REGIONAL COMPETITION IN PROPERTY TAXES, SCHOOL SPENDING,
AND TAX ABATEMENTS
By
Sung Hoon Kang

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
Agricultural, Food and Resource Economics – Doctor of Philosophy
2013

ABSTRACT
REGIONAL COMPETITION IN PROPERTY TAXES, SCHOOL SPENDING,
AND TAX ABATEMENTS
By
Sung Hoon Kang
In this dissertation, I examine the effects of changes in property tax rates, school
spending, and tax abatements on residential, commercial, and industrial property value growth in
Southeast Michigan using data for all 152 communities in the five counties surrounding Detroit
over the years 1983 through 2002. This is a period during which state government mandated
major changes to school finance. The major contribution of this research is to provide new
insights about the differential effects of policy changes on property classes. A key challenge in
this work is to properly address the potential endogenous relationship between policy changes
and property value growth. In this regard, my strategy is to consider the time period immediately
before and after the imposition of Proposal A, which resulted in significant differential changes
to both property tax rates and school spending for all communities in the state. In addition, I use
spatial econometric techniques to account for potential fiscal spillover effects of competitor
policy changes on one’s own property value growth. In this analysis, I find that: 1) residential
property values are more responsive to school spending changes than property tax rate changes;
2) commercial and industrial property values are more responsive to tax rate changes than school
spending changes; 3) commercial and industrial property values are more responsive to changes
in tax rates than are residential properties; and 4) regional competition plays an important role in
property value growth in the Southeast Michigan region.
I also examine the degree to which the use of industrial property tax abatements spurs
property value growth in the same region surrounding Detroit. My findings show that: 1)

localities that offer tax abatements yield statistically significant positive impacts on industrial
property value growth; 2) the impacts are larger in high tax than in low tax communities; 3) there
are positive spillover effects of tax abatements on residential and commercial property value
growth; 4) the fiscal benefits to local government of tax abatements are much smaller than the
cost of offering tax abatements; 5) this conclusion is remains intact even when I consider
spillover benefits of the tax abatements; and 6) tax abatements offered in competitor
communities do not appear to influence own industrial or other property value growth.

To my wife and son,
Shinhye Choi and Joshua Jiho Kang,
and my parents,
Sangwon Kang and Sunja Kim

iv

ACKNOWLEDGEMENTS

First and foremost, I must acknowledge and thank God for His guidance in my life. I
would like to gratefully and sincerely thank Dr. Mark Skidmore, my advisor and the chair of my
dissertation committee, for his guidance, understanding, mentoring, patience, support, and most
importantly, his friendship during my graduate studies at Michigan State University. His
mentorship was paramount in growing me as an economist as well as an independent thinker. I
also would like to thank my other committee members, Dr. Scott Loveridge, Dr. Robert Myers,
and Dr. Laura Reese for their guidance and support throughout my dissertation work. My
dissertation would not have been possible without their contributions. Their thoughtful
comments and suggestions have greatly improved my dissertation.
I would like to thank all my friends in our department who have been very helpful and
supportive all the way. I am especially grateful to Timothy Hodge, Christina Plerhoples, Ilona
Sipos, and Alexandra Peralta. Without their friendship and support, I could not imagine my
graduate student life.
Finally, I would like to deeply thank my wife, Shinhye Choi, from the bottom of my heart.
I am here because of Shinhye. Her encouragement, patience, support and endless love have
undeniably been the basis of my life. I am also very grateful to my parents, Sangwon Kang and
Sunja Kim, for their unbelievable support and faith in me.

v

TABLE OF CONTENTS

LIST OF TABLES....................................................................................................................... viii
LIST OF FIGURES ....................................................................................................................... xi
CHAPTER 1 Introduction............................................................................................................... 1
CHAPTER 2 Property Tax and Tax Abatement: A Review of Theory and Empirical Studies ..... 4
2.1 Property Taxation......................................................................................................... 4
2.1.1 Tax Competition - Fiscal Externality............................................................. 6
2.2 Tax Abatement............................................................................................................. 8
2.2.1 Firm Location............................................................................................... 11
CHAPTER 3 Property Tax and the IFT Abatement Program in Michigan.................................. 15
3.1 A History of the Property Tax in Michigan ............................................................... 15
3.2 Michigan’s IFT Abatement Program ......................................................................... 17
CHAPTER 4 The Effects of Changes in Property Tax Rates, School Spending, and Tax
Abatements on Residential, Commercial and Industrial Property Value growth.. 24
4.1 Introduction................................................................................................................. 24
4.2 Residential, Commercial and Industrial Property....................................................... 28
4.3 Data and Descriptive Statistics ................................................................................... 30
4.4 Empirical Model ......................................................................................................... 39
Procedure 4.1 ........................................................................................................ 41
4.4.1 Endogeneity ................................................................................................. 42
4.4.1.1 Property Taxes and School Spending ........................................... 42
4.4.1.2 Tax Abatements ............................................................................ 43
4.4.2 Regional Competition .................................................................................. 45
4.5 Empirical Results ........................................................................................................ 50
4.5.1 Residential Property..................................................................................... 52
4.5.1.1 Property Taxes and School Spending ........................................... 52
4.5.1.2 Tax Abatements ............................................................................ 53
4.5.2 Commercial Property ................................................................................... 58
4.5.2.1 Property Taxes and School Spending ........................................... 58
vi

4.5.2.2 Tax Abatements ............................................................................ 59
4.5.3 Industrial Property ....................................................................................... 63
4.5.3.1 Property Taxes and School Spending ........................................... 63
4.5.3.2 Tax Abatements ............................................................................ 63
4.5.3.3 IFT Benefit/Cost Analysis ............................................................ 72
4.5.3.4 Spillover Effects of Tax Abatements............................................ 79
4.6 Robust Evaluation....................................................................................................... 81
4.7 Conculsion and Policy Implications ......................................................................... 103
CHAPTER 5 Conclusion ............................................................................................................ 107
APPENDICES ............................................................................................................................ 109
APPENDIX 4.A Variable Sources ................................................................................ 110
APPENDIX 4.B First Stage IV Results......................................................................... 111
APPENDIX 4.C Determining the Weighted Average of "Competitor" Characteristics Migration Competitors....................................................................... 117
APPENDIX 4.D Instrument for School Spending: Changes in the Total Funding for
Public Schools .................................................................................... 119
APPENDIX 4.E Estimated Time Trend Slopes............................................................. 125
REFERENCES ........................................................................................................................... 130

vii

LIST OF TABLES

Table 3.1 The Percent of Communities Never Giving A Tax Abatement.................................... 23
Table 4.1 Summary Statistics and Variable Definitions............................................................... 32
Table 4.2 Summary Statistics Pre- and Post-Proposal A for Key Variables ................................ 34
Table 4.3 First Difference Results I-1
Dependnet Variable: ∆ Ln Resdiential Property Values Per Capita............................. 55
Table 4.4 First Difference Results I-2
Dependnet Variable: ∆ Ln Resdiential Property Values Per Capita............................. 56
Table 4.5 First Difference Results II-1
Dependnet Variable: ∆ Ln Commercial Property Values Per Capita ........................... 60
Table 4.6 First Difference Results II-2
Dependnet Variable: ∆ Ln Commercial Property Values Per Capita ........................... 61
Table 4.7 First Difference Results III-1
Dependnet Variable: ∆ Ln Industrial Property Values Per Capita ............................... 66
Table 4.8 First Difference Results III-2
Dependnet Variable: ∆ Ln Industrial Property Values Per Capita ............................... 67
Table 4.9 First Difference Results for High and Low Tax Communities
Dependnet Variable: ∆ Ln Industrial Property Values Per Capita ............................... 70
Table 4.10 Estimated Annual Forgone Revenues Resulting from Tax Abatements:
Fiscal Year 1984-2003................................................................................................ 75
Table 4.11 Estimated Generated Revenues Resulting from Tax Abatements:
Industrial Property ...................................................................................................... 76
Table 4.12 Estimated Generated Revenues Resulting from Tax Abatements:
Residential Property.................................................................................................... 77

viii

Table 4.13 Estimated Generated Revenues Resulting from Tax Abatements:
Commercial Property .................................................................................................. 78
Table 4.14 B/C Ratios of Tax Abatements I................................................................................. 79
Table 4.15 B/C Ratios of Tax Abatements II ............................................................................... 80
Table 4.16 First Difference Results - Robust Check I-1
Dependnet Variable: ∆ Ln Resdiential Property Values Per Capita........................... 82
Table 4.17 First Difference Results - Robust Check I-2
Dependnet Variable: ∆ Ln Resdiential Property Values Per Capita........................... 83
Table 4.18 First Difference Results - Robust Check II-1
Dependnet Variable: ∆ Ln Commercial Property Values Per Capita ......................... 85
Table 4.19 First Difference Results - Robust Check II-2
Dependnet Variable: ∆ Ln Commercial Property Values Per Capita ......................... 86
Table 4.20 First Difference Results - Robust Check III-1
Dependnet Variable: ∆ Ln Industrial Property Values Per Capita ............................. 88
Table 4.21 First Difference Results - Robust Check III-2
Dependnet Variable: ∆ Ln Industrial Property Values Per Capita ............................. 89
Table 4.22 First Difference Results - Robust Check IV-1
Dependnet Variable: ∆ Ln Resdiential Propery Values.............................................. 94
Table 4.23 First Difference Results - Robust Check IV-2
Dependnet Variable: ∆ Ln Residential Propery Values.............................................. 95
Table 4.24 First Difference Results - Robust Check V-1
Dependnet Variable: ∆ Ln Commercial Propery Values ............................................ 97
Table 4.25 First Difference Results - Robust Check V-2
Dependnet Variable: ∆ Ln Commercial Propery Values ............................................ 98

ix

Table 4.26 First Difference Results - Robust Check VI-1
Dependnet Variable: ∆ Ln Industrial Propery Values .............................................. 100
Table 4.27 First Difference Results - Robust Check VI-2
Dependnet Variable: ∆ Ln Industrial Propery Values .............................................. 101
Table 4.A.1 Variable Sources ..................................................................................................... 110
Table 4.B.1 First Stage IV Results from Column 2 in Table 4.3................................................ 111
Table 4.B.2 First Stage IV Results from Column 2 in Tables 4.5 and 4.7 ................................. 112
Table 4.B.3 First Stage IV Results from Column 2 in Table 4.16.............................................. 113
Table 4.B.4 First Stage IV Results from Column 2 in Tables 4.18 and 4.20 ............................. 114
Table 4.B.5 First Stage IV Results from Column 2 in Table 4.22.............................................. 115
Table 4.B.6 First Stage IV Results from Column 2 in Tables 4.24 and4.26 .............................. 116
Table 4.D.1 The Total Revenue Per Pupil as Funding for Public Schools: FY1992-1993 ........ 122
Table 4.D.2 The Total Revenue Per Pupil as Funding for Public Schools: FY1994-1995 ........ 123
Table 4.D.3 Changes in the Total Revenue Per Pupil for Public Schools
as the Key Instrument of School Spending: FY1994-1995 ................................... 124
Table 4.E.1 The Number of Granted Tax Abatements (G_IFT) and
of Granted Competitor Tax Abatements (CG_IFT)................................................. 125

x

LIST OF FIGURES

Figure 2.1 Fiscal Benefits from Industrial Property ..................................................................... 14
Figure 3.1 Number of Tax Abatements Granted .......................................................................... 21
Figure 3.2 Total Number of Tax Abatements Granted ................................................................. 22
Figure 4.1 Total Number of Tax Abatements Granted: 1983-2002.............................................. 37
Figure 4.2 Total Number of Tax Abatements Granted Per 1,000 Population: 1983-2002........... 38

xi

CHAPTER 1
Introduction

In 1994, Michigan voters approved Proposal A, which resulted in major changes in
education finance. This reform reduced property tax rates for K-12 education and shifted funding
sources to state government revenues from sales taxes, cigarette taxes, and a new statewide
property value-based six mill state education tax. Also, under this reform, tax rate reductions and
school spending increases were most dramatic in communities with the lowest tax bases, whereas
the changes were much more modest for wealthier communities. Proposal A was designed to
reduce the gap in school spending between poor and wealthy communities while reducing
reliance on local property taxes to fund K-12 education. As I discuss in detail later, the
exogenous changes in tax rates and school spending, resulting from the imposition of Proposal A,
play an important role to identify causal relationships between policy changes and property value
growth. In addition, I consider tax competition because if policy changes in competitor
communities affect one’s own property value growth, then a failure to account for competitor
activities results in biased estimates and potentially misleading inference.
My primary interest in this dissertation is to measure the effects of property taxes, school
spending, and tax abatements on residential, commercial, and industrial property value growth,
using data from a panel of 152 communities in the five counties (Macomb, Monroe, Oakland,
Washtenaw, and Wayne County) surrounding Detroit, Michigan between 1983 and 2002. To
accomplish this, it is important to address two empirical challenges. First, I must appropriately
address endogeneity of property taxes and school spending because simultaneity may lead to bias.
To address endogeneity, I follow Skidmore, et al. (2012) and use exogenous changes in property

1

tax rates and school spending imposed by Proposal A. The statewide imposition of Proposal A
allows me to use an instrumental variable approach while at the same time take into
consideration dynamic and spatial elements to explore relationships between the policy changes
and property value growth. In addition, the granting of industrial tax abatements is also
potentially endogenously determined. To explore the possibility of the endogeniety of tax
abatements, I use as an instrument a variable that measures the city mayor and township
supervisor election cycles. However, I find no evidence of endogeneity and thus treat tax
abatement variables as exogenous. To account for community specific time varying unobserved
fixed factors that could affect policy changes and property value growth, I include communityspecific time trends in the empirical specifications.
Given that the region I examine contains many communities that compete with each other,
it is important to consider appropriate ways to define and control for competitor/neighbor
communities in my evaluation. In traditional studies, contiguity and population are often used to
define neighbors or competitors. However, as noted by Case, et al. (1993), geographic proximity
may not reflect economic closeness. Brett and Pinkse (2000) define population neighbors to
characterize economic closeness. However, in the context of regional competition, actual
competitor communities may differ from neighbors as defined by contiguity or population in that
potential movers care about the tax-service packages offered in potential new communities, as
described by Tiebout (1956). That is, competitor communities can potentially be revealed by the
migration patterns of movers within the region (Skidmore, et al., 2012). Following Skidmore, et
al. (2012), I use a measure of “competitor communities” based on regional migration flow data.
As discussed in detail later, I consider three types of competitors (distance, population, and

2

migration competitors) where each type of competitor is defined using distance, population, and
migration flow information, respectively.
Using the regional competition framework, I find the following: 1) Residential property
values are more responsive to school spending changes than property tax rate changes; 2)
commercial and industrial property values are more responsive to tax rate changes than school
spending changes; 3) commercial and industrial property values are more sensitive to changes in
tax rates than are residential properties; 4) tax abatements have statistically significant effects on
industrial property value growth and their effects are larger in the high than the low tax
communities; 5) there are positive spillover effects of industrial tax abatements on residential
and commercial property value growth; 6) the local government fiscal benefits of tax abatements
are quite small as compared with the costs of offering tax abatements even when I include the
spillover benefits to residential and commercial properties; and 7) regional competition plays an
important role for property value growth; that is, changes in property tax rates/school spending
relative to competitor communities within the same region have significant impacts on one’s
own property value growth.
The remaining portions of this dissertation are organized as follows. The next chapter
provides a brief literature review of the most relevant research. In chapter 3, I provide context by
discussing the most relevant portions of the somewhat convoluted Michigan property tax history,
highlighting major changes in property tax policy. In addition, Industrial Facilities Tax (IFT)
abatement program in Michigan is described in detail in this chapter. Chapter 4 provides
estimates of the effects of property tax rates, school spending, and tax abatements on the value
growth of different types of property: Residential, commercial, and industrial property. Finally,
chapter 5 concludes with a summary of my main findings and policy implications.

3

CHAPTER 2
Property Tax and Tax Abatement: A Review of Theory and Empirical Studies

2.1 Property Taxation
For decades, the property tax and the optimal provision of public goods have been the
subjects of ongoing discussions and debate which can be summarized with the following
1

question: Is the property tax distortionary in nature or is it primarily a benefit tax? Hamilton
(1975) argued that property taxation is an efficient tax if new residents pay property tax rates
equal to the marginal costs for local public services they receive. If zoning requirements, which
2

set minimum house values in a community, are strictly binding, then no one has an incentive to
own homes with higher than the minimum required value because the owner would then be
required to pay higher property tax payments. Therefore, in equilibrium, all residents in a
homogenous community pay the same property tax, serving as an efficient “head tax”, and
receive the same level of public services. In this case, the property tax is efficient. Hamilton
(1976) further argued that under certain conditions the property tax serves as a pricing
mechanism for local public services even when property values are different across taxing
jurisdictions. Specifically, property taxes serve as a pricing mechanism when variations in
property values exactly equal the fiscal differences emerging from different tax/service packages
(perfect capitalization). This view is supported by Fischel (1987, 1995), Yinger, et al. (1988),
Palmon and Smith (1998), and others.

1

For a comprehensive review, see Zodrow (2000).

2

Each (homogenous) community in the region precludes movers who want to receive public
services at relatively low costs, which are less than their share of the costs through fiscal zoning.
4

On the other hand, Mieszkowski (1972) and Zodrow and Mieszkowski (1983, 1986)
suggest that the property tax distorts the allocation of capital: If one community levies higher
taxes on mobile factors such as capital, then the tax base moves to other communities with a
more attractive tax environment, thereby resulting in inefficiently low tax rates (and local public
services). In this case, the property tax is considered a tax on capital and results in “fiscal
externalities” (Wildasin, 1986, 1989). As I will discuss in detail in the next section, in this socalled capital tax view, tax competition is potentially harmful, in contrast to the positive view
presented by Tiebout (1956). In summary, local governments are more likely to set tax rates at
inefficiently low levels to gain locational advantages, thereby resulting in a lower level of
provision of local public services (Oates, 1972; Hoyt, 1991a; Krelove, 1993).
Fiscal zoning and capitalization play crucial roles in the degree to which the provision of
local public services is optimally provided. The body of research suggests that the property tax is
more like a distortionary capital tax when: 1) fiscal zoning is not strictly binding; and/or 2)
perfect capitalization does not occur (Mieszkowski and Zodrow, 1989). Because both
homeowners and capital are mobile, differences in property taxes and the quantity/quality of
public services can be reflected in property values. Further, the property tax is likely to distort
the allocation of people and capital across communities because the assumptions of binding
zoning and/or perfect capitalization are often rejected (Mieszkowski and Zodrow, 1989;
Wildasin, 1989). In the context of the present study, our regional dynamic spatial analysis offers
new evidence of fiscal externalities, and indirect evidence that the property tax is a capital tax, at
least in the context of Southeast Michigan.

5

2.2.1 Tax Competition – Fiscal Externality
A wide variety of public finance studies analyze provision of public goods. Samuelson
benefits (‫ ݅ܤ ܯ 1 ܰ∑ = ܤ ܯ‬to individuals from public goods equal marginal costs (‫ )ܥ ܯ‬of
)
݅=
(1954) formally showed that public goods are optimally provided when the sum of marginal

providing public goods. This condition came to be called the Samuelson condition (or the

Samuelson rule). However, he pointed out that no price mechanism exists to optimally provide
public goods because individuals can consume them at no costs. This is often referred to as
“free-rider” problem.
A prospective about the optimal provision of public goods came from Tiebout (1956). He
argued that if a resident is mobile, he reveals his preferences about policy packages by choosing
a community in which to live and local governments compete with one another for mobile
residents through their own policy packages. He argued that this competition among local
governments overcomes the free rider problem. However, according to Wilson's (1999) review
article, a tax increase in one region creates a positive externality for other regions under the
assumption of a fixed supply total capital stock because capital tends to flow to regions that have
relatively lower tax rates. In this case, an efficient outcome cannot be achieved because local
governments do not take account of this externality when they set property tax rates. This type of
externality is often referred to as “fiscal externalities” (Wildasin, 1986, 1989). In a general model,
tax competition can lead to lower tax rates and under-provision of public goods. In other words,
the tax and spending policies that each local government chooses can affect the allocation of a
mobile tax base among communities. This implies that government officials develop their own
jurisdiction through inducing residents, firms, and capital to move in by manipulating strategic
variables such as taxes and/or spending on public services. This framework also suggests that

6

changes in tax and spending policies are to some degree influenced by competitor communities,
and thus policy changes in competitor communities are not necessarily exogenous to own
community policy changes. Oates (1972) supported this argument that tax competition results in
providing public goods at a level below the optimal level. Hoyt (1991b) also showed that 1)
changes in the tax rates distort the flow of capital among the jurisdictions and therefore the tax
rates and public services are set at levels below the optimal level; and 2) this wasteful tax
competition is aggravated as the number of jurisdictions, involved in tax competition, increases.
This framework suggests that policy changes in competitor communities are not necessarily
exogenous to one’s own community policy changes.
In recent years, researchers have increasingly recognized that taking into account
neighboring competitor activities is important to identify and to measure strategic interactions
3

among communities in policy implementation. In much of this work, rather than attempt to
explicitly identify and account for neighbor activities, researchers make assumptions on which
neighbors are potentially most important (contiguous neighbors for example) and develop a
spatial weighting matrix to address spatial dependence via spatial autoregressive models. Anselin
(1988), LeSage and Pace (2009) and Brueckner (2003) provide a comprehensive overview of the
models and empirical studies that examine the strategic interactions between local governments
using explicit spatial econometric techniques.
Several papers use spatial econometric methods to examine the tax base effects of tax
rates in the context of tax competition. Defining competitor communities based on geographic
proximity such as distance and population size, Brett and Pinkse (2000) fail to find significant
effects of tax competition on the tax base. Buettner (2003) finds that average tax rates in
3

See Besley and Case (1995), Brett and Pinkse (2000), Buettner (2003), and Gérard et al., (2010)
as examples.
7

neighboring communities are a significant determinant of the local tax base, but only for small
jurisdictions. However, using a new approach that exploits migration patterns to determine
regional competitors, Skidmore, et al. (2012) find strong tax competition effects on property
value growth. That is, change in both the own and competitor tax rates are substantial
determinants of own property value growth, holding other factors constant. In this dissertation, to
examine regional competition effects, following Skidmore, et al. (2012), I define three types of
competitors: distance, population, and migration competitors. My findings in this analysis are
consistent of Skidmore, et al. (2012): 1) Migration appears to be better to explain regional
competiton effects; and 2) using migration patterns to account for competitor activities, I find
strong fiscal externalities indicating that regional competition plays a vital role for property value
growth.

