.

Ballard, Charles, L. (1991), “Marginal Efficiency Cost Calculations for Government
Exhaustive Expenditure and for Redistribution.” Public F inance-F inances Publiques
v46, n2: 177-185.

Ballard, Charles, L. (March, 1990), “ Marginal Welfare Cost Calculations: Differential
Analysis Vs. Balanced Budget Analysis.” Journal of Public Economics v41, n2: 263-
276.

Ballard, Charles, L. (December, 1988), “The Marginal Efficiency Cost of
Redistribution.” American Economic Review v78, n5: 1019-1033.

Ballard, Charles, L., Don Fullerton, John B. Shoven, and John Whalley, A General
Equilibrium Model for Tax Policy Evaluation. A National Bureau of Economic

Research. The University of Chicago Press, 1985.

 

Bruce, Donald, and William F. Fox (December, 2000), “E-Commerce in the Context of
Declining State Sales Tax Bases.” National Tax Journal v53, n4: 1373-1388.

The Bureau of Economic Analysis, http://www.bea.doc.gov/bea/regional/spi/.

Christiansen, Vidar (1994), “Cross-Border Shopping and the Optimum Commodity Tax
In a Competitive and a Monopoly Market.” Scandinavian Journal of Economics v96,
n3: 329-341.

Comia, Gary C., Kelly D. Edmiston, Steven M. Sheffrin, Terri A. Sexton, David L.
Sjoquist, and C. Kurt Zorn (December, 2000), “An Analysis of the Feasibility of
Implementing a Single Rate Sales Tax.” National Tax Journal v5 3, n4: 1327-1350.

Ferris, J. Stephen (December, 2000), “The Determinants of Cross Border Shopping:
Implications for Tax Revenues and Institutional Change.” National Tax Journal v53,
n4: 801-824.

Fullerton, Don (March, 1991), “Reconciling Recent Estimates of the Marginal Welfare
Cost of Taxation.” American Economic Review v81, n1: 302-308.

98

Fitz Gerald, John D (1989) “ The Distortionary Effects of Taxes on Trade in Border
Areas: the Republic of Ireland-UK Border.” In G. Winckler (ed.), Tax
Harmonization and Financial Liberalization in Europe. St. Martin’s Press.

Fox, William F ., and Matthew N. Murray (September, 1997), “The Sales Tax and
Electronic Commerce: So What’s New?” National Tax Journal v50, n3: 573-592.

Goolsbee, Austan (May, 2000), “In a World without Borders: The Impact of Taxes on
lntemet Commerce.” Quarterly Journal of Economics v115, n2: 561-576.

Goolsbee, Austan, and Jonathan Zittrain (September, 1999), “ Evaluating the Costs and
Benefits of Taxing lntemet Commerce.” National Tax Journal v52, n3: 413-428.

Gordon, Roger H., and Soren Bo Neilsen (November, 1997), “Tax Evasion in an Open
Economy: Value-Added vs.Income Taxation.” Journal of Public Economics v66, n2:
173-197.

Hellerstein, Walter (September, 1997), “Transaction Taxes and Electomic Commerce:
Designing State Taxes That Work in an Interstate Environment.” National Tax
Journal v50, n3: 593-606.

Kay, J.A. (February, 1980), “The Deadweight Loss from a Tax System.” Journal of
Public Economics v13, n1: 111-119.

Lockwood, Ben (September, 1993), “Commodity Tax Competition under Destination and
Origin Principles.” Journal of Public Economics v52, n2: 141-162.

Lovely, Mary E. (January, 1995), “Optimal Commodity Taxation with Costly
Noncompliance.” Public Finance Quarterly v23, n1: 115-130.

Lovely, Mary E. (February, 1994), “Crossing the Border: Does Commodity Tax Evasion
Reduce Welfare and Can Enforcement Improve It?” Canadian Journal of Economics
v27, n1: 157-174.

Mayshar, Joram (December, 1990), “On Measures of Excess Burden and Their
Application.” Journal of Public Economics v43, n3: 263-289.

McLure, Charles E. (December, 1997), “Electronic Commerce, State Sales Taxation, and
Intergovernmental Fiscal Relations.” National Tax Journal v50, n4: 731-749.

Michigan Department of Treasury, Office of Revenue and Tax Analysis. Michigan Sales
and Use Tax. Lansing, MI, 2001.

Michigan Department Treasury, Office of Revenue and Tax Analysis. Tax Revenue Loss
Estimates for Consumer Remote Sales. Lansing, MI, 2001.

99

Michigan Department of Treasury, http://www.michigangov/treasury.

Mikesell, John L. (December, 2000), “Remote Vendors and America Sales and Use
Taxation: The Balance between Fixing the Problem and Fixing the Tax.” National
Tax Journal v53, n4: 1273-1285.

Pazner, Elisha A., and Efraim Sadka (1980), “Excess Burden and Economics Surplus as
Consistent Welfare Indicators.” Public F inance-F inances Publiques v35, n3: 439-
449.

Poterba, James M. (June, 1996), “Retail Price Reactions to Changes in State and Local
Sales Taxes.” National Tax Journal v49, n2: 165-176.

Rosen, Harvey 3., Public Finance, 5th edition. Irwin/McGraw-Hill.

Sandmo, Agnar (1976), “Optimal Taxation-An Introduction to the Literature.” Journal of
Public Economics v6, nl-2: 37-54.

Sandmo, Agnar (September, 1974), “A Note on the Structure of Optimal Taxation.”
American Economic Review v64, n4: 701-706.

Scharf, Kimberly A. (February, 1999), “Scale Economics in Cross-Border Shopping and
Commodity Taxation.” International Tax and Public Finance v6, n1: 89-99.

Slemrod, Joel (September, 2002), “Michigan’s Sales and Use Taxes: Portrait and
Analysis.” mimeo., University of Michigan.

Stuart, Charles (June, 1984), “Welfare Costs per Dollar of Additional Tax Revenue in the
United States.” American Economic Review v74, n3: 352-362.

Trandel, Gregory A. (December, 1992), “Evading the Use Tax on Cross-Border Sales-
Pricing and Welfare Effects.” Journal of Public Economics, v49, n3: 313-331.

Tran-Nam, Binh, Chris Evans, and Michael Walpole (June, 2000), “Tax Compliance
Costs: Research Methodology and Empirical Evidence from Australia.” National
Tax Journal v53, n2: 229-252.

Triest, Robert K. (June, 1990), “The Relationship Between the Marginal Cost of Public
Funds and Marginal Excess Burden.” American Economic Review v80, n3: 557-566.

US. Census Bureau, United States Department of Commerce, http://www.census.gov/

Wedeland, Mike (February 8, 2003), Net Shoppers’ Tax-Free Days Nearly Gone. T_h_e
Detroit News and Free Press.

100

Weiner, Joann M. (September, 1997), “ Discussion of Papers on Telecommunication
Taxation.” National Tax Journal v50, n3: 623—630.

Williams, Seema, David Cooperstein, David Weisman and Thalika Oum
(September,l999) “Post-Web Retail.” TechStrategy Report. The Forrester.

Wooldridge, Jeffrey M. (November, 2001), Econometric Analysis of Cross Section and
Panel Data. MIT Press.

Yitzhaki, Shlomo (April, 1987), “On the Excess Burden of Tax Evasion.” Public Finance
Quarterly v15, n2: 123-137.

101

IWilli"?”lllilllilf

    

Ill