2.2 Tax Abatements
In reviewing the literature on tax abatements, it is clear that there is an ongoing debate
over the effectiveness of (temporal) tax incentives for local development. Proponents of tax
incentives argue that property taxes (or business taxes) have a negative influence on firm
location decisions and employment (Bartik, 1989; Charney, 1983; Mcguire, 1985; Papke, 1991;
Wasylenko, 1980). These studies suggest that tax abatements may attract new firms and/or to
retain existing firms. In addition, some studies suggest that tax abatements are relatively more
effective in struggling cities such as Detroit. Bartik (1991, 1994) argues that since the benefits of
additional job growth may be much greater in high than low unemployment communities,
redistributing jobs from low unemployment communities to high unemployment communities is
not a zero-sum game for the region as a whole. This conjecture is empirically tested by Goss and

8

Phillips (1999). Using detailed data on Nebraska’s tax incentive program, Goss and Phillips
(1999) find evidence that business tax incentives are important determinants for economic
growth, but this evidence is true only for low-unemployment counties. Although Goss and
Phillips (1999) fail to find support the argument put forth by Bartik (1991, 1994), they show
business incentives can have a positive influence on economic growth.
Opponents of tax abatement policies contend that taxes have a relatively small marginal
(or no significant) impact on firm location decisions and thus on economic activity in a state and
metropolitan region (Stephen, et al., 1994; Wolkoff, 1985). Wassmer (2007) argues that although
taxes could have a negative impact on firms, the response to tax abatements could be different
than the response to a tax. In the context of tax competition, Bartik (1992) points out that if tax
incentives in a given community tend to be matched by other communities within a given
metropolitan region, then the benefits from tax incentives are more likely to be reduced. Further,
Wassmer (2009) argues that due to the relatively low appeal of high tax communities, tax
incentives could be more effective when used there, but this is likely to be true only if neighbor
communities do not offer abatements.
The existing empirical literature offers somewhat mixed results, but most research
focusing on Southeast Michigan, finds tax incentives to be generally ineffective. Wassmer (1994)
fails to find evidence that property tax abatements stimulate economic development in the region.
Wassmer and Anderson (2001) also find that manufacturing property tax abatements had a
positive effect on property values, but only for the first three years following the adoption of the
abatement legislation in 1974; manufacturing tax abatements have significant negative effects on
economic activities after 1987. Based on these findings, they argue that industrial tax abatements
are effective only for a short-time period because copycat behaviors among communities are

9

more likely to happen in later periods. Using the duration model, Anderson and Wassmer (1995)
also offer evidence of competition phenomena among communities; the rate of tax abatement
adoption rises over time. These empirical findings suggest three important implications: 1) If
one community in a metropolitan area offers tax abatements and others do not, I can expect
significant effects of tax abatements on business activity, 2) there is evidence of mimicking
behavior in the offering of tax abatements, and 3) mimicking behaviors of local governments for
tax abatement policy reduce tax abatement effectiveness.
While these studies discuss the role of competition, they do not empirically account for
the effects of competitor policy changes on a given community’s economic development. To
isolate the effects of own tax abatements on industrial property value growth, it is important to
account for other influences, including the fiscal activities of competitor governments. Failing to
account for competitor fiscal activities could lead to biased and misleading results.
With regard to the effects of tax abatements on industrial property value growth, in this
study, I address the following questions in the context of Southeast Michigan:


Do tax abatements promote industrial property value growth?



Do tax abatements in competitor communities affect own industrial property value
growth? That is, is there a fiscal externality caused by tax abatements?



Are property tax rates a significant determinant for industrial property value growth?



Does tax competition play an important role for own industrial property value growth?
In the next section, I provide a theoretical overview of how fiscal activities of local

governments influence firm location decisions and discuss why officials might choose to use tax
abatements to promote economic development.

10

2.2.1 Firm Location
The Tiebout model has been modified and reapplied in various studies in the tax
competition literature. In the context of the present study, one important modification is an
extension of the model from mobile households to mobile firms. In particular, Fischel (1975),
White (1975), and Wellisch (2000) explored the linkages between firm location decisions and the
optimal provision of local public services. They argue that local services are optimally provided
if firms pay property tax rates equal to the marginal costs of the local services they receive and
physical externalities, such as pollution, caused by firms. This argument implies that, similar to
the original Tiebout hypothesis, competition among local governments for mobile firms results
in an optimal level of public goods with no added costs associated with firm location behavior.
Others have noted the potential for harm associated with competition spillover effects,
which are often referred to as fiscal externalities (Wildasin, 1989; Hoyt, 1991a; Wilson and
Wildasin, 2004). For example, suppose property tax rates in a community are cut such that they
are below average among communities. In this case, the reduction results in an inflow of capital
from higher tax communities. This capital inflow is the fiscal externality, implying that local
governments may have an incentive to offer too many tax abatements, which can result in
inefficiently low property tax rates driven by a desire to attract and/or retain business. Unless
local governments internalize this fiscal externality, local government competition leads to an
under provision of local public goods. Of course, the degree of inefficiency depends on degree of
firm mobility; higher mobility leads to greater fiscal externalities.
If mobile firms respond to changes in property taxation, tax abatements could play a
significant role in attracting new firms, retaining existing firms and encouraging them to

11

reinvestment in a given community. With the aim of promoting local economic growth in
Michigan, hundreds of millions of dollars in property tax revenues are forgone through the
offering of tax abatements each year. While tax abatements have been used extensively, they can
only be justified if firms generate fiscal benefits that outweigh the forgone revenues from tax
4

abatements. Also, in the context of competition, tax abatements are inefficient if they distort
capital flows—this is the result if local governments fail to take into account the fiscal
externalities associated with use of tax abatements.
To assess the effectiveness of tax abatements, the core questions are: 1) do industrial
firms generate significant fiscal benefits; and 2) do tax abatements result in fiscal externalities?
With regard to fiscal benefits, Fischel (1987) offers a clear explanation of how they could be
generated by a new plant or the rehabilitation of an old plant. Consider Figure 2.1, which is
taken from Fischel (1987, pp.307). Initially, I consider a case with no tax abatement and no
physical externality such as pollution. In Figure 2.1, a representative voter perceives the budget
line AB between private goods (assumed to be “before-tax” income) and local public goods,
where the initial equilibrium point is H. A representative voter pays AC in (property) taxes to
receive OD in local public goods. Suppose that industrial property is rehabilitated or improved
by new construction and the assessed value of property increases, which in turn results in
additional property tax revenues. In this scenario, the voter faces the new budget line AE; as
compared to point H, any point between X and Y makes the voter better off because the voter
receives a fiscal benefit via a reduction in property tax rates, an increase in local public services,
or both.

4

In this case, fiscal benefits can be considered as local revenues, generated by firms, in excess of
the costs of local public services they consume.
12

In this context, consider tax abatements like the IFT abatement program where property
tax rates are cut in half for new industrial facilities and the assessed value is frozen on qualifying
property. Also, assume the tax abatement is in effect for the specified time period. If firms make
investments in industrial property in response to the tax abatement, the available local revenues
are reduced as compared with the case where firms do the same without tax abatements.
However, after tax abatements expire, if firms remain in a community, then the assessed values
are reset based on true market value and therefore fiscal benefits in the later periods are
significant. However, there are several scenarios where firms will generate lower or no fiscal
benefits in the tax abatement program scenario: 1) New development or the rehabilitation of an
existing site may require additional local public services such as road maintenance and
police/fire protection; 2) firms may leave the community or go bankrupt prior to the expiration
date of the abatement; or 3) tax abatements result in inefficient regional competition. In these
scenarios, use of tax abatements could be ineffective and inefficient.
In the next chapter, I offer a detailed description of important features of property tax
system, which will help to lay out the empirical challenges, and the IFT abatement program in
Michigan.

13

Figure 2.1 Fiscal Benefits from Industrial Property
Private Goods

A

X
H

Y

C

O

D

B

14

E

Local Public Goods

CHAPTER 3
Property Tax and the IFT Abatement Program in Michigan

3.1 A History of the Property Tax in Michigan
To fully inform my empirical strategy and analysis, it is important to understand the
recent history of the property tax in Michigan. In particular, I discuss two key policy changes:
the Headlee Amendment and Proposal A. Prior to the implementation of Proposal A, property
5

tax revenues were limited by the “Headlee Amendment,” which was passed in 1978. That is, the
Headlee Amendment restricts property tax revenue growth to the rate of inflation (with an
adjustment for new construction). Any jurisdiction with potential revenue increases exceeding
the Headlee limit is required to reduce property tax rates to bring revenues into line with the
revenue growth restriction. This type of tax rate reduction is known as a “Headlee rollback.”

6

Feldman, et al. (2003) provide the following example to explain how the Headlee Amendment
works:
“For example, given an inflation rate in consumer prices of 2.5%, if the tax base
increased from $1,000,000 to $1,100,000 (excluding new construction), and if the tax rate
were one mill, the millage would have to be reduced to 0.932 so that the yield would be
the same as that generated by the one mill on the original tax base adjusted for inflation $1,025.”
Before the introduction of the taxable value cap, rapidly rising property values resulted in
numerous Headlee rollbacks. After Proposal A, however, rollbacks were greatly reduced in both
number and magnitude. Thus, after 1978 and before 1994 (Proposal A), the Headlee Amendment

5

The Headlee Amendment is named for its author, Richard H. Headlee.

6

Local residents can choose to exceed the Headlee limitation by referendum, but his has
occurred only rarely.
15

provided a mechanism for limiting property tax rates in a uniform manner across all properties in
a jurisdiction. Proposal A effectively instituted a new system for limiting effective property tax
rates, but the mechanism did not treat all properties in a jurisdiction uniformly.
I would like to emphasize several features of Proposal A: 1) A residential property (and a
7

qualified agricultural property) was entitled to exemptions from local school operating taxes ; 2)
the sales tax and cigarette tax were increased and a new statewide property value-based six mill
state education tax was imposed; and 3) unlike the Headlee Amendment, a cap on property value
8

assessment growth was imposed. These features imply the following:


While overall property tax rates were reduced, the homestead exemption meant that the
residential property tax rate was reduced more so than the commercial/industrial property
tax rate.



After the imposition of Proposal A, the funding for public schools shifted from local
governments to the state. Further, local governments’ control over property taxes was
reduced.



Poorer communities received relatively larger tax reductions and greater funding for local
school operating costs from the state as compared to wealthy communities.



During periods of rapidly rising home values, the assessment cap creates the gap between
state equalized valuation (SEV), which is the 50% of the estimated market value, and
taxable valuation (TV). Under the assessment growth cap, assessments are readjusted

7

This is commonly known as the “Homestead Exemption.”

8

As long on as a homeowner does not sell his/her home the assessed value is only allowed to
growth at the rate of inflation or 5%, whichever is lower. The taxable value cap is often referred
to as an assessment growth limit.

16

based on the market value (that is, SEV equals TV) only when properties are sold. This
implies that there are significant differentials in property taxes between old residents and
new residents (Skidmore, et al., 2010). Under the assessment growth cap, the gap
between SEV and TV for commercial and industrial property is smaller than that for
residential property (Feldman, et al., 2002).


Both the number and size of Headlee Rollbacks were reduced because of the imposition
of Proposal A, but rollbacks still occurred during the periods of rising property values.

With this overview of the Michigan property tax I highlight two important issues. First,
simultaneity between tax base and tax rates arises, for example, when a significant decline in the
tax base leads to increases in tax rates and/or decreases in school spending. In Michigan, central
cities such as Detroit have faced chronic financial challenges because of ongoing population
decline. Accordingly, struggling cities must either increase property tax rates to maintain
previous spending levels or cut spending on schools. In Michigan, the endogeneity problem is
exacerbated by the Headlee Amendment. Thus, in my study, endogeneity is of a paramount
concern. Second, as in Skidmore, et al. (2012), changes ushered in by Proposal A enable me to
identify the causal effects of changing tax rates and school spending on residential and business
property value growth. Importantly, these changes were exogenous for all communities because
a statewide referenda process was forced on voters by the legislature.

3.2 Michigan’s IFT Abatement Program
th

During the 20 century, Michigan’s economy was driven largely by manufacturing, and
in particular the auto industry was a dominant factor. Despite the relative prosperity associated
with the strong manufacturing base, there were also disadvantages associated with a heavy

17

reliance on manufacturing that emerged as manufacturing became a globally competitive
industry (Bartik, et al., 2003; Reese and Sands, 2012). First, central cities in Michigan suffered
an ongoing population loss, resulting in a declining property tax base. As the property tax base
eroded, local government officials increased property tax rates and/or decreased local public
services. State and local officials understood that the relatively high and growing property tax
rates in some cities could serve as an impediment to industrial development. In addition, wages
of manufacturing workers were much higher than those in nearby states and states in the south.
In addition, national economic recessions affected Michigan more than other states less
dependent on manufacturing. For example, during the double dip recession in the 1980s, many
older manufacturing facilities were closed and unemployment increased significantly (Block and
Belman, 2003).
In an effort to improve the business environment, state and local governments in
Michigan adopted various types of tax abatements and incentives: Industrial Property Tax
Abatements, Renaissance Zones, Neighbor Enterprise Zones, and the Michigan Economic
Growth Authority (Reese and Sands, 2012). One of the more popular tax incentive programs has
been the Industrial Facilities Tax (IFT) abatement program, often referred to as Public Act 198
(PA 198). The IFT abatement program was designed to encourage industrial firms to make
investments resulting in job creation and retention. For example, in 1973 the Chrysler
9

Corporation wanted to rehabilitate the dilapidated Mack Street Stamping Plant. However,
Chrysler officers argued that the company would not be able to rehabilitate the plant without tax
abatements, emphasizing that there were 5,000 jobs at stake. As a response to the request, local
officials in Detroit lobbied state officials for introduction of the industrial property tax abatement

9

This example is taken from Anderson and Wassmer (2000).
18

law, Public Act 198. This proposal was authorized by state authorities to begin in 1974 and
Detroit city government was able to grant the tax abatement to the Chrysler Corporation. Since
the IFT abatement program was adopted, many other local governments in Michigan have also
used the IFT program to reduce property taxes for eligible manufacturing firms in an effort to
promote economic development. According to Bartik, et al. (2003), this program represents the
largest single use of incentives to promote economic development in Michigan.
To grant an IFT abatement, a local government must first establish a plant rehabilitation
district, an industrial development district, or both. However, to qualify for establishing a district,
property taxes levied should be at least 30 mills. Once a district is established, an eligible
industrial firm can request tax abatements. To obtain tax abatements, the firm promises to make
the investment, creates new jobs and/or retains existing jobs in the community. After a request is
made by a firm, local government officials must decide whether or not to approve an application
within sixty days. Finally, once local leaders approve the application, state government must
approve the application prior to construction or installation of equipment.
An IFT abatement can apply to real and industrial/non-industrial personal property of
manufacturing firms such as buildings, building improvements, machinery and equipment. The
length of each tax abatement certificate is a maximum of twelve years from the completion date
of the facility. Once the tax abatement expires, any buildings, improvements, or
machinery/equipment become fully taxable. Thus, after tax abatement expiration, local
governments can begin collecting more tax revenues as a result of new investments, but in reality,
this does not always occur.
Figure 3.1 shows the number of tax abatements granted in Southeast Michigan over the
period 1974 through 2010. In the 1980s, the use of tax abatements increased, peaking in 1986.

19

Since then, there has been a general decline in the use of tax abatements. Despite the recent
decline, more than fifty tax abatement certificates have been granted every year since 1978 in
this region. With regard to the decline in the use of abatements, Reese, et al. (2009) conducted a
survey of city officials, finding that the main reason for the decline was that requests from firms
dropped over the period. In particular, during an economic downturn, firms are not able to invest
in new development/redevelopment, and this reality in turn leads to a reduction in abatement
requests. This implies that the decision to offer tax abatements is not the result of the policy per
se; rather, market conditions are crucial for demand on the part of firms for tax abatements.
As shown in Figure 3.1, the use of tax abatements tends to decrease during periods of
recession. On the other hand, it may be that local policy makers decide not to offer tax
abatements because they think that the abatement program is both costly and ineffective.
However, this is in generally not the case at least in Michigan (Reese, et al., 2009).
As shown in Figure 3.2, most communities issued less than twenty five IFT certificates in
total over the period of analysis. In general, communities were cautious with extending IFT
abatements. However, as shown in Figure 3.2, the following communities issued a high number
of abatements: Clinton Township, Detroit, Fraser, Sterling Heights, and Chesterfield Township.
In addition, as shown in Table 3.1, the number of communities that had never granted tax
abatements decreased over time.

20

0

50

# of Tax Abatements
100
150

200

Figure 3.1 Number of Tax Abatements Granted. For interpretation of the references to
color in this and all other figures, the reader is referred to the electronic version of this
dissertation.

1970

1980

1990
Year

21

2000

2010

250

Figure 3.2 Total Number of Tax Abatements Granted: 1983-2002. Each dot represents a
community.

Clinton

Quantiles of Total # of Tax Abatements
50
100
150
200

Detroit

0

Fraser
Sterling Heights
Chesterfield

0

.2

.4
.6
Fraction of the data

22

.8

1

Table 3.1 The Percentage of Communities Never Giving A Tax Abatement
Year
Community (%)
1983
58
1987
43
1991
41
1998
35
2002
34

23

CHAPTER 4
The Effects of Changes in Property Tax Rates, School Spending, and Tax Abatements
on Residential, Commercial, and Industrial Property Value growth

4.1 Introduction
Optimal provision of public goods has been discussed and debated in the literature for
decades. In early work, Samuelson (1954) formally derived the necessary conditions, but pointed
out that there is no decentralized pricing mechanism for the optimal provision of public goods
because of the so-called free rider problem. That is, there is no incentive for residents to reveal
true preferences if tax liabilities are based on stated preferences. However, Tiebout (1956)
argued that in the case of local public goods, residents reveal their preferences for public goods
by choosing to live in communities that have the most desirable tax-service packages. In this
case, taxes for local public goods are in some ways analogous to prices of goods allocated
through private markets. Oates (1969) conducted the first empirical test of Tiebout’s hypothesis.
Using data from a cross-section of communities in northeastern New Jersey, Oates (1969) found
that the net effect of tax reductions and education expenditure increases on property values is
close to zero. This result was used to argue that the local public service (education) was being
provided at close to the optimal level. Since Oates’ study, a large body of literature has been
devoted to examining the effects of tax rates and local public spending on property values

10

(see

for example, Bradbury et al., 2001; Brueckner, 1982, 1979; Guilfoyle, 1998; Haughwout, et al.,
2004; Lang and Jian, 2004; Oates, 1973; Palmon and Smith, 1998; Pollakowski, 1973). Despite
this now substantial body of research, little consideration has been given to the effects of tax

10

For comprehensive literature reviews on the subject of capitalization, see Yinger, et al. (1988)
and Ross and Yinger (1999).
24

rates and local public spending on different classes of property. Oates (1969) argued that if
residents are mobile and shop around for communities that provide preferable levels of local
public services at the lowest property tax liability, property taxes and the quality of local public
services are capitalized into property values. However, residential property owners may prefer
quite a different tax/service package than commercial or industrial interests. Further, businesses
and their investments are also mobile. It is therefore likely that changes in taxes and public
service spending have varying impacts on different classes of property, yet to my knowledge,
such an examination has not been the focus of existing research. The main purpose of this
chapter is to fill this gap in the literature by examining the responsiveness of property values of
different classes of property to changing tax and spending regimes in a regional competition
framework.
To address this question I use panel data from 152 communities in five counties
surrounding Detroit, in Southeast Michigan between 1983 and 2002. In Michigan, public schools
are primarily financed by a property tax, which is levied at different rates on different classes of
property. In this analysis, I focus on the three primary property classifications: Residential,
commercial and industrial.

11

As I discuss in greater detail later, mobile agents with different sets

of preferences make locational decisions based on their tax/service needs/preferences such that
the property values of various property classes likely respond differently to changes in taxes and
school spending. The work in this chapter extends Skidmore, et al. (2012) that examine how
changing taxes and school spending affects overall property value growth in Southeast Michigan.
The focus of the present work is to determine the degree to which residential, commercial, and

11

Agriculture, forest, and swampland property classifications are omitted.
25

industrial property values respond differently to changes in property tax rates and school
spending.
To estimate the effects of changing tax rates and school spending on property value
growth, I must take into account the endogeneity of property tax rates and school spending
decisions. Following Skidmore, et al. (2012), the instruments are based on the exogenous policy
shift brought on by the statewide imposition of Proposal A in 1994. Proposal A, which was
chosen among two education finance options via statewide referenda, was implemented to
reduce property tax burdens and improve funding equity in public schools through voter
approval.

12

Proposal A resulted in significant tax rate reductions across communities because

funding sources for public schools were shifted to state government revenues generated from
new sales and cigarette taxes, and a new statewide property value-based six mill state education
tax. Also, since this reform was designed to reduce the gap in school spending between poor and
wealthy communities, the former received relatively more funding from the state. As discussed
in detail later, these exogenous changes in tax rates and per pupil school spending for public
schools across communities enable me to identify causal relationships between policy changes
and property value growth of different property classes.
In the analysis, I also take into account tax competition. Specifically, if policy changes in
competitor communities affect one’s own property value growth, then a failure to account for
these leads to biased and inconsistent estimates. Thus, to avoid potential omitted variable bias
resulting from spillovers, I estimate the property value effects of tax rates and school spending in
the context of tax competition.

12

The legislature abolished the old system of school finance and then offered two proposals to
be considered by voters via statewide referenda. Proposal A was the successful proposal.
26

The analysis offered here reveals the following: 1) Residential property values are more
responsive to school spending changes than property tax rate changes; 2) commercial and
industrial property values are more responsive to tax rate changes than school spending changes;
3) commercial and industrial property values are more sensitive to changes in property taxes
relative to residential property values; and 4) there are significant fiscal externalities; that is, tax
competition plays an important role in property value growth in the region.
Another interest of this chapter is to examine the effectiveness of tax abatements in
Michigan. To do so, there are several issues that must be discussed. First, as I mentioned in
chapter 3, industrial facilities tax (IFT) abatements are mainly targeted at industrial facilities.
This suggests that that IFT abatements benefit only industrial firms. From the perspective of
local government, to justify IFT abatement program, the fiscal benefits of tax abatements must
offset their costs, which are the foregone tax revenues. Thus, to examine whether tax abatements
are effective, I should compare net present value of the stream of property tax revenue of firms
who receive tax abatements with foregone tax revenues resulting from granting abatements.
Further, if the fiscal benefits are substantial and local governments can lower the tax rates and/or
improve local public services to create favorable tax/spending environment for
residents/commercial firms, then it is possible for tax abatements to create spillover effects on
residential/commercial property values. Thus, in this chapter, I also examine the effects of tax
abatements on residential and commercial property value growth. This study reveals several
findings: 1) Localities that offer tax abatements yield statistically significant positive impacts on
industrial property value growth; 2) these impacts are larger in high tax than in low tax
communities; 3) the benefits of tax abatements are much smaller than the cost of offering tax
abatements; that is, tax abatements are cost-ineffective; 4) there are positive spillover effects of

27

tax abatements on residential and commercial property value growth, but tax abatements are still
cost-ineffective even when the spillover benefits are considered; and 5) tax abatements offered in
competitor communities do not appear to affect the own industrial property value growth.

4.2 Residential, Commercial and Industrial Property
The property tax is an important source of local revenues, but it is useful to note that
property tax revenues are generated from different classes of property. Further, owners of
different property classes sometimes require different types of local public services. Thus,
changes in property taxes and public services may result in different levels of capitalization
across property classifications. In this chapter, I seek to measure the degree to which policy
changes result in different rates of capitalization across residential, commercial and industrial
property classes.
According to the Tiebout hypothesis, mobile residents sort themselves across local
communities in accordance with their policy preferences. This hypothesis implies that taxes
function as a pricing mechanism for local public services as residents are willing to pay their
share of the costs for these services in the communities they choose. This notion can also be
applied to mobile firms (commercial and industrial firms). Fischel (1975), White (1975), and
Wellisch (2000) argue that if firms are perfectly mobile and shop around among communities
that offer different policy packages in a way that is analogous to mobile residents, local
government competition results in the efficient provision of public goods. Thus, values of
business property also reveal the policy preferences of firms through their locational decisions. If
agents make locational decisions based on policy preferences and have different policy
needs/preferences, then property value responses to policy changes will differ across classes of

28

property. This argument leads to a question: How do fiscal policy preferences differ among
agents? In a Cost Of Community Services (COCS) study

13

in Scio Township, Southeast

Michigan, Crane, et al. (1996) show that residents pay less than their share of costs of providing
local public services, especially school services, whereas commercial/industrial firms contribute
more revenue than the costs of local services they receive.

14

Ladd (1975) also argues that the

higher fraction of non-residential property lowers the property tax burden on residents for service
and consequently they demand relatively high levels of school services. Thus, residents may be
more concerned about school spending than the property tax when commercial/industrial
properties contribute to school service fiscal capacity. Also, in this case, commercial/industrial
firms are more likely to respond to tax policy because they receive less in public services than
the tax revenues they generate. Furthermore, Bartik (1991) argues that if the cost function of
firms is similar across communities within the region, then property tax differentials could be an
important location determinant for businesses, which are assumed to be motivated by profit
maximization.

15

Luce (1994) examines the effects of fiscal policy on the location of employment

and households simultaneously in the Philadelphia area, using a multiple-equation crosssectional model. He finds empirical evidence that school spending is only a significant
determinant of household location choices. He also shows that property taxes and local public

13

COCS studies provide insights on the impact of different land uses on revenues and
expenditures of local governments. For more details, see Freedgood (2004).
14

The findings of COCS studies are consistent with those above showing high community costs
associated with residential land use and lower costs associated with commercial and industrial
land use. For more examples, see Freedgood (2004).
15

For empirical examples, see Wasylenko (1980), Charney (1983), Mcguire (1985), Bartik
(1989), Papke (1991).
29

services affect the location decision of firms, but the effect of the property tax is larger than
public services. However, he fails to find an impact of school spending on firm behavior.
Based on this discussion and the previous research presented above, I pose two hypotheses:
H1: Residential property values are more responsive to changes in school spending than
changes in property taxes.
H2: Commercial and Industrial property values are more responsive to changes in property
taxes than changes in school spending.
As I describe more fully below, the findings confirm that while residential property
values are more responsive to school spending changes than property tax policy changes,
business property values are more responsive to changes in property taxes. This study provides
new evidence that property values across property classes respond differently to changes in local
government policies.

4.3 Data and Descriptive Statistics
To examine the effects of policy changes in property taxes and school spending on the
value growth of classes of property, I use data from a panel of 152 communities in the five
county region surrounding Detroit over the 1983-2002 period. All variables in the data set are
available from United State Census Bureau, Michigan Department of Treasury, Michigan
Department of Education, and the Federal Bureau of Investigation (Table 4.A.1 in APPENDIX
4.A).
The dependent variables are the values of three different property classifications:
residential, commercial, and industrial property, where state equalized valuations (SEV)

16

In Michigan, SEV is defined as 50 percent of the estimated market value.
30

16

are

used as a proxy for property values.

17

Across the jurisdictions, residential property accounts for

68 percent of total aggregate SEV on average, whereas business property, (which is composed of
both commercial and industrial property) accounts for 18 percent. With regard to industrial
property, nine percent of the observations for industrial property values are truncated at zero. To
account for the corner solution nature of industrial property values, in the framework of fixed
effects and endogenous effects of fiscal policy, I might have used a Tobit model to estimate the
effects of tax abatements on industrial property value growth. However, the Tobit specification
does not provide qualitatively similar results in comparison with those from the core
specifications. It may be that decisions regarding whether to permit industrial firms at all verses
how much industrial activity is allowed are not the same. For example, industrial firms often
generate negative externalities for residents such as pollution and congestion. This externality
can inhibit industrial firms from moving into a community. In such a case, the Tobit model
generates biased estimates and thus is not appropriate (Cragg, 1971).
Following Skidmore, et al. (2012), I match current year of reported property values with
lagged values of the other variables of interest because SEV reflects market values in the
previous year. In addition, to account for the full impact of the changes in tax rates and school
spending brought on by Proposal A, I define the transition period from 1993 to 1995 because
Proposal A was partially implemented in 1994, but did not fully take effect until 1995.
The control variables I use are property tax rates, school spending, tax abatements and
crime rates in Michigan. Summary statistics are reported in Tables 4.1 and 4.2.

17

I note that SEV captures both price responses (capitalization) and quantity responses (the
number of houses/businesses) to the impacts of changes in fiscal policies. The focus of this
chapter is to examine differential responses of residential, commercial and industrial property
value growth to fiscal policy changes, so it does not consider the quantity and price responses
separately. However, I note that this is an important topic for future research.
31

Table 4.1 Summary Statistics and Variable Definitions
Std.
Variables
Obs.
Mean
Definition
Dev.
Dependent Variable
Residential
Property Values
Commercial
Property Values
Industrial
Property Values

2,888

17,944

1,7029

2,888

2,869

3,406

2,888

1,838

4,474

G_IFT

2,888

0.474

1.098

E_IFT

2,888

0.167

0.597

2,888

21.96

10.13

2,888

25.44

7.19

School Spending

2,888

5,340

1,867

Crime Rates

2,888

38.51

58.72

1/2 of Estimated Residential
Market Value per Capita
1/2 of Estimated Commercial
Market Value per Capita
1/2 of Estimated Industrial
Market Value per Capita

Own Policy Variables

Residential
Property Tax
Non-Residential
Property Tax

The Number of Granted Tax
Abatements Per 1,000 Population
The Number of Expired Tax
Abatements Per 1,000 Population
1/2 of Residential Property Tax Rates
Per $1,000 of Taxable Value
1/2 of Non-Residential Property Tax
Rates Per $1,000 of Taxable Value
General Fund School Expenditures
per Pupil
Uniform Crime Index per 1,000
Capita

Competitor Policy Variables
Migration Competitor Variables
CG_IFT

2,888

0.291

0.156

CE_IFT

2,888

0.827

0.090

2,888

20.91

8.90

2,888

24.10

5.30

School Spending

2,888

5,150

1,668

Crime Rates

2,888

26.28

8.30

Residential
Property Tax
Non-Residential
Property Tax

32

The Number of Granted Tax
Abatements Per 1,000 Population
The Number of Expired Tax
Abatements Per 1,000 Population
Competitor Residential
Property Tax Rates
Competitor Non-Residential
Property Tax Rates
Competitor General Fund
School Expenditures per Pupil
Competitor Uniform Crime Index

Table 4.1 (cont’d)
Variables

Obs.

Mean

Std.
Dev.

Definition

Distance Competitor Variables
CG_IFT

2,888

0.498

0.689

The Number of Granted Tax
Abatements Per 1,000 Population

CE_IFT

2,888

0.177

0.392

The Number of Expired Tax
Abatements Per 1,000 Population

Residential
Property Tax

2,888

22.13

9.70

Competitor Residential
Property Tax Rates

Non-Residential
Property Tax

2,888

25.55

6.33

Competitor Non-Residential
Property Tax Rates

School Spending

2,888

5,322

1,759

Competitor General Fund
School Expenditures per Pupil

Crime Rates

2,888

36.48

28.62

Competitor Uniform Crime Index

Population Competitor Variables
CG_IFT

2,888

0.481

0.564

The Number of Granted Tax
Abatements Per 1,000 Population

CE_IFT

2,888

0.173

0.334

The Number of Expired Tax
Abatements Per 1,000 Population

Residential
Property Tax

2,888

21.89

9.28

Competitor Residential
Property Tax Rates

Non-Residential
Property Tax

2,888

25.34

5.69

Competitor Non-Residential
Property Tax Rates

School Spending

2,888

5,316

1,707

Competitor General Fund
School Expenditures per Pupil

Crime Rates

2,888

39.31

31.01

Competitor Uniform Crime Index

33

Table 4.2 Summary Statistics Pre- and Post-Proposal A for Key Variables
Std.
Std.
Variables
Obs.
Mean
Obs.
Mean
Dev.
Dev.
Before Proposal A
After Proposal A
Dependent Variable
Residential
Property Values
Commercial
Property Values
Industrial
Property Values

1,672

11,702

9,496

1,216

26,527

20,920

1,672

2,157

2,402

1,216

3,847

4,240

1,672

1,531

4,288

1,216

2,260

4,687

1,672

29.45

5.38

1,216

11.67

4.57

1,672

29.45

5.38

1,216

19.93

5.52

1,672

4,099

1,163

1,216

7,047

1,177

1,672

28.32

1.96

1,216

10.71

1.81

1,672

28.32

1.96

1,216

18.29

1.75

1,672

3,885

780

1,216

6,890

738

1,672

29.67

4.15

1,216

11.75

3.68

1,672

29.67

4.15

1,216

19.89

4.00

1,672

4,094

1,025

1,216

7,012

986

1,672

29.39

3.05

1,216

11.56

2.75

1,672

29.39

3.05

1,216

19.78

3.28

1,672

4,088

957

1,216

7,003

863

Own Policy Variable
Residential
Property Tax
Non-Residential
Property Tax
School Spending
Competitor Policy Variables
Migration Comp.
Residential
Property Tax
Non-Residential
Property Tax
School Spending
Distance Comp.
Residential
Property Tax
Non-Residential
Property Tax
School Spending
Population Comp.
Residential
Property Tax
Non-Residential
Property Tax
School Spending

34

With regard to property tax rates, in Michigan, the overall property tax rate

18

consists of three

primary components: The county tax rate, the city or township tax rate, and the school tax rate.
The same county tax rate is levied on all property within a given county, but other tax rates are
determined by cities, townships, and school districts.

19

As discussed in earlier, there is

substantial variation in property tax rates before and after Proposal A. Prior to Proposal A, the
average millage across the region was 29.45 mills. After Proposal A was passed, the average
millage decreased by more than nine mills, but the reductions varied substantially across
jurisdictions. Generally, low tax base/high tax rate jurisdictions received a larger tax rate
reduction than high tax base/low tax rate jurisdictions. In addition, as a result of Proposal A,
principal residence properties are not subject to local school taxes. Thus, residential property tax
rates were reduced more than those for commercial or industrial property. Finally, after Proposal
A all properties were subject to a new six mill statewide education tax. Regardless, all property
classifications in all communities experienced a substantial reduction in property tax rates as a
result of Proposal A.
Proposal A also resulted in a shift of education finance responsibilities from the local
level to state government, and statewide education spending increased (Feldman, et al., 2003). In
this new system of school finance, poorer communities experienced significant increases in
school funding, whereas wealthy communities were allowed to impose property tax millage rates
to maintain their original spending levels. The end result was that the gap in school spending

18

This tax rate is the statutory property tax millage rate. One mill is defined as $1 per $1,000 of
taxable value. Because statutory property tax rates are twice effective property tax rates for 50
percent of the estimated market value, I use 1/2 of statutory property tax rates as effective
property tax rates in this chapter.
19

After 1994, a six mill statewide education tax was also imposed.
35

between low and high spending communities was reduced, but the overall level of school
spending in the region increased. I use changes in school spending per pupil resulting from
Proposal A as a measure of exogenous changes in local public services.
The next key control variable is the number of tax abatements offered in each year (the
number of IFT certificates) over the period 1983 through 2002. There are four issues that warrant
discussion here. First, to control for community size, tax abatements are standardized by
population size. Standardization is important because the number of tax abatements is likely to
be greater in communities with larger populations. For example, as shown in Figure 4.1, the
total number of tax abatements for the period 1983 to 2002 is greater in Detroit, which is the
most populated city in the region. However, as shown in Figure 4.2, once I standardize by
population, I see that Detroit offers relatively few tax abatements. Second, I use the first
differenced cumulative number of tax abatements to capture effects over time. Next, whereas tax
abatements benefits are targeted at industrial firms, it is possible for such abatements to create
spillover effects on residential/commercial property values if a development project receiving tax
abatements generates enough fiscal benefits to attract new residents/commercial firms and/or
increase the value of existing firms. Thus, I also measure the effects of abatements on residential
and commercial property values. Finally, I must also take into account the (cumulative) effects of
expired tax abatements because: 1) tax abatements may have an impact on industrial property
value growth after they expire; and 2) a new tax abatement is sometimes granted immediately
following the expiration of an existing abatement such that the granting and expiration of
abatements may be correlated. This means that a failure to control for expirations may result in
omitted variable bias. To fully assess the effectiveness of tax abatements and to avoid the

36

omitted variable bias, I, therefore, include a variable for the cumulative number of expired tax
abatements in all specifications.
Figure 4.1 Total Number of Tax Abatements Granted: 1983-2002

Detroit

Legend
Southeast Michigan
Total # of Abatements
0.0-3.0
3.1-11.0
11.1-25.0
25.1-59.0
59.1-109.0
109.1-166.0
166.1-greater

0

37

12,000 24,000

48,000 m

Figure 4.2 Total Number of Tax Abatements Granted Per 1,000 Population: 1983-2002

Detroit

Legend
Southeast Michigan
Total # of Abatements
Per 1,000 Population
0.000-0.142
0.143-0.497
0.498-0.981
0.982-1.646
1.647-2.770
2.771-4.541
4.542-greater

0

38

12,000 24,000

48,000 m

Finally, I use the number of crimes per 1,000 to capture temporal unobservable shocks to
the quality of living in communities, which may be correlated with tax base. The studies, which
examine tax base effects of the (business) property tax, often include the unemployment rate to
control for temporal changes to the local economy. Unfortunately, this variable is not available
for every community for each year I consider in the analysis. In addition, to avoid omitted
variable bias I include community fixed effects and community specific time trends. This
inclusion controls for unobserved community fixed factors and community specific time varying
factors in the empirical specifications.

4.4 Empirical Model
I estimate the effects of changes in property tax rates, school spending, and tax
abatements on property value growth for each property class using a first difference
specification:

20

ܰ
ܰ
݆
ܸܲ݅ = ‫ + ߛ ݅ܺ + ߚ ݅ ܫ‬෍ ܹ ݅ ‫ + ߜ ݆ ܫ‬቎ ݅ − ෍ ܹ ݅ ݆ܺ ቏ߠ
‫ݐ ܶܨ‬
ܺ‫ݐ‬
‫ݐ‬
݆ ‫ݐ ܶܨ‬
݆ ‫ݐ‬
‫ݐ‬
݆= 1
݆= 1

+݉ ݅+ ߤ݅ ‫ , ݅݁ + ݐ‬j=1,2,3
‫ݐ +ݐ‬
‫ݐ‬

݆
where ܸܲ݅ represents the natural logarithm of the property value ݆(݆=1, 2, 3 for residential,
‫ݐ‬

(4.1)

commercial, and industrial property, respectively) for community ݅in period, ‫ ݅ ܫ‬is a vector
‫ݐ ܶܨ ݐ‬

abatements (E_IFT) for community ݅in period, ܺ݅ is a vector of variables for community ݅in
‫ݐ ݐ‬
of variables that include the number of granted tax abatements (G_IFT) and expired tax

20

This is the so-called “Spatially Lagged X Model (SLX)”, which is a sub-category of the
Spatial Durbin Model (SDM).

39

period ‫ݐ‬that includes the natural logarithm of aggregate effective tax rates of all overlying taxing
and the natural logarithm of crime rate per 1,000 population, ∑ܰ 1 ܹ ݅ ‫ ݆ ܫ‬and
݆ ‫ݐ ܶܨ‬
݆=

authorities that apply within the community, the natural logarithm of school spending per pupil,
∑ܰ 1 ߱ ݅ ݆ܺ represent the analogous set of variables for competitor communities where
݆ ‫ݐ‬
݆=

(ܹ ݅ = ܹ ݀[݅ ݆ , ܹ ‫ ,) ݆ ݅[ ݐ ݎ݋ , ݆ ݅[݌‬௜ is jurisdiction fixed effects, ߠ݅
, ]
, ]
ܹ݉ , ] ݉
‫ݐ‬represents the
݆

competitors are defined by the spatial weighting matrix, described in equations (4.6)-(4.8) below,
community-specific time trends for community and ‫ ݐ‬represents the time indicator variables
݅
௧
21

to capture time effects.

22

To eliminate unobserved community effects, I estimate equation (4.1) using a firstܰ
ܰ
݆
∆ܸܲ݅ = ∆‫ ∆ + ߛ ݅ܺ∆ + ߚ ݅ ܫ‬෍ ܹ ݅ ‫ ∆ + ߜ ݆ ܫ‬቎ ݅ − ෍ ܹ ݅ ݆ܺ ቏ߠ
‫ݐ ܶܨ‬
ܺ‫ݐ‬
‫ݐ‬
݆ ‫ݐ ܶܨ‬
݆ ‫ݐ‬
‫ݐ‬
݆= 1
݆= 1

difference (FD) procedure and the specifications, thus, are based on the following:

+ߤ݅+ ∆‫ , ݅݁∆ + ݐ‬j=1,2,3 (4.2)
‫ݐ‬
‫ݐ‬

Also, I employ a cluster approach in which standard errors are clustered at the community level
to address temporal autocorrelation. Cluster-standard errors perform well when the number of
clusters is reasonably large (Bertrand, et al., 2000; Kezdi, 2004). Further, I take into account
spatial autocorrelation. If there is spatial autocorrelation, the last error term in the equation (4.2)

21

This accounts for unobserved community factors changing over time that may affect property
value growth. For example, community-specific time trends can proxy for the impacts of
changes in quality of life or environment that may affect property values. I also note that the
main results remain unchanged if I omit the community-specific time trends in the specifications.
22

The first difference specification also addresses the positive autocorrelation of error terms I
find in the data.
40

can be expressed as: ∆݁݅ = ∆ ∑ܰ 1 ܹ ݅ ݁݅ ߮ + ∆‫ . ݅ݒ‬Thus, in this case, I can rewrite the
‫ݐ‬
݆ ‫ݐ‬
‫ݐ‬
݆=
ܰ
ܰ
݆
∆ܸܲ݅ = ∆‫ ∆ + ߛ ݅ܺ∆ + ߚ ݅ ܫ‬෍ ܹ ݅ ‫ ∆ + ߜ ݆ ܫ‬቎ ݅ − ෍ ܹ ݅ ݆ܺ ቏ߠ
‫ݐ ܶܨ‬
ܺ‫ݐ‬
‫ݐ‬
݆ ‫ݐ ܶܨ‬
݆ ‫ݐ‬
‫ݐ‬
݆= 1
݆= 1
specification as:

spatial autocorrelation, φ = 0:

+ߤ݅+ ∆‫ , ݅ݒ∆ + ߮ ݅݁ ݅ ܹ 1 ܰ∑ ∆ + ݐ‬j=1,2,3 (4.3)
‫ݐ‬
݆ ‫ݐ‬
‫ݐ‬
݆=

To test for spatial autocorrelation, I use a regression-based test using the null hypothesis of no

(1) Obtain the residuals (∆݁݅ ) for all (i, t) pairs by estimating equation (4.2).
Ƹ‫ݐ‬

Procedure 4.1

(2) Estimate ∆ ∑ܰ 1 ܹ ݅ ݁ଓ using the residuals (∆݁݅ ) and replace the ∆݁݅ in equation
Ƹ‫ݐ‬
݆ෞ‫ݐ‬
‫ݐ‬
݆=
(4.2) with [∆ ∑ܰ 1 ܹ ݅ ݁ଓ ]߮ + ∆‫ . ݅ݒ‬The specifications for the test are based on
݆ෞ‫ݐ‬
‫ݐ‬
݆=

(3) Use pooled OLS to estimate parameters (ߚ, ߛ, ߜ, ߠ, and ߮) in equation (4.3) and perform
equation (4.3).
a ‫ݐ‬test on ߮
ො.

23

As a result of the test for spatial autocorrelation, I detect significant spatial autocorrelation for
some specifications. In these specifications, I estimate the effects of policy changes on
residential, commercial, and industrial property value growth taking into account spatial
autocorrelation. In some specifications, I include contemporaneous and (one-, two-, and three-

Another version of this test is proposed by Born and Breitung (2011): 1) Estimate the
following equation: ∆݁݅ = ∆ ∑ܰ 1 ܹ ݅ ݁݅ ߮ + ∆‫ ; ݅ݒ‬and then 2) perform a t test on ߮ Born
Ƹ‫ݐ‬
ො.
݆ Ƹ‫ݐ‬
‫ݐ‬
݆=
and Breitung (2011) show that a regression-based test for spatial autocorrelation is
asymptotically equivalent to the traditional test such as the Moran’s I and LM tests. These
methods are very similar to test for serial correlation (Wooldridge, 2010, pp198-199).
23

41

year) lagged values of all policy variables to examine the length of time it takes for the policy
changes be fully reflected in property values.

4.4.1 Endogeneity
In the analysis, simultaneity between tax base and the policy variables (the tax/spending
policy and tax abatement variables) is a primary concern.

4.4.1.1 Property Taxes and School Spending
The endogeneity problem occurs when a significant decline in the tax base leads to
increases in tax rates and/or decreases in school spending. Also, as I discussed earlier, this
problem is exacerbated by the Headlee Amendment. To overcome the endogeneity of property
taxes and school spending, following Skidmore, et al. (2012), I exploit the imposition of
Proposal A, which resulted in differential changes in property tax rates and school spending
across communities in Michigan. Recall that Proposal A, authorized by voters in the state in
1994, had the following characteristics: 1) Prior to its imposition, local governments relied
heavily on property tax revenues to fund K-12 education. As previously discussed, Proposal A
resulted in significant changes in property taxes and total funding per pupil for public schools.
Specifically, Proposal A shifted school funding responsibilities to state government, which was
paid for with revenues from sales taxes, cigarette taxes, and a new statewide property valuebased six mill state education tax; 2) local property tax rates were reduced because of the
reduction in local school operating millage rates; 3) for homeowners, for schools only the six
mills state education tax is levied, thereby reducing tax rates even further; and 4) poorer
communities received relatively larger tax rate reductions and greater funding for school

42

spending relative to wealthier communities. These exogenous changes in the taxes and the per
pupil school spending for public schools resulting from Proposal A enable me to identify causal
relationships between policy changes and property value growth. In this context, I use change in
the natural logarithm of the tax rate and change in the natural logarithm of the per pupil school
spending resulting from the imposition of Proposal A as the two key identifying instruments. To
explore lagged effects of property taxes and school spending, I included one-, two- and threeyear lags of the logarithms of tax rates and school spending in equations (4.1), which are
three-years (߬ 1 , ߬ 2 , and ߬ 3 ) and the logarithms of per pupil spending lagged one-,
‫−ݐ‬
‫−ݐ‬
‫−ݐ‬
potentially endogenous. Since I also treat both logarithms of tax rates lagged one-, two-, and

two-, and three-years (ܵ‫ , 2 −ݐܵ , 1 −ݐ‬and ܵ‫ ) 3 −ݐ‬as potentially endogenous, I use one-, two-,
and three-year lagged changes in the logarithms of tax rates and per pupil school spending
resulting from Proposal A as additional instruments, respectively.

24

4.4.1.2 Tax Abatements
To determine whether tax abatements are endogenous, the key questions are: Is the tax
abatement decision of local policy makers in a community driven by the demand of firms or the
supply of local governments (local policy)? If local governments are willing to offer tax
abatements, but they cannot grant them because firms do not request them, then local offers of
tax abatements are a function of requests by firms. This case is more likely when firms do not
consider building new plants, rehabilitation of old plants, or installation of machinery/equipment
because of a difficult economic climate. However, there is also the possibility that local
24

To further improve the efficiency of the instrumental variables technique, I added property tax
and school spending five-years lagged to the set of instrumental variables for the current property
tax and school spending, respectively.
43

governments decide to stop offering tax abatements because officials think that tax abatements
are costly and ineffective in terms of spurring economic development. As reported earlier, Reese,
et al. (2009) argue that the use of tax abatements is more likely to be driven by firm demand. In
such a case, controlling for community fixed effects and community-specific year trends may
enable me to identify the causal linkage between tax abatements and industrial property value
growth. Nevertheless, as discussed next, I offer a formal test of exogeneity.
I examine whether the tax abatement variable is exogenous using the Hausman Test,
which requires me to identify at least one valid instrument that is correlated with tax abatement
use but not directly with industrial property value growth. The instrument I use is the city mayor
and township supervisor election cycle.

25

Depending on how voters interpret tax abatement

policies, the political costs and benefits of offering tax abatements may vary over the election
cycle. For example, if voters believe that tax abatements are beneficial (harmful) in a
community, elected local officials may be more (less) likely to offer tax abatements in the year
prior to the election. Thus, the instrument for tax abatements is the political variable, which is a
dummy variable that equals 1 in one year prior to mayoral or township supervisor elections and 0
otherwise.

26

As shown in Table 4.B.3-4.B.6, APPENDIX B, as compared with other years, tax

abatements are more likely to be granted the year before the election and this estimated effect is
statistically significant. Thus, the election cycle variable is a good predictor of tax abatements.

25

The election data are from several sources: City/Township Charter, County election official
homepages, city official homepages, and Michigan Townships Association homepages. In
general, each city has its own two or four-year election cycles for mayor, but all townships have
four-year election cycles for supervisors (as well as clerks, treasures and trustees).
26

I also tried to use the dummy variables for election year and/or one year after election as
instruments. However, since these dummy variables do not have any predictive power, I
excluded them in the list of instruments.
44

To test whether I have the weak instruments problem, I conduct F-tests for the excluded
instruments for the first-stage regressions, which are statistically significant with the exception of
one regression (column1 in Tables 4.B.3-4.B.6, APPENDIX B). To further examine the weak
instrument problem in the presence of non-iid errors, as Baum, et al. (2007) suggested, I use a
Kleibergen-Paap (KP) Wald rk F statistic for the weak identification (ID) test. The KP Wald rk
statistics from all first-stage regressions are less than two. The critical values for Stock-Yogo
weak ID test provided by Stock and Yogo (2005) are not available because I have five
endogenous and nine instrumental variables (they provide critical values for up to three
endogenous variables). Thus, I use the informal threshold of 10, the “Staiger and Stock rule of
thumb”, as a critical value to test the null hypothesis that the instruments are weak (Staiger and
Stock, 1997). Using the KP Wald rk F statistics, for all first stage regressions, I fail to reject the
null hypothesis, indicating that the joint strength of the excluded instruments is weak even
though the election cycle variable is a statistically significant predictor of tax abatements.
Nevertheless, I conduct the Hausman test to determine whether endogeneity is present. In the test,
I fail to find evidence of endogeneity, and therefore prefer the specification treating the tax
abatement variable as exogenous. However, as a robustness check, I also present instrumental
variable specifications in which the tax abatement variable is treated as endogenous.

4.4.2 Regional Competition
To address regional competition, I considered three approaches for determining
competitors. Traditional methods for defining competitor communities are based on information
on distance and population. In addition to these traditional approaches, as in Skidmore, et al.

45

(2012), I use another approach that is based on intra-regional migration patterns. I therefore
consider three definitions of competitors: distance, population, and migration:


Distance (four nearest) competitors: community j is a competitor of i if it is one of the
four closest jurisdictions to community i.



Population competitors: community j is a competitor of i if it is one of the four closest in
population size to community i.



Migration competitors: community j is a competitor of i if many who had previously
lived in community i migrated to community j.

27

To calculate average competitor variables, I need to use an appropriate weighting matrix. Each
weighting matrix, corresponding to the three competitor definitions as described above, is based
on the following:

1
݀ , ]
ܹ ݀[݅ ݆ = ‫ × ݆ ݅[ ܫ‬൭
, ]
൱ (4.4)
          
         
݀ , ]
∑݆‫݆ ݅[ ܫ‬

1
‫] , ݌‬
ܹ ‫ × ݆ ݅[ ܫ = ݆ ݅[݌‬ቆ
, ]
‫ ݆ ݅[݌‬ቇ
∑݆‫] , ܫ‬

(4.5)

1
݉ , ]
ܹ ‫ = ݆ ݅[ ݐ‬Out-migrants݉ [݅ ݆ ൭
 (4.6)
     
‫ ∑ ] , ݐ‬Out-migrants݉ [݅ ݆ ൱
݆
‫] , ݐ‬

݀ , ]is
where ‫ ݆ ݅[ ܫ‬an indicator variable, which takes the value of 1 if a community ݆is a “distance

‫] , ݌‬
competitor” of ݅and 0 otherwise, ‫ ݆ ݅[ ܫ‬is an indicator variable, which takes the value of 1 if a

27

The findings remain consistent if I use “migration competitors” based on the following
definition: community j is a competitor of i if it is one of the four closest in the number of outmigrants to community i. This definition is consistent with other competitor definitions.
46

݉
community ݆is a “population competitor” of ݅and 0 otherwise, and Out-migrants‫ ݆ ݅[ ݐ‬is out, ]
migrants per capita from a community ݅to a community ݆ which is a “migration competitor”.
,

To determine competitor communities based on intra-regional migration patterns, I need

data on out-migrants at the community level. These data should include information on where
out-migrants moved. If such data were available, then I can calculate “competitor” variables as
weighted averages using the ratio of out-migrants to total out-migrants who moved to a
community in all “competitor” communities. Unfortunately, these data are not available and
hence must be estimated.
Following the procedure of Skidmore, et al. (2012), this estimation was achieved. First, I
have data on in-migration at the subdivision level, but these data do not indicate specifically from
where the in-migrants came. Second, to estimate origin, I use data on out-migration at the county
level that includes information on where out-migrants moved. However, I need to make several
assumptions to calculate out-migration:


Assumption 1: Cities and townships within a given county have the same out-migration
as the county as a whole.



Assumption 2: Since Census sources do not provide data on out-migrants who moved
from one community to another within the same county, I follow Schachter et al. (2003)
and further assume that in the sample 24.9 percent of county population moved from one
community to another within the same county.

47

Next, with these two assumptions, I calculate out-migrants from one community to another by
multiplying out-migrants at the county level by the ratio of in-migrants to total in-migrants at the
28

݉
Out-migrants‫( = ݐ‬Out-Migrants݆ × (In-Migrants Ratio݅ )
)
݆
community level:

where each ݅ and ݆represents community and county, respectively.
,

(4.7)

To illustrate more concretely, assume that 1,000 people move from Wayne to Macomb

County.

29

Recall that my goal is that I use information on out-migrants at the county level and

in-migrants at the community level to estimate the number of out-migrants who move from one
community to another. Let me further suppose that there are a total of 2,000 in-migrants to
Macomb County from elsewhere in the United States. If there were 100 in-migrants to Clinton
Township (Clinton Township is located in Macomb County), the ratio of in-migrants to total inmigrants would be 0.05 (=100/2,000). To estimate out-migrants at the community level, I
multiplied total out-migrants (1,000) at the county level by the ratio of in-migrants to total inmigrants for all communities. To obtain the estimated number of out-migrants from Wayne
County to the Township of Clinton, I multiplied 1,000 by 0.05 for an estimate of 50 out-migrants.
I used this method to calculate the out-migrants to other counties in the region for each
community in the sample. I then transformed all the variables using the ratio between the
estimated number of out-migrants at the community level and total out-migrants summed over
all “competitor” communities. From this ratio, I generated weighted averages for “competitor”
variables. A more detailed explanation of the methods used to determine the competitors
28

To control for community size, population inflow and outflow are divided by community
population to calculate the per capita inflows and outflows.
29

This illustration is taken from Skidmore, et al. (2012).
48

weighted average is provided in APPENDIX 4.C. I acknowledge that because of data limitations
on migration activity, I can only generate an approximation of competitor communities.
However, as I demonstrate in the analysis, this approach seems to perform better than the
approaches typically used in the literature, at least in the context of Southeast Michigan.
Once I obtain competitor variables as weighted averages, I include them in my
specifications to investigate the effects of regional competition. There are several items worth
noting before I turn to a discussion of the empirical analysis: 1) To examine the effects of the
changes in competitor tax/spending policies on residential and business property value growth, I
use the relative changes in property tax rates and school spending between one’s own
community and competitor communities; and 2) in the estimates that take into account
competitor activity, I transform the instruments in a way that is analogous to the method I used to
transform the other explanatory variables. As Skidmore, et al. (2012) mentioned, in this study,
using the relative fiscal position of a community is important because the instruments, used to
identify exogenous variation in changes in the tax/spending policies, are based on the imposition
of Proposal A. Proposal A led to changes in property taxes and school spending across all
communities, but the magnitudes of changes varied. These magnitudes reflecting the relative
fiscal position of a community could play an important role for the own industrial property value
growth. That is, depending on the relative magnitudes of tax reductions between the own
community and competitor communities, the effect of the own tax reduction on property value
growth could be significant or not. For example, if the property tax reductions in competitor
communities are larger than the reduction in the own property tax rates, then the effect of the
own tax reduction may not be significant. Therefore, following Skidmore, et al. (2012), to

49

examine regional competition effects, I use the relative changes in property taxes and school
‫݊ݓܱ( ݈ = ܶܲܥ‬
݊
ܲ‫)ݏ݁ݔܽܶ ݐݎ݁݌݋ݎܲ ݐ ݐ݁݌ ݉݋ܥ( ݈ − )ݔܽܶ ݐݎ݁݌݋ݎ‬
‫ݕ‬
݊
݅ ‫ݎ݋‬
‫ݕ‬

spending between one’s own community and competitor communities, illustrated as follows:
‫ܿܵ( ݈ = ܵܥ‬
݊ ℎ‫ܿܵ ݐ ݐ݁݌ ݉݋ܥ( ݈ − ) ݅݀݊݁݌݈ܵ݋݋‬

݊݃
݊
݅ ‫ ݎ݋‬ℎ‫݅݀݊݁݌݈ܵ݋݋‬

݊݃)
‫)ݏ݁ ݁ ݅ݎܥ ݐ ݐ݁݌ ݉݋ܥ( ݈ − )ݏ݁ ݁ ݅ݎܥ( ݈ = ܥܥ‬
݊
݉ ܴܽ‫ݐ‬
݊
݅ ‫ݐܴܽ ݉ ݎ݋‬

(4.8)
(4.9)
(4.10)

Equations (4.8), (4.9), and (4.10) imply that changes in property tax rates, school

spending, and crime rates relative to competitors are important to own property value growth.

30

If competition is important for property value growth, then the sign of coefficients on CPT, CS,
and CC can be expected to be positive, negative and positive, respectively.

4.5 Empirical Results
To examine the property value effects of changes in property tax rates and school
spending, I estimate equation (4.2) for each of the three classes of property. Because of possible
endogeneity, I use an instrumental variable technique using the instrumental variables discussed
above. Tables 4.B.1 and 4.B.2 in APPENDIX B display the first stage regression results,
showing that instrumental variables are strong predictors of endogenous variables. The Fstatistics for excluded instruments for all endogenous variables from the residential and business
(commercial/industrial) property value equations are between 1,013 and 8,314, indicating that I
do not have the weak instruments problem. Also, the Sargan-Hansen test for Overidentifying
Restrictions shows that the instrumental variables are valid for all sets of regressions.
Core regression results are shown in Tables 4.3- 4.8. I present four regression
specifications in all tables. The first specification (column 1, Tables 4.3, 4.5, and 4.7) only
30

As noted in the discussion of equation (4.1), all variables including those illustrated by
equations (4.8), (4.9), and (4.10) are first-differenced.
50

includes current values of one’s own explanatory variables. In the second specification, I control
for competitor activities by including competitor explanatory variables (columns 2-4, Tables 4.3,
4.5, and 4.7). In this specification, I define competitor communities based on distance,
population size, and intra-migration flows. In the third specification, I include one-year, two-year,
and three-year lagged values of the own policy variables in the first specification (column1,
Tables 4.4, 4.6, and 4.8). In addition to the lags of the own policy variables, in the last
specification I include one-year, two-year, and three-year lagged values of competitor policy
variables (columns 2-4, Tables 4.4, 4.6, and 4.8). The inclusion of these lags allows me to
examine residual effects of changes in the own and competitor policy on property value growth,
showing the length of time it takes for policy changes to be fully reflected in industrial property
values.
In all tables, regional competition appears to be better explained when “competitors” are
based on migration flow information: 1) The degree of the impacts of property taxes on one’s
own residential, commercial, and industrial property value growth are larger; and 2) the results
show much clearer effects of school spending and school spending relative to competitors on the
own residential property value growth. These findings show that how competitors are defined is
important, and is consistent with the finding of Skidmore, et al. (2012) that intra-regional
migration patterns are a better approach for identifying competitors. Thus, because the results
improve considerably when migration flow information is used to determine competitors, I focus
on columns 2 to discuss about regional competition effects.
Finally, as mentioned earlier, I prefer regression specifications that treat tax abatements
as exogenous. However, I also present the regression results from specifications with the tax
abatement variable treated as endogenous in Tables 4.16-4.21 and discuss these estimates in the

51

robustness section. With regard to tax abatements, I take into account the cumulative number of
granted tax abatements (G_IFT) and the cumulative number of expired tax abatements (E_IFT).
The coefficients on G_IFT and E_IFT provide estimates of the effects of another granted and
expired IFT certificate on industrial property value growth, holding other factors constant,
respectively.

4.5.1 Residential Property
4.5.1.1 Property Taxes and School Spending
Consider first the residential property estimates presented in Table 4.3. In columns 1,
property taxes have significant effects on residential property value growth. In columns 2-4, I
present the estimates that take into account regional competition. In column 2, the coefficient on
the property tax variable is much larger as compared to that displayed in columns 1. In addition,
unlike columns 1, school spending now shows a significant effect on residential property value
growth. The coefficient on the school spending variable is also larger than the coefficient for
property tax. This initial set of estimates suggest that: 1) failing to account for competition
effects biases the estimates toward zero; and that 2) the elasticity of property values with respect
to property taxes is much smaller than school spending. In other words, resident property values
are more responsive to changes in school spending than changes in property tax rates.
Turning to Table 4.4 that present regression results in which the lagged policy variables
are included, in column 1, I also find evidence that the initial impact of changes in property taxes
on residential property values is negative and statistically significant, but the impacts dissipate
after about two years. However, in these regressions, school spending changes are not a
substantial determinant of residential property value growth. The regression presented in column

52

2 includes current and lagged own and competitor policy variables. Here, the initial impact of tax
policy changes is significant, but this impact dissipates over time, though the sign of the
coefficients on three-year lagged property taxes is unexpectedly positive. I also find school
spending to be a significant factor for residential property value growth. Similar to the tax rate
effects on property value growth, the initial impact of school spending changes tends to diminish
over time. For the long run effects, holding other factors constant, a 10 percent tax reduction and
a 10 percent school spending increase will increase property values by 3.8 and 18.0 percent,
respectively. These results suggest that it takes about three years for the policy changes to fully
generate residential tax base responses. These estimates also show that school spending is a
much more important than tax rate changes to residential property value growth. Further, when I
take into account regional tax competition, the net effects of tax policy changes depend on one’s
standing relative to competitors. For example, holding competitor tax rates constant, a tax
reduction improves one’s own community’s relative tax position as compared to competitor
communities, thereby further increasing tax base growth; the estimates suggest that a 10 percent
tax reduction further increases property value growth by 3.2 percent when I consider tax
competition effects. However, in this case, if competitor communities reduce tax rates, then the
own community’s relative tax position worsens, thus the net effects of a tax reduction in the own
community become smaller. In all specifications, crime rates have no significant effect on
residential property value growth.

4.5.1.2 Tax Abatements
Consider first the estimates of the coefficients on the number of granted tax abatements
(∆G_IFT) and the number of expired tax abatements (∆E_IFT), presented in Table 4.3. In Table

53

4.3, regardless of specifications, I find positive and statistically significant coefficients on
∆G_IFT and ∆E_IFT. Focusing on column 2,

31

the result shows that 1) on average, an increase

of one IFT certificate (per 1,000 population) will increase residential property values by 0.08
percent; and 2) the expiration of a tax abatement (per 1,000 population) is associated with a 0.05
percent increase in residential property values. It implies that tax abatements create positive
spillover effects on residential property value growth. In addition, I also find statistically
(∆CG_IFT and ∆CG_IFT), but the signs of these coefficients are unexpectedly positive. There
significant coefficients on the number of granted and expired competitor tax abatements

are two possible explanations for this finding. First, competitor tax abatement variables may be
highly correlated with another unobservable variable, which is not included in this analysis and
this excluded variable could cause the unexpected sign. Second, the unexpected sign on
competitor tax abatement variables may be because of a high positive correlation between the
use of tax abatements in the own and competitor communities. For example, APPENDIX 4.E
shows the estimated time trend slopes in the number of granted tax abatements (G_IFT) and of
granted competitor tax abatements (CG_IFT). For each community, G_IFT and CG_IFT have a
similar positive time trend. Also, the estimated correlation between the two variables is 0.57.
This high correlation may cause this unexpected result.
The findings remain consistent when I account for competitor policy effects as well as the
lagged effects of policy variables although there are no residual effects of tax abatements
(columns 2-4, Table 4.4).

31

This column 2 presents the regression results from the specification using migration patterns
to account for competitor activities.
54

Table 4.3 First Difference Results I-1
Dependent Variable: ∆ Ln Residential Property Values Per Capita
Specification

‡

(1) FDIV

(2) FDIV

(3) FDIV

(4) FDIV

Include Comp. Variables

No

Yes

Yes

Yes

The Types of Competitors
The Own Variables

–

Migration

Distance

Population

0.073***
(0.025)
0.043***
(0.015)
-0.081***
(0.020)
-0.009
(0.078)
0.001
(0.002)

0.076***
(0.025)
0.045***
(0.015)
-0.270***
(0.040)
0.973***
(0.184)
0.004
(0.007)

0.067**
(0.028)
0.043***
(0.013)
-0.123***
(0.021)
-0.015
(0.100)
0.001
(0.002)

0.070***
(0.025)
0.040***
(0.015)
-0.062*
(0.036)
0.021
(0.140)
0.0002
(0.002)

1,976

0.169***
(0.051)
0.379***
(0.136)
0.233***
(0.046)
-0.995***
(0.191)
-0.003
(0.006)
1,976

0.054*
(0.032)
-0.033
(0.035)
0.097***
(0.030)
0.037
(0.127)
0.0004
(0.001)
1,976

-0.032
(0.037)
-0.081**
(0.033)
-0.017
(0.038)
-0.034
(0.120)
0.001
(0.001)
1,976

∆ G_IFT
∆ E_IFT
∆ Ln Property Tax
∆ Ln School Spending
∆ Ln Crime rate
The Competitor Variables
∆ CG_IFT
∆ CE_IFT
∆ Ln Property Tax
∆ Ln School Spending
∆ Ln Crime rate
N

Notes:
1. All regressions include a series of time indicator and community-specific time trend variables.
2. Robust cluster standard errors in parentheses
3. “Property Tax” indicates the residential property tax.
4. Spatial autocorrelation is taken into consideration in the specifications with the symbol (‡) in
the first row.
5. * significant at 10%; ** significant at 5%; *** significant at 1%

55

Table 4.4 First Difference Results I-2
Dependent Variable: ∆ Ln Residential Property Values Per Capita
‡
Specification
(1) FDIV
(4) FDIV
(2) FDIV
(3) FDIV
Include Comp. Variables
No
Yes
Yes
Yes
The Types of Competitors
–
Distance
Population
Migration
The Own Variables
0.074**
0.062*
0.072**
0.082***
∆ G_IFT
(0.032)
(0.033)
(0.030)
(0.030)
0.024
0.025
0.021
0.036
∆ G_IFT t-1
(0.021)
(0.020)
(0.021)
(0.022)
-0.008
-0.005
-0.014
-0.010
∆ G_IFT t-2
(0.017)
(0.016)
(0.016)
(0.018)
-0.006
-0.008
-0.009
-0.007
∆ G_IFT t-3
(0.020)
(0.020)
(0.020)
(0.019)
0.044***
0.046***
0.040***
0.048***
∆ E_IFT
(0.017)
(0.016)
(0.015)
(0.017)
-0.008
-0.015
-0.010
-0.017
∆ E_IFT t-1
(0.017)
(0.018)
(0.016)
(0.019)
-0.003
-0.007
-0.007
-0.012
∆ E_IFT t-2
(0.012)
(0.012)
(0.012)
(0.012)
0.024
0.017
0.026*
0.027*
∆ E_IFT t-3
(0.015)
(0.013)
(0.015)
(0.015)
-0.090***
-0.133***
-0.078*
-0.251***
∆ Ln Property Tax
(0.021)
(0.024)
(0.040)
(0.056)
-0.042***
-0.067***
-0.050*
-0.147***
∆ Ln Property Tax t-1
(0.012)
(0.014)
(0.026)
(0.046)
-0.024*
-0.056***
-0.038
-0.130***
∆ Ln Property Tax t-2
(0.013)
(0.017)
(0.026)
(0.049)
-0.001
-0.006
-0.048**
0.149***
∆ Ln Property Tax t-3
(0.013)
(0.016)
(0.021)
(0.049)
-0.044
-0.143
-0.078
1.156***
∆ Ln School Spending
(0.095)
(0.129)
(0.160)
(0.248)
0.052
0.074
0.013
0.657***
∆ Ln School Spending t-1
(0.069)
(0.081)
(0.131)
(0.245)
0.022
-0.007
0.043
0.318
∆ Ln School Spending t-2
(0.049)
(0.074)
(0.136)
(0.273)
0.101*
0.203***
0.028
0.382
∆ Ln School Spending t-3
(0.057)
(0.070)
(0.111)
(0.250)
0.002
0.002
0.001
0.009
∆ Ln Crime rate
(0.002)
(0.002)
(0.002)
(0.008)

56

Table 4.4 (cont’d)
Specification
The Competitor Variables

(1) FDIV

(2) FDIV

(3) FDIV

‡

(4) FDIV

1,520

0.176***
(0.053)
0.155***
(0.046)
0.009
(0.053)
0.027
(0.051)
0.414***
(0.156)
0.231
(0.188)
-0.236
(0.198)
0.291
(0.205)
0.211***
(0.063)
0.134***
(0.048)
0.118**
(0.049)
-0.143***
(0.053)
-1.240***
(0.234)
-0.635***
(0.230)
-0.300
(0.266)
-0.277
(0.253)
-0.007
(0.008)
1,520

0.039
(0.041)
-0.002
(0.038)
-0.033
(0.033)
0.038
(0.039)
-0.039
(0.042)
-0.021
(0.035)
-0.053*
(0.030)
0.050
(0.033)
0.111***
(0.031)
0.064***
(0.022)
0.058***
(0.022)
0.010
(0.026)
0.144
(0.137)
0.007
(0.098)
0.081
(0.098)
-0.197**
(0.086)
0.001
(0.001)
1,520

-0.014
(0.044)
0.079
(0.059)
-0.022
(0.038)
0.001
(0.049)
-0.076**
(0.038)
-0.086***
(0.032)
-0.063*
(0.036)
0.008
(0.036)
-0.008
(0.041)
0.014
(0.024)
0.023
(0.023)
0.055***
(0.022)
0.036
(0.137)
0.031
(0.108)
-0.032
(0.121)
0.060
(0.092)
0.002
(0.001)
1,520

∆ CG_IFT
∆ CG_IFT t-1
∆ CG_IFT t-2
∆ CG_IFT t-3
∆ CE_IFT
∆ CE_IFT t-1
∆ CE_IFT t-2
∆ CE_IFT t-3
∆ Ln Property Tax
∆ Ln Property Tax t-1
∆ Ln Property Tax t-2
∆ Ln Property Tax t-3
∆ Ln School Spending
∆ Ln School Spending t-1
∆ Ln School Spending t-2
∆ Ln School Spending t-3
∆ Ln Crime rate
N

Notes:
1. All regressions include a series of time indicator and community-specific time trend variables.
2. Robust cluster standard errors in parentheses
3. “Property Tax” indicates the residential property tax.
4. Spatial autocorrelation is taken into consideration in the specifications with the symbol (‡) in
the first row.
5. * significant at 10%; ** significant at 5%; *** significant at 1%

57

4.5.2 Commercial Property
4.5.2.1 Property Taxes and School Spending
The core results for commercial property are presented in Tables 4.5 and 4.6. Consider
first Table 4.5. Column 1 shows no statistically significant relationship between property tax
rates and commercial property value growth. Also, school spending is not a significant factor for
commercial property value growth. However, these estimates may be biased because
specifications do not account for competitor community activities. In column 2, when I account
for regional competition, property tax is a marginally significant factor in business property
value growth. School spending shows a positive but insignificant effect. In addition, competitor
tax has only marginally significant effects on one’s own business property value growth and
competitor school spending does not have a significant effect. However, turning to Table 4.6,
once I take into account both competitor activities as well as lagged effects of the policy
variables, I find property tax rates to be a significant factor for commercial property value
growth. In the long run, the effects of tax changes reach a peak two years after policy changes
and then dissipate in subsequent years. The long run tax elasticity of commercial property is –
1.77. Moreover, this long run property tax elasticity is much larger than that of residential
property. Also, school spending is found to be marginally significant and its coefficient is 0.783,
which is much smaller than that of property tax rates. For all specifications, I fail to find
coefficients on crime rates to be statistically significant at the five percent level. In summary,
commercial property values are more sensitive to changes in tax policy than residential property
values, and property tax rates are much more important than school spending for commercial
property value growth. I also find strong tax competition effects for commercial property value

58

growth. This implies that it is important to keep tax rates competitive for local commercial
businesses.

4.5.2.2 Tax Abatements
statistically significant coefficients on ∆G_IFT, but I fail to find a statistically significant
Consider first Table 4.5. In Table 4.5, for all specifications, I find positive and

coefficient on ∆E_IFT. As shown in column 2, the result shows that on average, an increase of
one IFT certificate (per 1,000 population) will increase commercial property values by 0.07

percent. This result indicates that tax abatements may also create positive spillover effects on
commercial property value growth.
abatements and the range of coefficients on ∆G_IFT is from 0.64 to 0.07. However, there are no
Turning to Table 4.6, for all specifications, I find a positive spillover effect of tax

residual effects of tax abatements. Also, for all specification I find statistically significant

are unexpectedly negative. As shown in column 2, the negative coefficient on ∆E_IFT is -0.04,

coefficient on the number of expired tax abatements (∆E_IFT), but the signs of these coefficients

indicating that the long run (net) spillover effect of tax abatements is smaller in column 2, Table
4.5 than that in column 2, Table 4.6. Finally, I fail to find evidence that competitor tax
abatements are a substantial determinant for the own commercial property value growth.

59

Table 4.5 First Difference Results II-1
Dependent Variable: ∆ Ln Commercial Property Values Per Capita
Specification

‡

(1) FDIV

(2) FDIV

(3) FDIV

(4) FDIV

Include Comp. Variables

No

Yes

Yes

Yes

The Types of Competitors
The Own Variables

–

Migration

Distance

Population

0.066**
(0.026)
0.016
(0.023)
-0.021
(0.053)
0.077
(0.112)
0.006*
(0.003)

0.071***
(0.025)
0.015
(0.023)
-0.274*
(0.160)
0.375
(0.399)
0.021
(0.015)

0.061**
(0.027)
0.015
(0.023)
-0.087
(0.083)
-0.015
(0.163)
0.003
(0.004)

0.066***
(0.024)
0.017
(0.023)
0.059
(0.109)
-0.046
(0.283)
-0.001
(0.004)

1,976

0.170
(0.115)
-0.022
(0.179)
0.287*
(0.155)
-0.268
(0.372)
-0.015
(0.014)
1,976

0.083
(0.067)
-0.012
(0.046)
0.120
(0.088)
0.145
(0.195)
0.002
(0.002)
1,976

-0.010
(0.079)
-0.007
(0.060)
-0.085
(0.096)
0.104
(0.263)
0.007**
(0.003)
1,976

∆ CG_IFT
∆ CE_IFT
∆ Ln Property Tax
∆ Ln School Spending
∆ Ln Crime rate
The Competitor Variables
∆ CG_IFT
∆ CE_IFT
∆ Ln Property Tax
∆ Ln School Spending
∆ Ln Crime rate
N

Notes:
1. All regressions include a series of time indicator and community-specific time trend variables.
2. Robust cluster standard errors in parentheses
3. “Property Tax” indicates the non-residential property tax.
4. Spatial autocorrelation is taken into consideration in the specifications with the symbol (‡) in
the first row.
5. * significant at 10%; ** significant at 5%; *** significant at 1%

60

Table 4.6 First Difference Results II-2
Dependent Variable: ∆ Ln Commercial Property Values Per Capita
‡
Specification
(1) FDIV
(4) FDIV
(2) FDIV
(3) FDIV
Include Comp. Variables
No
Yes
Yes
Yes
The Types of Competitors
–
Distance
Population
Migration
The Own Variables
0.067**
0.064*
0.068**
0.074**
∆ G_IFT
(0.032)
(0.034)
(0.028)
(0.031)
0.009
0.003
0.010
0.020
∆ G_IFT t-1
(0.029)
(0.029)
(0.031)
(0.031)
-0.003
0.005
-0.002
-0.013
∆ G_IFT t-2
(0.036)
(0.034)
(0.036)
(0.039)
-0.022
-0.020
-0.019
-0.037
∆ G_IFT t-3
(0.028)
(0.027)
(0.028)
(0.029)
0.027
0.023
0.022
0.015
∆ E_IFT
(0.031)
(0.031)
(0.030)
(0.032)
0.024
0.014
0.022
0.025
∆ E_IFT t-1
(0.026)
(0.027)
(0.026)
(0.029)
-0.027*
-0.029*
-0.035**
-0.036**
∆ E_IFT t-2
(0.014)
(0.016)
(0.016)
(0.016)
0.040
0.040
0.041
0.022
∆ E_IFT t-3
(0.034)
(0.034)
(0.034)
(0.039)
0.003
-0.052
0.075
-0.431***
∆ Ln Property Tax
(0.051)
(0.089)
(0.107)
(0.158)
0.011
-0.006
0.051
-0.572**
∆ Ln Property Tax t-1
(0.039)
(0.061)
(0.075)
(0.223)
0.028
0.099
0.003
-0.762***
∆ Ln Property Tax t-2
(0.047)
(0.078)
(0.075)
(0.222)
0.033
0.077
0.105
-0.093
∆ Ln Property Tax t-3
(0.053)
(0.079)
(0.095)
(0.224)
0.086
-0.052
-0.124
0.783*
∆ Ln School Spending
(0.115)
(0.181)
(0.287)
(0.449)
0.032
0.022
0.093
0.468
∆ Ln School Spending t-1
(0.099)
(0.144)
(0.273)
(0.423)
-0.037
0.130
-0.289
0.620
∆ Ln School Spending t-2
(0.111)
(0.152)
(0.255)
(0.395)
0.061
0.115
-0.221
-0.430
∆ Ln School Spending t-3
(0.099)
(0.149)
(0.232)
(0.339)
0.007*
0.004
-0.001
0.017
∆ Ln Crime rate
(0.004)
(0.005)
(0.004)
(0.017)

61

Table 4.6 (cont’d)
Specification
The Competitor Variables

(2) FDIV

(3) FDIV

(4) FDIV

1,520

0.194
(0.129)
-0.041
(0.091)
-0.026
(0.099)
-0.139
(0.113)
-0.378
(0.282)
0.586*
(0.314)
0.136
(0.307)
-0.407
(0.410)
0.472***
(0.152)
0.580***
(0.216)
0.805***
(0.220)
0.136
(0.222)
-0.625
(0.445)
-0.436
(0.428)
-0.582
(0.386)
0.532
(0.338)
-0.009
(0.017)
1,520

0.062
(0.084)
-0.040
(0.084)
0.007
(0.069)
-0.068
(0.075)
0.026
(0.057)
-0.060
(0.067)
0.085
(0.064)
0.107
(0.103)
0.114
(0.093)
0.043
(0.069)
-0.110
(0.075)
-0.066
(0.090)
0.215
(0.221)
0.086
(0.194)
-0.208
(0.199)
-0.107
(0.235)
0.002
(0.003)
1,520

0.057
(0.088)
0.042
(0.078)
-0.182**
(0.087)
0.083
(0.080)
0.006
(0.068)
-0.054
(0.065)
-0.093
(0.072)
0.026
(0.066)
-0.068
(0.093)
-0.033
(0.074)
0.027
(0.071)
-0.079
(0.072)
0.213
(0.264)
-0.016
(0.239)
0.250
(0.205)
0.267
(0.197)
0.009***
(0.003)
1,520

∆ CG_IFT
∆ CG_IFT t-1
∆ CG_IFT t-2
∆ CG_IFT t-3
∆ CE_IFT
∆ CE_IFT t-1
∆ CE_IFT t-2
∆ CE_IFT t-3
∆ Ln Property Tax
∆ Ln Property Tax t-1
∆ Ln Property Tax t-2
∆ Ln Property Tax t-3
∆ Ln School Spending
∆ Ln School Spending t-1
∆ Ln School Spending t-2
∆ Ln School Spending t-3
∆ Ln Crime rate
N

‡

(1) FDIV

Notes:
1. All regressions include a series of time indicator and community-specific time trend variables.
2. Robust cluster standard errors in parentheses
3. “Property Tax” indicates the non-residential property tax.
4. Spatial autocorrelation is taken into consideration in the specifications with the symbol (‡) in
the first row.
5. * significant at 10%; ** significant at 5%; *** significant at 1%

62

4.5.3. Industrial Property
4.5.3.1 Property Taxes and School Spending
The regression results for tax policy and school spending are robust and consistent with
the findings of the previous section (4.5.2): 1) The results show a significant negative
relationship between property taxes and industrial property value growth in the context of
regional competition; 2) competitor community property taxes play an important role for the own
industrial property; 3) school spending and school spending relative to competitors are not
important determinants of industrial property value growth; and 4) crime rates are not a
significant factor for industrial property value growth. These findings suggest that the own
community’s relative tax position plays a crucial role for industrial property value growth. That
is, if the own community reduces its tax rates more than competitor communities such that its
relative tax position is improved, then in addition to the effects of the own tax reduction,
industrial property values are further increased by competition effects. The estimates suggest that
a 10 percent tax reduction increases industrial property values by 17.3 percent in the long run.
However, in the case where the own community’s relative tax position is worsened by tax
reductions in competitor communities, industrial property value growth in the own community is
hurt. Finally, industrial property values are more responsive to tax changes than school spending
changes. Thus, the property tax is a more important factor than school spending for industrial
property value growth.

4.5.3.2 Tax Abatements
The cumulative effects of the abatement program on industrial property value growth can
be significant if the abatements affect the manufacturing firms’ decision to relocate, expand, or

63

rehabilitate plants. While tax abatements attract attention as a critical economic development
policy tool, most studies on Michigan tax abatements conclude that they are inefficient and
largely ineffective. The rationale in many of these studies is that tax abatements result in
inefficient intra-community competition. While these studies recognize that regional competition
can potentially be important assessing the effectiveness of tax abatements, they do not directly
test the hypothesis that use of tax abatements in competitor communities influences the own
industrial property value growth. A primary motivation of this section is to examine the
effectiveness of the IFT abatement program in the context of regional competition.
In column 1, I find a positive and statistically significant coefficient on ∆G_IFT. This result

The core regression results are presented in Tables 4.7 and 4.8. Consider first Table 4.7.

indicates that on average, an increase of one IFT certificate (per 1,000 population) will increase
32

upon the expiration of a tax abatement. The coefficient on ∆E_IFT is 0.110, indicating that the
industrial property values by 0.16 percent . I also find marginally significant positive effects

expiration of a tax abatement (per 1,000 population) is associated with a 0.11 percent increase in
industrial property values. Thus, these findings suggest that communities benefit from tax
abatements when they are offered (partial increase in tax base due to new facility) and then again
when they expire (tax base increases by full amount of the project). The findings remain
consistent when I control for competitor policy effects, but I do not find any evidence that
competitor tax abatements are important factors for the own industrial property value growth.
although the coefficient on ∆G_IFT is slightly larger in column 2.

Further, the results are not sensitive to the way in which competitor communities are defined,

32

To better understand this result, suppose there is community A with 1,000 people. In this
community A, one IFT certificate is associated with a 0.16 percent increase in industrial property
values.
64

In Table 4.8, the regression results are based on the specifications that include lags. In
column 1, there are no significant initial or residual effects of changes in tax abatements on
industrial property value growth. In columns 2-4, I present the results from specifications that
take into account both the lagged effects of policy changes and competitor activities. Columns 2respectively. In columns 2-4, the coefficients on ∆G_IFT are positive and significant, and their
4 show regression results that incorporate migration, distance, and population competitors,

magnitudes are similar across columns. I find a negative coefficient on the second year lagged

tax abatement variable (∆G_IFTt-2) that are marginally significant in column 4, indicating that
the long run (net) effects of tax abatements are larger in column 2 than those in column 4. I also
find evidence that both the own and competitor property taxes play important roles for the own
industrial property value growth, but only in column 2. These findings suggest that the regression
results are sensitive to the definition of competitor communities; economic closeness appears to
be better measured by migration flow information, thus affecting the results of industrial
property value effects for the own and competitor tax abatements. I therefore focus on the
regression results for the specification that includes the migration competitor variables (column
2).

In column 2, the coefficient on ∆G_IFT is 0.160 and marginally significant. Its

magnitude is slightly larger than that displayed in column 1 in Table 4.7 and very close to that in
property value growth as well as effects of the expired tax abatements. For ∆E_IFT, I find a

column 2 in Table 4.7. However, I fail to find residual effects of tax abatements on industrial

∆CG_IFT and ∆CE_IFT to be different from zero: that is, I find no evidence of fiscal

positive coefficient, but it is not statistically significant. Further, I fail to find coefficients on

externalities resulting from tax abatements.

65

Table 4.7 First Difference Results III-1
Dependent Variable: ∆ Ln Industrial Property Values Per Capita
Specification

‡

(1) FDIV

(2) FDIV

(3) FDIV

(4) FDIV

Include Comp. Variables

No

Yes

Yes

Yes

The Types of Competitors
The Own Variables

–

Migration

Distance

Population

0.158**
(0.071)
0.110*
(0.062)
-0.017
(0.106)
0.264
(0.477)
0.004
(0.005)

0.168**
(0.071)
0.111*
(0.062)
-0.315
(0.341)
1.138*
(0.655)
0.010
(0.023)

0.162**
(0.070)
0.109*
(0.062)
0.015
(0.170)
0.258
(0.488)
0.003
(0.006)

0.152**
(0.068)
0.119*
(0.062)
-0.315
(0.261)
-0.872
(0.814)
-0.005
(0.007)

1,976

0.105
(0.158)
-0.401
(0.282)
0.297
(0.297)
-0.932
(0.893)
-0.007
(0.022)
1,976

0.005
(0.209)
-0.061
(0.084)
-0.048
(0.150)
0.041
(0.306)
0.001
(0.005)
1,976

0.120
(0.138)
-0.106
(0.143)
0.317
(0.216)
1.074*
(0.609)
0.010
(0.007)
1,976

∆ G_IFT
∆ E_IFT
∆ Ln Property Tax
∆ Ln School Spending
∆ Ln Crime rate
The Competitor Variables
∆ CG_IFT
∆ CE_IFT
∆ Ln Property Tax
∆ Ln School Spending
∆ Ln Crime rate
N

Notes:
1. All regressions include a series of time indicator and community-specific time trend variables.
2. Robust cluster standard errors in parentheses
3. “Property Tax” indicates the non-residential property tax.
4. Spatial autocorrelation is taken into consideration in the specifications with the symbol (‡) in
the first row.
5. * significant at 10%; ** significant at 5%; *** significant at 1%

66

Table 4.8 First Difference Results III-2
Dependent Variable: ∆ Ln Industrial Property Values Per Capita
Specification

‡

(1) FDIV

(2) FDIV

(3) FDIV

(4) FDIV

Include Comp. Variables

No

Yes

Yes

Yes

The Types of Competitors

–

Migration

Distance

Population

0.140
(0.087)
-0.074
(0.056)
-0.089
(0.058)
-0.046
(0.055)
0.079
(0.077)
0.005
(0.043)
-0.052
(0.050)
0.021
(0.040)
0.030
(0.114)
0.212
(0.188)
-0.008
(0.160)
0.040
(0.069)
0.240
(0.510)
-0.851
(0.648)
-0.212
(0.173)
0.0780
(0.283)
0.004
(0.006)

0.160*
(0.086)
-0.058
(0.056)
-0.093
(0.063)
-0.053
(0.053)
0.081
(0.078)
0.006
(0.042)
-0.063
(0.053)
-0.0001
(0.039)
-0.612
(0.394)
-0.903**
(0.428)
-0.554
(0.522)
-0.041
(0.493)
0.853
(0.856)
0.873
(1.017)
0.328
(0.662)
0.061
(0.553)
0.010
(0.028)

0.155*
(0.088)
-0.063
(0.053)
-0.088
(0.064)
-0.026
(0.061)
0.079
(0.076)
-0.006
(0.047)
-0.047
(0.052)
0.030
(0.044)
0.053
(0.178)
0.429
(0.437)
0.0001
(0.137)
0.167
(0.140)
0.160
(0.546)
-1.047
(0.760)
-0.137
(0.248)
0.099
(0.358)
0.007
(0.007)

0.158*
(0.089)
-0.081
(0.061)
-0.107*
(0.063)
-0.010
(0.064)
0.074
(0.078)
0.023
(0.049)
-0.055
(0.054)
0.030
(0.046)
-0.136
(0.251)
0.079
(0.289)
0.175
(0.221)
0.147
(0.163)
-0.997
(0.884)
-1.535
(1.411)
0.678
(0.995)
1.300**
(0.580)
-0.004
(0.007)

The Own Variables
∆ G_IFT
∆ G_IFT t-1
∆ G_IFT t-2
∆ G_IFT t-3
∆ E_IFT
∆ E_IFT t-1
∆ E_IFT t-2
∆ E_IFT t-3
∆ Ln Property Tax
∆ Ln Property Tax t-1
∆ Ln Property Tax t-2
∆ Ln Property Tax t-3
∆ Ln School Spending
∆ Ln School Spending t-1
∆ Ln School Spending t-2
∆ Ln School Spending t-3
∆ Ln Crime rate

67

Table 4.8 (cont’d)
Specification
The Competitor Variables

(2) FDIV

(3) FDIV

(4) FDIV

1,520

0.152
(0.218)
-0.058
(0.152)
0.251
(0.315)
0.050
(0.169)
-0.384
(0.547)
0.538
(0.644)
-0.116
(0.564)
-0.492
(0.803)
0.661*
(0.364)
1.069**
(0.461)
0.562
(0.525)
0.071
(0.482)
-0.628
(0.951)
-1.824
(1.637)
-0.551
(0.722)
0.006
(0.693)
-0.005
(0.027)
1,520

-0.063
(0.265)
0.052
(0.174)
-0.121
(0.119)
0.001
(0.190)
0.019
(0.102)
-0.253
(0.156)
0.029
(0.111)
0.124
(0.140)
-0.071
(0.156)
-0.335
(0.413)
-0.048
(0.142)
-0.227
(0.158)
-0.094
(0.329)
0.509
(0.556)
-0.162
(0.359)
-0.095
(0.329)
-0.002
(0.006)
1,520

0.262
(0.170)
0.077
(0.240)
0.115
(0.166)
0.305**
(0.152)
-0.067
(0.152)
-0.213
(0.165)
-0.141
(0.183)
0.436
(0.269)
0.197
(0.205)
0.099
(0.169)
-0.214
(0.263)
-0.138
(0.154)
1.206*
(0.637)
0.685
(1.018)
-0.953
(0.934)
-1.126**
(0.525)
0.013*
(0.007)
1,520

∆ CG_IFT
∆ CG_IFT t-1
∆ CG_IFT t-2
∆ CG_IFT t-3
∆ CE_IFT
∆ CE_IFT t-1
∆ CE_IFT t-2
∆ CE_IFT t-3
∆ Ln Property Tax
∆ Ln Property Tax t-1
∆ Ln Property Tax t-2
∆ Ln Property Tax t-3
∆ Ln School Spending
∆ Ln School Spending t-1
∆ Ln School Spending t-2
∆ Ln School Spending t-3
∆ Ln Crime rate
N

‡

(1) FDIV

Notes:
1. All regressions include a series of time indicator and community-specific time trend variables.
2. Robust cluster standard errors in parentheses
3. “Property Tax” indicates the non-residential property tax.
4. Spatial autocorrelation is taken into consideration in the specifications with the symbol (‡) in
the first row.
5. * significant at 10%; ** significant at 5%; *** significant at 1%

68

Bartik (1991, 1994) and Wassmer (2007) argue that tax abatements may be more
effective when they are used in high tax communities because industrial firms appear to avoid
such communities. To directly test this hypothesis, I split the sample into two groups based on
average property tax rates before the imposition of Proposal A: High tax communities and low
tax communities. Based on these two groupings, I create indicator variables for the two groups
and then interact the indicator variables with the tax abatement variables and include them in the
specifications. The regression results are summarized in Table 4.9. This table only reports the
regression coefficients with robust standard errors on tax abatement variables for the high tax
community and low tax community, but note that all the other control variables are included in
the specifications.

33

For the results in Table 4.9, let me focus on column 6 that provides the

respect of the effects of tax abatements in high tax communities, the coefficient on ∆G_IFT is

estimates from the specification that accounts for lagged effects and competition effects. With

column 2 in Table 4.8. I also find a positive and statistically significant coefficient on ∆E_IFT,
0.133 and just marginally significant. Its magnitude is slightly smaller than that displayed in

which is 0.186. However, I fail to find significant effects of tax abatements when they are

offered and when they expire in the low tax community. I also find that the negative coefficient
on the second year lagged tax abatement variable (∆G_IFTt-2) that is marginally significant. This
result suggests that industrial property values in the high tax community are increased when tax
abatements are offered and then again when they expire, but again this is not the case in the low
property tax community. Also, I fail to find residual effects of tax abatements on industrial
property value growth and evidence of fiscal externalities for both the high and low tax
communities.
33

The results for control variables are consistent with previous results.
69

Table 4.9 First Difference Results for High and Low Tax Communities
Dependent Variable: ∆ Ln Industrial Property Values Per Values
Specification

‡

‡

(1) FDIV

(2) FDIV

(3) FDIV

(4) FDIV

(5) FDIV

(6) FDIV

(7) FDIV

(8) FDIV

Include Comp. Variables

No

No

Yes

Yes

Yes

Yes

Yes

Yes

The Types of Competitors

–

-

Migration

Distance

Pop.

Migration

Distance

Pop.

0.153**
(0.075)

0.145*
(0.075)

0.133*
(0.077)

0.181***
(0.065)

0.178***
(0.066)

0.192***
(0.067)

0.133*
(0.076)
-0.003
(0.062)
-0.008
(0.070)
-0.068
(0.072)
0.186**
(0.074)
-0.003
(0.058)
-0.079
(0.070)
0.021
(0.051)

0.121
(0.075)
-0.033
(0.063)
0.021
(0.062)
-0.048
(0.074)
0.198***
(0.075)
0.005
(0.056)
-0.036
(0.060)
0.091
(0.062)

0.105
(0.080)
-0.021
(0.069)
-0.037
(0.069)
-0.043
(0.080)
0.180**
(0.079)
0.040
(0.059)
-0.081
(0.064)
0.015
(0.060)

The Tax Abatement Variables in High Tax Communities
∆ G_IFT

0.140*
(0.075)

∆ G_IFT t-1
∆ G_IFT t-2
∆ G_IFT t-3
∆ E_IFT
∆ E_IFT t-1
∆ E_IFT t-2
∆ E_IFT t-3

0.176***
(0.065)

0.103
(0.076)
-0.029
(0.061)
-0.003
(0.063)
-0.058
(0.071)
0.174**
(0.076)
0.008
(0.054)
-0.056
(0.059)
0.056
(0.054)

70

Table 4.9 (cont’d)
Specification

(1) FDIV

(2) FDIV

(3) FDIV

‡

(4) FDIV

(5) FDIV

(6) FDIV

The Tax Abatement Variables in Low Tax Communities
0.185
0.178
0.192
0.187
0.180*
0.192
∆ G_IFT
(0.119)
(0.152)
(0.119)
(0.117)
(0.108)
(0.151)
-0.124
-0.113
∆ G_IFT t-1
(0.084)
(0.086)
-0.179*
-0.180*
∆ G_IFT t-2
(0.094)
(0.103)
-0.018
-0.021
∆ G_IFT t-3
(0.092)
(0.087)
0.039
-0.014
0.036
0.036
0.041
-0.024
∆ E_IFT
(0.106)
(0.139)
(0.107)
(0.105)
(0.106)
(0.137)
0.029
0.037
∆ E_IFT t-1
(0.062)
(0.062)
-0.025
-0.025
∆ E_IFT t-2
(0.086)
(0.086)
-0.026
-0.036
∆ E_IFT t-3
(0.059)
(0.059)
N
1,976
1,520
1,976
1,976
1,976
1,520
Notes:
1. All regressions include a series of time indicator variables and individual-specific time trend variables.
2. Robust cluster standard errors in parentheses
3. “Property Tax” indicates the non-residential property tax.
4. Spatial autocorrelation is taken into consideration in the specifications with the symbol (‡) in the first row.
* significant at 10%; ** significant at 5%; *** significant at 1%

71

‡

(7) FDIV

(8) FDIV

0.192
(0.150)
-0.104
(0.080)
-0.194*
(0.101)
0.022
(0.104)
-0.034
(0.138)
0.016
(0.071)
-0.029
(0.091)
-0.045
(0.072)
1,520

0.222
(0.148)
-0.131
(0.089)
-0.177*
(0.102)
0.032
(0.093)
-0.036
(0.139)
0.022
(0.070)
-0.007
(0.091)
0.040
(0.074)
1,520

I highlight two implications of these findings. First, the results indicate that for each
additional IFT certificate (per 1,000 population), industrial property values increase by 0.16
percent, but there are no residual effects. Second, the effects of tax abatements are larger in the
high tax communities than in low tax communities. Third, I find no evidence of tax abatement
competition effects even though it may be true that local governments mimic the tax abatement
policies of competitor communities within a regional competition framework.

4.5.3.3 IFT Benefit/Cost Analysis
To examine the effectiveness of tax abatements, I conduct a simple benefit/cost (BC)
analysis. The purpose of the BC analysis is to determine whether the benefits to local
governments of tax abatements exceed the costs. I focus the analysis on the costs to local
governments in that tax abatements incur costs in the form of forgone tax revenues during the
period of abatement. That is, the costs of providing tax abatements are the tax revenues that
might have been collected had the abatements not been offered. Thus, to fully assess the efficacy
of tax abatements, it is important to estimate and compare the benefits and costs. If the benefitcost ratio for offering tax abatements is larger than 1, then tax abatements are effective and
worthwhile from the perspective of local government finance. I recognize that there may other
benefits (job creation for residents, etc.) and costs (increased congestion, etc.), but consideration
of these is beyond the scope of the present analysis.
The calculations provide an estimate of what happens to industrial property values when
officials increase the use of tax abatements. The empirical findings suggest that an additional
abatement in an average community and a high tax community yields $0.105 and $0.115
increases in per capita industrial property value, respectively. The net present values of property

72

value benefits of the abatement are presented in Table 4.11. To calculate the annual increase in
industrial property value from the abatement, I assume that 1) the period of the abatement is 12
years; 2) a firm stays for 12 years after the abatement expires; and 3) the discount rate is 5%. In
addition, because the costs of offering abatements are expressed as (forgone) revenues, to offer
an appropriate comparison, using the sample average of non-homestead millage rates, I estimate
34

revenues generated by the abatement for each year. The estimated annual revenues from the
abatement for an average community and a high tax community are $0.0004 and $0.0005,
respectively (Table 4.11). For costs, I use data on abatements from the State of Michigan. In
Michigan, as a result of the IFT program 14,434 projects were abated between the fiscal years
1984-2003 and over this period, $6,746 in per capita taxable value were abated: that is, on
average, annual taxable value of property abated for each project is $0.467 per capita (Citizens
Research Council of Michigan, 2007). Using the sample average non-homestead millage rate, I
calculate average forgone revenues from the tax abatements as a measure of the costs of the
abatements to local governments (Table 4.10). The forgone revenues depend on the type of
project because industrial firms receive a tax abatement of 100% of the improvement of the
property for a replacement facility and a tax abatement of 50% of the total mills levied on a
property for a new facility. However, I do not have information on which of the projects are
categorized as a replacement facility and which are categorized as a new facility. Thus, I perform
simulations for the B/C analysis using the following scenarios:

34

Tax abatements may indirectly result in other benefits such as job creations and retentions,
thereby promoting local economic development. However, in the B/C analysis, I focus on the
direct fiscal benefits and costs of the abatements for localities because they provide a clear
picture of changes in the fiscal health of localities after the abatements are granted. Thus, to
assess the efficacy of the tax abatements, I examine how much tax revenues would be raised and
lost if another tax abatement is offered.
73

 Scenario 1: Projected taxable values of the property for a replacement facility are 100% of
total abated taxable values
 Scenario 2: Projected taxable values of the properties for a replacement facility and for a
new facility are 70% and 30% of total abated taxable values, respectively
 Scenario 3: Projected taxable values of the properties for a replacement facility and for a
new facility are 50% and 50% of total abated taxable values, respectively
 Scenario 4: Projected taxable values of the properties for a replacement facility and for a
new facility are 30% and 70% of total abated taxable values, respectively
 Scenario 5: Projected taxable values of the property for a new facility are 100% of total
abated taxable values
For each scenario, the per capita forgone revenues for each project as a result of tax abatements
are provided in Table 4.10. Using benefits and costs of tax abatements, I calculate the B/C ratios.
As shown in Table 4.14, for all scenarios, B/C ratios are always less than one, which suggests
that on average the benefits of another tax abatement is far less than the costs. Based on this
evaluation, I conclude that tax abatements are not cost-effective from a local government finance
perspective. Also, this conclusion is maintained even for the high tax communities where tax
abatements have relatively larger effects on industrial property value growth. This B/C analysis
shows that tax abatements cannot be substantiated based on an evaluation from a local
government perspective.

74

Table 4.10 Estimated Annual Forgone Revenues Resulting from Tax Abatements:
Fiscal Year 1984-2003
Estimated Annual Forgone Revenues Per Capita Per Project
Scenarios
Average Communities

High Tax Communities

Scenario 1

0.212

0.253

Scenario 2

0.180

0.215

Scenario 3

0.159

0.190

Scenario 4

0.138

0.165

Scenario 5

0.106

0.127

Note:
1. Annual average taxable value of property values abated per capita per project is $0.470
(=$4,497,855/9,561,519). Also, for average communities and high tax communities, the
sample average of non-homestead millage rates is 50.88 and 60.77, respectively.
2. To calculate the forgone revenues resulting from tax abatements, I use information on the
number of projects and the taxable values of property abated from Citizens Research Council
of Michigan (2007)
3. I assume that the period of tax abatements is 12 years.

75

Table 4.11 Estimated Generated Revenues Resulting from Tax Abatements: Industrial Property

Average
Pop.

NonHomestead
Millage
Rates

(1)

Comm.

Average
Property
Values
per
Capita

(2)

(3)

The Number
of Tax
Abatements
Granted per
1,000
1
Population

Coefficients
on Tax
2
Abatements

(4)

(5)

=1/[(2)/1,000]
Average
Comm.

1,838

27,980

50.88

0.036

0.160

Property
Value
Benefits
Per Capita
Per
3
Project

Annual Industrial
Property Value
Benefits
Per Capita Per
4
Project

ܶ
1
(1)×(4)×(5)
(6)ൗ෍
×0.01
‫)50.1( 1 =ݐ‬௧
(6)

0.150

(7)

0.0076

Estimated
Generated
Annual
Revenues
Per Capita
Per
Project
(8)
[(7)×(3)
/1,000]
0.0004

High
0.319
Tax
588
16,353
60.77
0.061
0.115
0.0083
0.0005
=(0.133+0.186)
Comm.
Note:
1. I assume that one additional tax abatement is granted in an average community and a high tax community.
2. I use the coefficients on tax abatement variables in column 2, Table 4.8 and in column 6, Table 4.9.
3. Net present value of industrial property value benefits per capita per project
4. I assume that the discount rate is 5%. Also, I consider the case that the period of tax abatements is 12 years and firms stay for 12
years after abatements expire (T=24). The formula of net present value in column (6) is: NPV (0.05, 24 years) =
24
(7)
෍
.
‫)50.1( 1 =ݐ‬௧

76

Table 4.12 Estimated Generated Revenues Resulting from Tax Abatements: Residential Property

Average
Pop.

Homestead
Millage
Rates

(1)

Comm.

Average
Property
Values
per
Capita

The Number
of Tax
Abatements
Granted per
1,000
1
Population

(2)

(3)

(4)

Coefficients
on Tax
2
Abatements

(5)

=1/[(2)/1,000]
Average
Comm.

17,944

27,980

43.92

0.036

0.130
=(0.082+0.048)

Property
Value
Benefits
Per Capita
Per
3
Project

Annual Residential
Property Value
Benefits
Per Capita Per
4
Project

ܶ
1
(1)×(4)×(5)
(6)ൗ෍
×0.01
‫)50.1( 1 =ݐ‬௧
(6)

0.835

(7)

0.061

Estimated
Generated
Annual
Revenues
Per Capita
Per
Project
(8)
[(7)×(3)
/1,000]
0.0027

Note:
1. I assume that one additional tax abatement is granted in an average community.
2. I use the coefficients on tax abatement variables in column 2, Table 4.4.
3. Net present value of residential property value benefits per capita per project
4. I assume that the discount rate is 5%. Also, I consider the case that the period of tax abatements is 12 years and firms stay for 12
years after abatements expire (T=24). The formula of net present value in column (6) is: NPV (0.05, 24 years) =
24
(7)
෍
.
‫)50.1( 1 =ݐ‬௧

77

Table 4.13 Estimated Generated Revenues Resulting from Tax Abatements: Commercial Property

Average
Pop.

NonHomestead
Millage
Rates

(1)

Comm.

Average
Property
Values
per
Capita

(2)

(3)

The Number
of Tax
Abatements
Granted per
1,000
1
Population

Coefficients
on Tax
2
Abatements

(4)

(5)

=1/[(2)/1,000]
Average
Comm.

2,869

27,980

60.77

0.036

0.039
=(0.074-0.036)

Property
Value
Benefits
Per Capita
Per
3
Project

Annual Commercial
Property Value
Benefits
Per Capita Per
4
Project

ܶ
1
(1)×(4)×(5)
(6)ൗ෍
×0.01
‫)50.1( 1 =ݐ‬௧
(6)

0.040

(7)

0.003

Estimated
Generated
Annual
Revenues
Per Capita
Per
Project
(8)
[(7)×(3)
/1,000]
0.0002

Note:
1. I assume that one additional tax abatement is granted in an average community.
2. I use the coefficients on tax abatement variables in column 2, Table 4.6.
3. Net present value of commercial property value benefits per capita per project
4. I assume that the discount rate is 5%. Also, I consider the case that the period of tax abatements is 12 years and firms stay for 12
years after abatements expire (T=24). The formula of net present value in column (6) is: NPV (0.05, 24 years) =
24
(7)
෍
.
‫)50.1( 1 =ݐ‬௧

78

Table 4.14 B/C Ratios of Tax Abatements I
Average Community

High Tax Community

Scenarios
Costs

1

Benefits

2

B/C Ratio

Costs

1

2

Benefits

B/C Ratio

Scenario 1

0.212

0.0093

0.044

0.253

0.0121

0.048

Scenario 2

0.180

0.0093

0.052

0.215

0.0121

0.056

Scenario 3

0.159

0.0093

0.059

0.190

0.0121

0.064

Scenario 4

0.138

0.0093

0.068

0.165

0.0121

0.074

Scenario 5

0.106

0.0093

0.088

0.127

0.0121

0.096

Note:
1. The costs of tax abatements are taken from Table 4.10.
2. For benefits of tax abatements, I assume that the total length of tenure of firms after the tax
abatement is granted is 24 years: that is, the estimated annual revenues from abatements per
capita per project can be calculated by: [column (8), Table 4.11] × 24 years.

4.5.3.4 Spillover Effects of Tax Abatements
In previous sections (4.5.1) and (4.5.2), I find evidence that there are positive spillover
effects of tax abatements on residential and commercial property value growth. Thus, I
recalculate the B/C ratios considering positive spillover effects of tax abatements on residential
35

and commercial property values . As shown in Table 4.15, for all scenarios, B/C ratios are still
less than one, suggesting that tax abatements are costly and inefficient at least in Southeast
Michigan.

35

For residential and commercial property, the estimated generated revenues resulting from tax
abatements are presented in Tables 4.12 and 4.13.
79

Table 4.15 B/C Ratios of Tax Abatements II
Average Community
Scenarios
Costs

1

Benefits

2

B/C Ratio

Scenario 1

0.212

0.077

0.364

Scenario 2

0.180

0.077

0.429

Scenario 3

0.159

0.077

0.486

Scenario 4

0.138

0.077

0.560

Scenario 5

0.106

0.077

0.729

Note:
1. The costs of tax abatements are taken from Table 4.10.
2. For benefits of tax abatements, I assume that the total length of tenure of firms after the tax
abatement is granted is 24 years: that is, the estimated annual revenues from abatements per
capita per project can be calculated by: [column (8), Tables 4.11] × 24 years + [column (8),
Tables 4.12] × 24 years + [column (8), Tables 4.13] × 24 years. I note that in the benefits
here, I include spillover benefits of the tax abatements to residential and commercial
properties.

80

4.6. Robustness Evaluation
To conduct robustness checks, I first consider the case that tax abatements are
endogenous. Tables 4.16-4.21 provide the regression results from specifications treating the tax
abatement variable (∆ G_IFT) as endogenous. I use the city mayor and township supervisor
election cycles as an instrument to address potential endogeneity. Tables 4.16-4.21 correspond to
Tables 4.3-4.8 and present the results that treat tax abatements as endogenously determined. As
shown in Tables 4.16-4.21, I find no evidence that the own tax abatement variable (∆ G_IFT)
have significant positive impacts on residential, commercial, and industrial property value
growth. The lack of statistical significance may be due to the fact that the instrument I use for tax
abatements is weak. In the future study, it will be useful to consider adding other valid
instruments to improve identification. For other tax abatement variables, the results are
consistent. Finally as shown in column 2 of Tables 4.17, 4.19, and 4.21, I report consistent
findings for property tax rates and school spending: 1) Property tax rates affect the own
residential, commercial, and industrial property value growth; 2) school spending is a important
factor only for residential property value growth and 3) in the regional competition context, it is
important for local communities to maintain a competitive tax position for industrial property
value growth.

81

Table 4.16 First Difference Results – Robust Check I-1
Dependent Variable: ∆ Ln Residential Property Values Per Capita
Specification

‡

(1) FDIV

(2) FDIV

(3) FDIV

(4) FDIV

Include Comp. Variables

No

Yes

Yes

Yes

The Types of Competitors
The Own Variables

–

Migration

Distance

Population

0.021
(0.145)
0.049**
(0.021)
-0.079***
(0.020)
-0.009
(0.078)
0.002
(0.002)

-0.193
(0.150)
0.073**
(0.029)
-0.281***
(0.043)
0.881***
(0.198)
0.005
(0.007)

-0.061
(0.121)
0.056***
(0.020)
-0.124***
(0.021)
-0.005
(0.099)
0.001
(0.002)

0.018
(0.139)
0.045**
(0.020)
-0.060*
(0.035)
0.029
(0.136)
0.0002
(0.002)

1,976

0.140***
(0.054)
0.443***
(0.141)
0.257***
(0.049)
-0.882***
(0.206)
-0.004
(0.006)
1,976

0.007
(0.033)
-0.027
(0.038)
0.107***
(0.033)
0.023
(0.125)
0.001
(0.001)
1,976

-0.030
(0.037)
-0.086**
(0.036)
-0.017
(0.037)
-0.043
(0.114)
0.002
(0.001)
1,976

∆ G_IFT
∆ E_IFT
∆ Ln Property Tax
∆ Ln School Spending
∆ Ln Crime rate
The Competitor Variables
∆ CG_IFT
∆ CE_IFT
∆ Ln Property Tax
∆ Ln School Spending
∆ Ln Crime rate
N

Notes:
1. All regressions include a series of time indicator and community-specific time trend variables.
2. Robust cluster standard errors in parentheses
3. “Property Tax” indicates the residential property tax.
4. Spatial autocorrelation is taken into consideration in the specifications with the symbol (‡) in
the first row.
5. I treat the tax abatement variable (G_IFT) as endogenous.
6. * significant at 10%; ** significant at 5%; *** significant at 1%

82

Table 4.17 First Difference Results – Robust Check I-1
Dependent Variable: ∆ Ln Residential Property Values Per Capita
‡
Specification
(1) FDIV
(4) FDIV
(2) FDIV
(3) FDIV
Include Comp. Variables
No
Yes
Yes
Yes
The Types of Competitors
–
Distance
Population
Migration
The Own Variables
0.233
0.046
-0.015
0.090
∆ G_IFT
(0.209)
(0.102)
(0.244)
(0.134)
0.100
0.036
-0.534*
-0.088
∆ G_IFT t-1
(0.201)
(0.149)
(0.293)
(0.128)
0.193
0.195
0.234
-0.295**
∆ G_IFT t-2
(0.278)
(0.184)
(0.296)
(0.146)
0.455*
0.073
0.201
-0.105
∆ G_IFT t-3
(0.265)
(0.131)
(0.315)
(0.126)
-0.004
0.040
0.030
0.060**
∆ E_IFT
(0.050)
(0.026)
(0.053)
(0.028)
-0.035
-0.022
0.098*
0.007
∆ E_IFT t-1
(0.049)
(0.033)
(0.059)
(0.025)
-0.025
-0.032
-0.055
0.024
∆ E_IFT t-2
(0.055)
(0.035)
(0.048)
(0.034)
-0.028
0.004
0.022
0.047
∆ E_IFT t-3
(0.050)
(0.021)
(0.048)
(0.036)
-0.082***
-0.129***
-0.025
-0.296***
∆ Ln Property Tax
(0.024)
(0.024)
(0.072)
(0.067)
-0.042**
-0.070***
0.037
-0.164***
∆ Ln Property Tax t-1
(0.017)
(0.016)
(0.072)
(0.063)
-0.022
-0.054***
0.010
-0.125**
∆ Ln Property Tax t-2
(0.015)
(0.021)
(0.057)
(0.060)
-0.018
0.0002
0.016
0.106*
∆ Ln Property Tax t-3
(0.023)
(0.016)
(0.066)
(0.062)
-0.030
-0.137
-0.056
1.165***
∆ Ln School Spending
(0.093)
(0.129)
(0.302)
(0.293)
0.075
0.100
0.120
0.644**
∆ Ln School Spending t-1
(0.068)
(0.082)
(0.333)
(0.325)
0.053
0.013
-0.246
0.274
∆ Ln School Spending t-2
(0.056)
(0.079)
(0.282)
(0.339)
0.119
0.228***
-0.583**
0.189
∆ Ln School Spending t-3
(0.080)
(0.069)
(0.262)
(0.304)
0.002
0.002
0.0004
0.012
∆ Ln Crime rate
(0.002)
(0.003)
(0.005)
(0.009)

83

Table 4.17 (cont’d)
Specification
The Competitor Variables

(1) FDIV

(2) FDIV

(3) FDIV

(4) FDIV

1,520

0.142**
(0.071)
0.079
(0.076)
-0.063
(0.073)
-0.006
(0.064)
0.479**
(0.194)
0.272
(0.217)
-0.257
(0.236)
0.393*
(0.212)
0.251***
(0.074)
0.141**
(0.064)
0.116*
(0.062)
-0.089
(0.069)
-1.231***
(0.279)
-0.613**
(0.309)
-0.241
(0.329)
-0.054
(0.299)
-0.010
(0.008)
1,520

0.056
(0.047)
-0.007
(0.042)
0.016
(0.035)
0.049
(0.046)
-0.046
(0.048)
-0.022
(0.035)
-0.056
(0.035)
0.040
(0.034)
0.107***
(0.034)
0.072**
(0.033)
0.049
(0.034)
-0.005
(0.034)
0.147
(0.138)
-0.025
(0.105)
0.052
(0.119)
-0.242**
(0.104)
-0.0001
(0.002)
1,520

0.054
(0.108)
0.066
(0.091)
-0.197**
(0.100)
0.099
(0.091)
-0.012
(0.080)
-0.114
(0.099)
0.022
(0.105)
0.038
(0.086)
-0.037
(0.064)
-0.039
(0.063)
-0.065
(0.056)
-0.025
(0.058)
0.158
(0.286)
-0.024
(0.306)
0.248
(0.246)
0.524**
(0.239)
0.008**
(0.004)
1,520

∆ CG_IFT
∆ CG_IFT t-1
∆ CG_IFT t-2
∆ CG_IFT t-3
∆ CE_IFT
∆ CE_IFT t-1
∆ CE_IFT t-2
∆ CE_IFT t-3
∆ Ln Property Tax
∆ Ln Property Tax t-1
∆ Ln Property Tax t-2
∆ Ln Property Tax t-3
∆ Ln School Spending
∆ Ln School Spending t-1
∆ Ln School Spending t-2
∆ Ln School Spending t-3
∆ Ln Crime rate
N

Notes:
1. All regressions include a series of time indicator and individual-specific time trend variables.
2. Robust cluster standard errors in parentheses
3. “Property Tax” indicates the residential property tax.
4. Spatial autocorrelation is taken into consideration in the specifications with the symbol (‡) in
the first row.
5. I treat tax abatement variables (G_IFT, G_IFT t-1, G_IFT t-2, and G_IFT t-3) as endogenous.
6. * significant at 10%; ** significant at 5%; *** significant at 1%
84

Table 4.18 First Difference Results II-1
Dependent Variable: ∆ Ln Commercial Property Values Per Capita
Specification

‡

‡

(1) FDIV

(2) FDIV

(3) FDIV

(4) FDIV

Include Comp. Variables

No

Yes

Yes

Yes

The Types of Competitors
The Own Variables

–

Migration

Distance

Population

-0.175
(0.238)
0.042
(0.040)
-0.016
(0.052)
0.100
(0.108)
0.006**
(0.003)

-0.079
(0.231)
0.029
(0.036)
-0.259
(0.160)
0.334
(0.387)
0.021
(0.015)

-0.109
(0.232)
0.032
(0.037)
-0.102
(0.087)
0.008
(0.164)
0.003
(0.004)

-0.066
(0.229)
0.030
(0.036)
0.050
(0.106)
-0.034
(0.281)
-0.001
(0.004)

1,976

0.142
(0.111)
-0.032
(0.185)
0.277*
(0.155)
-0.193
(0.356)
-0.015
(0.014)
1,976

0.039
(0.069)
-0.005
(0.047)
0.148
(0.095)
0.129
(0.198)
0.003
(0.003)
1,976

-0.005
(0.081)
-0.020
(0.067)
-0.071
(0.094)
0.106
(0.255)
0.008***
(0.003)
1,976

∆ G_IFT
∆ E_IFT
∆ Ln Property Tax
∆ Ln School Spending
∆ Ln Crime rate
The Competitor Variables
∆ CG_IFT
∆ CE_IFT
∆ Ln Property Tax
∆ Ln School Spending
∆ Ln Crime rate
N

Notes:
1. All regressions include a series of time indicator and community-specific time trend variables.
2. Robust cluster standard errors in parentheses
3. “Property Tax” indicates the non-residential property tax.
4. Spatial autocorrelation is taken into consideration in the specifications with the symbol (‡) in
the first row.
5. I treat the tax abatement variable (G_IFT) as endogenous.
6. * significant at 10%; ** significant at 5%; *** significant at 1%

85

Table 4.19 First Difference Results II-2
Dependent Variable: ∆ Ln Commercial Property Values Per Capita
‡
Specification
(1) FDIV
(4) FDIV
(2) FDIV
(3) FDIV
Include Comp. Variables
No
Yes
Yes
Yes
The Types of Competitors
–
Distance
Population
Migration
The Own Variables
-0.194
-0.027
-0.080
-0.002
∆ G_IFT
(0.218)
(0.203)
(0.226)
(0.177)
-0.349
-0.299
-0.416
-0.288
∆ G_IFT t-1
(0.268)
(0.212)
(0.255)
(0.206)
0.021
0.112
0.110
-0.163
∆ G_IFT t-2
(0.372)
(0.302)
(0.305)
(0.250)
0.048
0.074
0.001
0.055
∆ G_IFT t-3
(0.280)
(0.242)
(0.265)
(0.247)
0.062
0.034
0.044
0.027
∆ E_IFT
(0.053)
(0.043)
(0.050)
(0.042)
0.079
0.056
0.086*
0.070
∆ E_IFT t-1
(0.052)
(0.044)
(0.051)
(0.052)
-0.020
-0.036
-0.042
-0.009
∆ E_IFT t-2
(0.055)
(0.046)
(0.046)
(0.042)
0.031
0.031
0.041
0.018
∆ E_IFT t-3
(0.046)
(0.041)
(0.046)
(0.044)
0.003
-0.039
0.039
-0.506***
∆ Ln Property Tax
(0.052)
(0.095)
(0.123)
(0.166)
0.008
-0.046
0.032
-0.544**
∆ Ln Property Tax t-1
(0.046)
(0.071)
(0.119)
(0.238)
0.033
0.086
0.016
-0.818***
∆ Ln Property Tax t-2
(0.052)
(0.083)
(0.084)
(0.223)
0.049
0.085
0.106
-0.128
∆ Ln Property Tax t-3
(0.057)
(0.085)
(0.104)
(0.243)
0.079
-0.036
-0.035
0.720
∆ Ln School Spending
(0.118)
(0.189)
(0.306)
(0.458)
0.031
0.080
0.134
0.326
∆ Ln School Spending t-1
(0.107)
(0.162)
(0.334)
(0.463)
-0.029
0.089
-0.232
0.454
∆ Ln School Spending t-2
(0.110)
(0.163)
(0.271)
(0.438)
-0.010
0.129
-0.334
-0.563
∆ Ln School Spending t-3
(0.099)
(0.148)
(0.259)
(0.366)
0.008*
0.006
-0.001
0.020
∆ Ln Crime rate
(0.004)
(0.005)
(0.004)
(0.018)

86

Table 4.19 (cont’d)
Specification
The Competitor Variables

(1) FDIV

(2) FDIV

(3) FDIV

(4) FDIV

1,520

0.136
(0.146)
-0.119
(0.129)
-0.059
(0.120)
-0.123
(0.123)
-0.325
(0.312)
0.643*
(0.372)
0.162
(0.337)
-0.493
(0.424)
0.541***
(0.158)
0.553**
(0.229)
0.875***
(0.222)
0.181
(0.243)
-0.544
(0.456)
-0.270
(0.467)
-0.353
(0.433)
0.633*
(0.384)
-0.011
(0.018)
1,520

0.069
(0.095)
-0.137
(0.097)
0.030
(0.082)
-0.082
(0.081)
0.034
(0.064)
-0.041
(0.067)
0.092
(0.071)
0.082
(0.103)
0.083
(0.109)
0.109
(0.086)
-0.097
(0.092)
-0.069
(0.107)
0.089
(0.239)
0.012
(0.225)
-0.104
(0.224)
-0.267
(0.268)
0.002
(0.003)
1,520

0.058
(0.101)
0.062
(0.083)
-0.195**
(0.095)
0.099
(0.086)
-0.013
(0.080)
-0.093
(0.088)
-0.026
(0.091)
0.019
(0.079)
-0.034
(0.114)
-0.009
(0.117)
0.013
(0.084)
-0.067
(0.088)
0.110
(0.281)
-0.052
(0.310)
0.214
(0.222)
0.314
(0.233)
0.010***
(0.003)
1,520

∆ CG_IFT
∆ CG_IFT t-1
∆ CG_IFT t-2
∆ CG_IFT t-3
∆ CE_IFT
∆ CE_IFT t-1
∆ CE_IFT t-2
∆ CE_IFT t-3
∆ Ln Property Tax
∆ Ln Property Tax t-1
∆ Ln Property Tax t-2
∆ Ln Property Tax t-3
∆ Ln School Spending
∆ Ln School Spending t-1
∆ Ln School Spending t-2
∆ Ln School Spending t-3
∆ Ln Crime rate
N

Notes:
1. All regressions include a series of time indicator and community-specific time trend variables.
2. Robust cluster standard errors in parentheses
3. “Property Tax” indicates the non-residential property tax.
4. Spatial autocorrelation is taken into consideration in the specifications with the symbol (‡) in
the first row.
5. I treat tax abatement variables (G_IFT, G_IFT t-1, G_IFT t-2, and G_IFT t-3) as endogenous.
6. * significant at 10%; ** significant at 5%; *** significant at 1%
87

Table 4.20 First Difference Results – Robust Check III-1
Dependent Variable: ∆ Ln Industrial Property Values
Specification

‡

(1) FDIV

(2) FDIV

(3) FDIV

(4) FDIV

Include Comp. Variables

No

Yes

Yes

Yes

The Types of Competitors
The Own Variables

–

Migration

Distance

Population

0.158
(0.557)
0.110
(0.083)
-0.019
(0.108)
0.255
(0.479)
0.004
(0.005)

0.371
(0.561)
0.090
(0.085)
-0.340
(0.340)
1.213*
(0.687)
0.008
(0.024)

0.121
(0.529)
0.114
(0.082)
0.012
(0.171)
0.255
(0.492)
0.003
(0.006)

0.322
(0.583)
0.101
(0.087)
-0.304
(0.254)
-0.888
(0.805)
-0.005
(0.007)

1,976

0.129
(0.183)
-0.447
(0.322)
0.318
(0.295)
-1.025
(0.925)
-0.006
(0.023)
1,976

0.010
(0.213)
-0.061
(0.082)
-0.044
(0.168)
0.037
(0.306)
0.001
(0.005)
1,976

0.113
(0.135)
-0.088
(0.134)
0.299
(0.216)
1.069*
(0.589)
0.009
(0.006)
1,976

∆ G_IFT
∆ E_IFT
∆ Ln Property Tax
∆ Ln School Spending
∆ Ln Crime rate
The Competitor Variables
∆ CG_IFT
∆ CE_IFT
∆ Ln Property Tax
∆ Ln School Spending
∆ Ln Crime rate
N

Notes:
1. All regressions include a series of time indicator and community-specific time trend variables.
2. Robust cluster standard errors in parentheses
3. “Property Tax” indicates the non-residential property tax.
4. Spatial autocorrelation is taken into consideration in the specifications with the symbol (‡) in
the first row.
5. I treat the tax abatement variable (G_IFT) as endogenous.
6. * significant at 10%; ** significant at 5%; *** significant at 1%

88

Table 4.21 First Difference Results – Robust Check III-2
Dependent Variable: ∆ Ln Industrial Property Values
Specification

(1) FDIV

(2) FDIV

(3) FDIV

(4) FDIV

Include Comp. Variables

No

Yes

Yes

Yes

The Types of Competitors

–

Migration

Distance

Population

-0.028
(0.382)
-0.106
(0.450)
-0.385
(0.601)
-0.037
(0.500)
0.105
(0.087)
0.019
(0.087)
-0.008
(0.107)
0.020
(0.071)
0.039
(0.118)
0.200
(0.187)
0.003
(0.163)
0.050
(0.072)
0.242
(0.511)
-0.899
(0.659)
-0.227
(0.183)
0.093
(0.280)
0.005
(0.007)

0.082
(0.370)
-0.699
(0.558)
-0.464
(0.527)
-0.458
(0.563)
0.120
(0.112)
0.118
(0.106)
-0.007
(0.107)
0.057
(0.095)
-0.666
(0.414)
-0.903**
(0.444)
-0.762
(0.607)
-0.349
(0.643)
0.530
(1.029)
0.721
(0.989)
0.030
(0.778)
-0.019
(0.640)
0.009
(0.032)

-0.167
(0.438)
-0.056
(0.500)
-0.914
(0.666)
-0.436
(0.603)
0.158
(0.102)
0.022
(0.105)
0.060
(0.132)
0.077
(0.099)
0.019
(0.190)
0.457
(0.428)
-0.023
(0.152)
0.121
(0.166)
0.127
(0.556)
-1.285
(0.855)
-0.175
(0.306)
-0.077
(0.425)
0.006
(0.008)

-0.156
(0.460)
-0.474
(0.500)
-0.421
(0.533)
-0.336
(0.554)
0.140
(0.104)
0.098
(0.094)
-0.009
(0.104)
0.070
(0.094)
-0.216
(0.260)
0.029
(0.287)
0.142
(0.238)
0.136
(0.191)
-0.924
(0.922)
-1.502
(1.421)
0.702
(1.074)
1.275**
(0.566)
-0.003
(0.007)

The Own Variables
∆ G_IFT
∆ G_IFT t-1
∆ G_IFT t-2
∆ G_IFT t-3
∆ E_IFT
∆ E_IFT t-1
∆ E_IFT t-2
∆ E_IFT t-3
∆ Ln Property Tax
∆ Ln Property Taxt-1
∆ Ln Property Taxt-2
∆ Ln Property Taxt-3
∆ Ln School Spending
∆ Ln School Spending t-1
∆ Ln School Spending t-2
∆ Ln School Spending t-3
∆ Ln Crime rate

89

Table 4.21 (cont’d)
Specification
The Competitor Variables

(1) FDIV

(2) FDIV

(3) FDIV

(4) FDIV

1,520

-0.027
(0.253)
-0.283
(0.295)
0.061
(0.316)
-0.031
(0.220)
-0.415
(0.564)
0.802
(0.708)
0.218
(0.707)
-0.756
(0.908)
0.716*
(0.388)
1.047**
(0.482)
0.771
(0.613)
0.411
(0.650)
-0.251
(1.132)
-1.697
(1.533)
-0.204
(0.845)
0.179
(0.755)
-0.004
(0.030)
1,520

-0.193
(0.272)
0.108
(0.172)
-0.210
(0.137)
0.044
(0.212)
0.036
(0.131)
-0.228
(0.161)
-0.010
(0.133)
0.169
(0.144)
0.013
(0.198)
-0.405
(0.405)
0.052
(0.199)
-0.119
(0.219)
0.123
(0.381)
0.738
(0.641)
-0.155
(0.495)
0.247
(0.454)
0.001
(0.007)
1,520

0.271
(0.169)
0.077
(0.234)
0.101
(0.167)
0.313*
(0.162)
-0.094
(0.149)
-0.251
(0.222)
-0.140
(0.221)
0.427*
(0.249)
0.291
(0.227)
0.146
(0.188)
-0.166
(0.284)
-0.094
(0.184)
1.128
(0.691)
0.604
(1.035)
-0.999
(1.002)
-1.127**
(0.525)
0.013*
(0.007)
1,520

∆ CG_IFT
∆ CG_IFT t-1
∆ CG_IFT t-2
∆ CG_IFT t-3
∆ CE_IFT
∆ CE_IFT t-1
∆ CE_IFT t-2
∆ CE_IFT t-3
∆ Ln Property Tax
∆ Ln Property Tax t-1
∆ Ln Property Tax t-2
∆ Ln Property Tax t-3
∆ Ln School Spending
∆ Ln School Spending t-1
∆ Ln School Spending t-2
∆ Ln School Spending t-3
∆ Ln Crime rate
N

Notes:
1. All regressions include a series of time indicator and community-specific time trend variables.
2. Robust cluster standard errors in parentheses
3. “Property Tax” indicates the residential property tax.
4. I fail to find spatial autocorrelation for all specifications.
5. I treat tax abatement variables (G_IFT, G_IFT t-1, G_IFT t-2, and G_IFT t-3) as endogenous.
6. * significant at 10%; ** significant at 5%; *** significant at 1%
90

Further, in addition to treating tax abatements as endogenous, I explore several
alternative approaches and specifications. First, migration and population competitor weights in
the analysis are based on the data from the 2000 Census. Because the post-reform values may
reflect migration flows in response to school finance reform, for robustness evaluation, I use the
values before the imposition of Proposal A to define competitor communities; that is, I use the
data on migration patterns that was obtained from the 1990 Census and an “average” population
of each community over the years before the implementation of Proposal A (1983-1993). Second,
to explore finding the appropriate definition of “true” competitors/neighbors, additionally, I
employ information on the composition of race and income per capita (in 1985 dollars) to define
Race competitors: community ݆is a competitor of ݅if it is one of the four closest in the

competitors. Race and income competitors are defined as:




ratio of black/African American to community ݅
.

Income competitors: community ݆is a competitor of ݅if it is one of the four closest in
income per capita to community ݅
.

Next, the population and crime variables may be endogenous to the fiscal changes brought on by
Proposal A, which is important for my identification strategy because potential movers may
choose where to live based on physical amenities and fiscal amenities such as crime, pollution,
and tax/spending. Furthermore, it is possible that population and crime are also highly correlated
with residential and business property values (Brett and Pinkse, 2000). In the case, I need
instruments to control for the potential endogeneity of the population and crime rate variables.
Since I do not have appropriate instruments for population and crime, I use property values,
which are not normalized by population, as the dependent variable and omit the crime rate
variable as the independent variable in the specifications. Lastly, under Proposal A, poorer

91

communities received greater funding for local operating costs from the state than wealthy
communities. Proposal A was designed to reduce the gap between low and high spending
communities and the overall level of school funding in the region increased. Thus, in an effort to
find an appropriate instrument for school spending, I use changes in the per pupil school funding
(the per pupil state aid and the per pupil local revenue) resulting from Proposal A as an
alternative instrument. APPENDIX 4.D provides a more detailed description of how I generate
the instrument for changes in school spending. For reference, I report, in Tables 4.B.5 and 4.B.6,
APPENDIX 4.B, the first stage regressions, showing that the alternative instrument is a
statistically significant predictor. However, the F-statistics for excluded instruments for two of
the four endogenous variables, school spending and competitor school spending (CS), are
between four and six (columns 3 and 5 on both Tables 4.B.5 and 4.B.6). To further examine the
weak instrument problem in the presence of non-i.i.d errors, as Baum, et al. (2007) suggested, I
use a Kleibergen-Paap (KP) Wald rk F statistic for the weak identification (ID) test.

36

The KP

Wald rk F statistics from the first-stage regressions for the residential and business
(commercial/industrial) property value equations are 1.07 and 1.4, respectively. The critical
values for Stock-Yogo weak ID test provided by Stock and Yogo (2005) are not available
because in these specifications because I have five endogenous and nine instrumental variables
(they provide critical values for up to three endogenous variables). Thus, I use the informal
threshold of 10, the “Staiger and Stock rule of thumb”, as a critical value to test the null
hypothesis that the instruments are weak (Staiger and Stock, 1997). Using the KP Wald rk F
statistics, I fail to reject the null hypothesis, indicating that the joint strength of the excluded

36

The KP Wald rk F statistic is analogous to the Cragg-Donald (CD) F statistic, but unlike the
CD F statistic, it accounts for heteroskedasticity, autocorrelation, or clustering (Baum, et al.,
2007).
92

instruments is weak. Therefore, the robust estimates may be biased by weak instruments.
Regardless, I think that this robust evaluation advances understanding of regional competition.
As shown in Tables 4.5-4.8, the robustness check results are consistent with the
traditional instrumental variable results presented in the core analysis, except for industrial
property value effects of the tax rates,
somewhat smaller.

38

37

though the magnitudes of the estimated effects are

Also, intra-regional migration patterns appear to be a better approach for

identifying competitors, at least in the case of Southeast Michigan.

37

In this robustness check result, I fail to find a statistically significant effect of the tax rates on
industrial property value growth.
38

To calculate the long run effects of property taxes and school spending on residential property
value growth in column 2, Table 4.23, I use the Stata LINCOM command computing the
aggregate coefficients: 1) the aggregate coefficient on current-, one-, and two-year lagged
property taxes is – 0.347, which is statistically significant although the coefficient on one year
lagged property taxes is not significant and 2) the aggregate coefficient on one- and three-year
lagged school spending is 0.683, which is statistically significant. For the long run effects of
school spending, I do not include the coefficient on the two-year lagged school spending, which
is marginally significant, because the aggregate coefficient on current-, one-, and two year
lagged school spending is not different from zero. Also, using a same technique, I calculate the
long run effects of property taxes on commercial property value growth in column 2, Tables 4.25:
– 0.438.
93

Table 4.22 First Difference Results – Robust Check IV-1: Dependent Variable - ∆ Log Residential Property Values
‡
‡
Specification
(1) FDIV
(2) FDIV
(4) FDIV
(5) FDIV
(3) FDIV
(6) FDIV
Include Competitor Variables
No
Yes
Yes
Yes
Yes
Yes
The Types of Competitors
–
Migration
Distance
Population
Income
Race
The Own Variables
-0.178
-0.191
-0.186
-0.257**
-0.175
-0.093
∆ G_IFT
(0.141)
(0.118)
(0.128)
(0.123)
(0.147)
(0.137)
0.007
0.006
0.007
0.011
0.006
-0.007
∆ E_IFT
(0.022)
(0.020)
(0.021)
(0.021)
(0.022)
(0.022)
-0.105***
-0.274***
-0.174***
-0.091**
-0.111***
-0.167***
∆ Ln Property Tax
(0.019)
(0.046)
(0.023)
(0.036)
(0.034)
(0.051)
0.032
0.495
0.052
0.146
0.108
-0.575
∆ Ln School Spending
(0.164)
(0.302)
(0.133)
(0.299)
(0.409)
(0.384)
The Competitor Variables
0.012
-0.011
-0.024
0.004
-0.022
∆ CG_IFT
(0.045)
(0.024)
(0.041)
(0.038)
(0.040)
0.365***
0.008
-0.089**
-0.001
-0.002
∆ CE_IFT
(0.110)
(0.024)
(0.038)
(0.002)
(0.002)
0.191***
0.129***
-0.008
0.007
0.058
∆ Ln Property Tax
(0.050)
(0.033)
(0.035)
(0.033)
(0.050)
-0.404
-0.315
-0.068
-0.073
0.588*
∆ Ln School Spending
(0.285)
(0.256)
(0.296)
(0.333)
(0.322)
Notes:
1. All regressions include a series of time indicator and community-specific time trend variables.
2. Robust cluster standard errors in parentheses
3. “Property Tax” indicates the residential property tax.
4. Spatial autocorrelation is taken into consideration in the specifications with the symbol (‡) in the first row.
5. I treat tax abatement variables (G_IFT) as endogenous.
6. * significant at 10%; ** significant at 5%; *** significant at 1%

94

Table 4.23 First Difference Results – Robust Check IV-2
Dependent Variable: ∆ Ln Residential Property Values
Specification

(1) FDIV

Include Comp.
No
Variables
The Types of
–
Competitors
The Own Variables
0.025
∆ G_IFT
(0.145)
-0.101
∆ G_IFT t-1
(0.156)
0.060
∆ G_IFT t-2
(0.229)
0.344
∆ G_IFT t-3
(0.219)
-0.035
∆ E_IFT
(0.035)
-0.006
∆ E_IFT t-1
(0.034)
-0.005
∆ E_IFT t-2
(0.042)
-0.012
∆ E_IFT t-3
(0.038)
∆ Ln Property
-0.12***
Tax
(0.024)
∆ Ln Property
-0.027*
Tax t-1
(0.015)
∆ Ln Property
-0.026**
Tax t-2
(0.013)
∆ Ln Property
-0.016
Tax t-3
(0.019)
-0.024
∆ Ln School
Spending
(0.185)
∆ Ln School
0.072
Spending t-1
(0.133)
∆ Ln School
0.040
Spending t-2
(0.128)
-0.206
∆ Ln School
Spending t-3
(0.153)

‡

‡

(2) FDIV

(3) FDIV

(4) FDIV

(5) FDIV

(6) FDIV

Yes

Yes

Yes

Yes

Yes

Migr.

Dist.

Pop.

Race

Income

-0.175*
(0.092)
-0.091
(0.084)
-0.072
(0.086)
0.086
(0.078)
0.004
(0.019)
0.004
(0.020)
0.006
(0.018)
0.003
(0.021)
-0.235***
(0.050)
-0.027
(0.050)
-0.085**
(0.039)
0.002
(0.042)
0.226
(0.193)
0.274*
(0.149)
-0.087
(0.167)
0.410**
(0.206)

-0.276***
(0.102)
-0.026
(0.127)
0.065
(0.190)
-0.020
(0.124)
0.013
(0.022)
-0.004
(0.030)
-0.015
(0.030)
0.005
(0.019)
-0.178***
(0.026)
-0.056***
(0.016)
-0.069***
(0.019)
-0.026
(0.018)
-0.258*
(0.154)
0.093
(0.117)
-0.208
(0.127)
0.097
(0.107)

0.210
(0.296)
-0.540**
(0.254)
0.311
(0.272)
0.307
(0.277)
-0.055
(0.059)
0.078
(0.059)
-0.061
(0.056)
0.033
(0.048)
-0.158*
(0.085)
0.061
(0.080)
-0.003
(0.064)
-0.023
(0.073)
-0.288
(0.801)
1.096*
(0.656)
0.652
(0.619)
-1.097**
(0.486)

-0.002
(0.109)
-0.018
(0.108)
-0.004
(0.151)
0.194
(0.156)
-0.023
(0.024)
-0.010
(0.028)
-0.0003
(0.027)
0.006
(0.026)
-0.141***
(0.036)
-0.017
(0.030)
-0.035
(0.028)
-0.035
(0.032)
-0.083
(0.367)
0.279
(0.202)
-0.011
(0.227)
-0.288
(0.283)

0.039
(0.104)
-0.012
(0.104)
0.030
(0.127)
0.284*
(0.157)
-0.037
(0.028)
-0.012
(0.027)
-0.005
(0.027)
-0.005
(0.028)
-0.137**
(0.054)
-0.069
(0.046)
-0.027
(0.037)
0.005
(0.057)
-0.579
(0.381)
-1.037***
(0.399)
-0.203
(0.323)
-0.103
(0.280)

95

Table 4.23 (cont’d)
‡

‡

Specification
(1) FDIV
The Competitor Variables

(2) FDIV

(3) FDIV

(4) FDIV

(5) FDIV

(6) FDIV

∆ CG_IFT

-0.001
(0.044)
0.051
(0.038)
0.050
(0.041)
0.095**
(0.042)
0.478***
(0.173)
0.287*
(0.159)
-0.155
(0.166)
-0.173
(0.185)
0.156***
(0.057)
0.001
(0.055)
0.077*
(0.043)
-0.003
(0.050)
-0.282*
(0.158)
-0.354***
(0.130)
-0.113
(0.147)
-0.394*
(0.206)
1,520

-0.0562
(0.0367)
-0.0349
(0.0398)
-0.0217
(0.0351)
-0.0063
(0.0360)
-0.0117
(0.0340)
0.0391
(0.0322)
-0.0210
(0.0270)
0.0105
(0.0315)
0.1507***
(0.0372)
0.0653***
(0.0253)
0.0674**
(0.0343)
0.0447
(0.0379)
0.1868
(0.2064)
-0.0956
(0.2214)
0.0114
(0.2123)
-0.1995
(0.1818)
1,520

0.047
(0.121)
0.021
(0.090)
-0.147
(0.110)
0.124
(0.121)
0.061
(0.097)
-0.187*
(0.099)
0.027
(0.108)
0.077
(0.104)
0.028
(0.074)
-0.043
(0.070)
-0.064
(0.066)
-0.005
(0.065)
0.521
(0.727)
-0.856
(0.616)
-0.225
(0.525)
0.790*
(0.433)
1,520

-0.060
(0.038)
-0.009
(0.033)
0.014
(0.035)
-0.010
(0.040)
0.0004
(0.002)
-0.003
(0.002)
0.001
(0.002)
-0.002
(0.002)
0.022
(0.036)
-0.019
(0.029)
0.010
(0.027)
0.021
(0.030)
-0.066
(0.321)
-0.257
(0.176)
0.017
(0.159)
0.177
(0.220)
1,520

0.0441
(0.0568)
-0.0195
(0.0486)
0.0226
(0.0517)
0.1391*
(0.0753)
-0.0007
(0.0026)
-0.0023
(0.0021)
-0.0033
(0.0022)
-0.0019
(0.0029)
0.0154
(0.0546)
0.0294
(0.0462)
0.0017
(0.0339)
-0.0163
(0.0524)
0.5187
(0.3268)
1.0820***
(0.3635)
0.1845
(0.3066)
0.0009
(0.2436)
1,520

∆ CG_IFT t-1
∆ CG_IFT t-2
∆ CG_IFT t-3
∆ CE_IFT
∆ CE_IFT t-1
∆ CE_IFT t-2
∆ CE_IFT t-3

∆ Ln Property
Tax
∆ Ln Property
Tax t-1
∆ Ln Property
Tax t-2
∆ Ln Property
Tax t-3
∆ Ln School
Spending
∆ Ln School
Spending t-1
∆ Ln School
Spending t-2
∆ Ln School
Spending t-3
N
1,520
Notes:
1. All regressions include a series of time indicator and community-specific time trend variables.
2. Robust cluster standard errors in parentheses
3. “Property Tax” indicates the residential property tax.
4. Spatial autocorrelation is taken into consideration in the specifications with the symbol (‡) in
the first row.
5. I treat tax abatement variables (G_IFT, G_IFT t-1, G_IFT t-2, and G_IFT t-3) as endogenous.
6. * significant at 10%; ** significant at 5%; *** significant at 1%

96

Table 4.24 First Difference Results – Robust Check V-1: Dependent Variable - ∆ Log Commercial Property Values
Specification
(1) FDIV
(2) FDIV
(3) FDIV
(4) FDIV
(5) FDIV
(6) FDIV
Include Competitor Variables
No
Yes
Yes
Yes
Yes
Yes
The Types of Competitors
–
Migr.
Dist.
Pop.
Income
Race
The Own Variables
-0.352
-0.171
-0.353
-0.180
-0.290
-0.236
∆ G_IFT
(0.312)
(0.228)
(0.308)
(0.274)
(0.292)
(0.303)
-0.003
-0.024
-0.007
-0.010
-0.004
-0.011
∆ E_IFT
(0.050)
(0.035)
(0.048)
(0.045)
(0.045)
(0.045)
-0.023
-0.363**
-0.165
-0.174
-0.190**
-0.030
∆ Ln Property Tax
(0.066)
(0.183)
(0.102)
(0.155)
(0.093)
(0.151)
0.503
0.208
-0.166
-1.200
-0.112
-1.309
∆ Ln School Spending
(0.501)
(0.532)
(0.467)
(0.982)
(0.713)
(1.009)
The Competitor Variables
-0.014
0.069
0.002
0.020
-0.036
∆ CG_IFT
(0.089)
(0.063)
(0.077)
(0.078)
(0.102)
0.268
0.030
0.039
0.007**
0.0003
∆ CE_IFT
(0.188)
(0.059)
(0.077)
(0.003)
(0.003)
0.333*
0.277**
0.128
0.185*
-0.034
∆ Ln Property Tax
(0.196)
(0.120)
(0.139)
(0.110)
(0.134)
0.081
1.256*
1.617*
0.564
1.725*
∆ Ln School Spending
(0.532)
(0.730)
(0.898)
(0.549)
(0.992)
Notes:
1. All regressions include a series of time indicator and community-specific time trend variables.
2. Robust cluster standard errors in parentheses
3. “Property Tax” indicates the non-residential property tax.
4. I fail to find spatial autocorrelation for all specifications.
5. I treat tax abatement variables (G_IFT) as endogenous.
6. * significant at 10%; ** significant at 5%; *** significant at 1%

97

Table 4.25 First Difference Results – Robust Check IV-2
Dependent Variable: ∆ Ln Commercial Property Values
Specification

(1) FDIV

Include Comp.
No
Variables
The Types of
–
Competitors
The Own Variables
0.304
∆ G_IFT
(0.361)
-0.144
∆ G_IFT t-1
(0.371)
0.555
∆ G_IFT t-2
(0.529)
0.537
∆ G_IFT t-3
(0.445)
-0.091
∆ E_IFT
(0.077)
0.015
∆ E_IFT t-1
(0.075)
-0.087
∆ E_IFT t-2
(0.094)
-0.009
∆ E_IFT t-3
(0.079)
∆ Ln Property
-0.025
Tax
(0.072)
∆ Ln Property
0.043
Tax t-1
(0.053)
∆ Ln Property
0.068
Tax t-2
(0.059)
∆ Ln Property
0.011
Tax t-3
(0.076)
0.532
∆ Ln School
Spending
(0.609)
∆ Ln School
0.430
Spending t-1
(0.366)
∆ Ln School
0.609
Spending t-2
(0.437)
-0.194
∆ Ln School
Spending t-3
(0.314)

‡

(2) FDIV

(3) FDIV

(4) FDIV

(5) FDIV

(6) FDIV

Yes

Yes

Yes

Yes

Yes

Migr.

Dist.

Pop.

Race

Income

-0.076
(0.224)
-0.175
(0.207)
0.036
(0.263)
0.222
(0.216)
-0.027
(0.052)
0.048
(0.040)
-0.035
(0.044)
0.008
(0.047)
-0.438**
(0.183)
-0.284
(0.185)
-0.169
(0.194)
-0.102
(0.166)
0.758
(0.576)
0.378
(0.338)
0.547
(0.431)
0.507
(0.504)

0.661
(0.425)
0.016
(0.429)
0.631
(0.742)
0.707
(0.519)
-0.146
(0.101)
-0.041
(0.097)
-0.092
(0.129)
-0.011
(0.099)
-0.039
(0.139)
0.094
(0.103)
0.240
(0.151)
0.153
(0.159)
0.801
(0.717)
0.412
(0.569)
1.477**
(0.615)
-0.180
(0.368)

0.229
(0.300)
-0.330
(0.249)
0.344
(0.293)
0.312
(0.249)
-0.065
(0.059)
0.046
(0.051)
-0.069
(0.060)
0.029
(0.049)
-0.131
(0.147)
0.105
(0.132)
0.068
(0.114)
0.063
(0.139)
-0.513
(0.876)
1.131*
(0.685)
0.709
(0.651)
-0.549
(0.489)

0.002
(0.313)
-0.130
(0.285)
0.227
(0.381)
0.172
(0.277)
-0.030
(0.055)
0.039
(0.048)
-0.056
(0.059)
0.023
(0.046)
-0.132
(0.107)
-0.037
(0.093)
0.005
(0.101)
0.057
(0.141)
0.172
(0.826)
-0.108
(0.657)
-0.219
(0.487)
0.085
(0.589)

-0.034
(0.297)
-0.206
(0.252)
0.391
(0.355)
0.188
(0.260)
-0.029
(0.059)
0.058
(0.048)
-0.065
(0.057)
0.018
(0.047)
0.235*
(0.132)
0.118
(0.121)
0.150
(0.150)
0.162
(0.147)
0.310
(0.730)
0.002
(0.762)
0.525
(0.806)
0.658
(0.630)

98

Table 4.25 (cont’d)
‡

Specification
(1) FDIV
The Competitor Variables

(2) FDIV

(3) FDIV

(4) FDIV

(5) FDIV

(6) FDIV

∆ CG_IFT

-0.037
(0.123)
-0.069
(0.096)
0.044
(0.126)
0.120
(0.102)
0.723**
(0.367)
1.038**
(0.415)
-0.546
(0.355)
-0.088
(0.600)
0.467***
(0.173)
0.287
(0.180)
0.274
(0.188)
0.151
(0.165)
-0.256
(0.413)
-0.291
(0.292)
-0.229
(0.357)
-0.421
(0.470)
1,520

0.336**
(0.140)
-0.140
(0.129)
0.142
(0.110)
0.171
(0.144)
0.045
(0.096)
-0.125
(0.106)
0.207**
(0.096)
0.102
(0.114)
0.052
(0.145)
0.001
(0.113)
-0.277
(0.208)
-0.225
(0.230)
-0.212
(0.644)
0.438
(0.557)
-0.861
(0.704)
-0.013
(0.528)
1,520

0.073
(0.113)
0.020
(0.082)
-0.133
(0.106)
0.145
(0.120)
0.068
(0.102)
-0.165*
(0.093)
-0.020
(0.096)
0.061
(0.100)
0.069
(0.138)
-0.063
(0.127)
-0.025
(0.108)
-0.053
(0.125)
0.698
(0.775)
-0.838
(0.651)
-0.269
(0.518)
0.446
(0.452)
1,520

-0.164*
(0.092)
0.044
(0.091)
-0.039
(0.107)
0.002
(0.097)
0.010**
(0.005)
-0.001
(0.005)
0.001
(0.004)
0.007
(0.004)
0.131
(0.105)
0.091
(0.092)
0.040
(0.102)
0.016
(0.112)
0.101
(0.621)
0.618
(0.599)
0.328
(0.374)
0.135
(0.491)
1,520

0.090
(0.133)
-0.062
(0.109)
0.150
(0.133)
0.199*
(0.114)
0.003
(0.004)
0.007
(0.005)
0.003
(0.006)
-0.001
(0.006)
-0.277**
(0.117)
-0.090
(0.110)
-0.103
(0.127)
-0.113
(0.117)
0.040
(0.771)
0.357
(0.727)
-0.400
(0.679)
-0.624
(0.607)
1,520

∆ CG_IFT t-1
∆ CG_IFT t-2
∆ CG_IFT t-3
∆ CE_IFT
∆ CE_IFT t-1
∆ CE_IFT t-2
∆ CE_IFT t-3

∆ Ln Property
Tax
∆ Ln Property
Tax t-1
∆ Ln Property
Tax t-2
∆ Ln Property
Tax t-3
∆ Ln School
Spending
∆ Ln School
Spending t-1
∆ Ln School
Spending t-2
∆ Ln School
Spending t-3
N
1,520
Notes:
1. All regressions include a series of time indicator and community-specific time trend variables.
2. Robust cluster standard errors in parentheses
3. “Property Tax” indicates the non-residential property tax.
4. Spatial autocorrelation is taken into consideration in the specifications with the symbol (‡) in
the first row.
5. I treat tax abatement variables (G_IFT, G_IFT t-1, G_IFT t-2, and G_IFT t-3) as endogenous.
6. * significant at 10%; ** significant at 5%; *** significant at 1%

99

Table 4.26 First Difference Results – Robust Check VI-1: Dependent Variable - ∆ Log Industrial Property Values
Specification
(1) FDIV
(2) FDIV
(3) FDIV
(4) FDIV
(5) FDIV
(6) FDIV
Include Competitor Variables
No
Yes
Yes
Yes
Yes
Yes
The Types of Competitors
–
Migr.
Dist.
Pop.
Income
Race
The Own Variables
0.726
1.296
0.349
1.330
1.076
1.653
∆ G_IFT
(0.840)
(0.943)
(0.693)
(1.029)
(0.952)
(1.239)
-0.022
-0.078
0.019
-0.049
-0.039
-0.112
∆ E_IFT
(0.123)
(0.151)
(0.097)
(0.154)
(0.139)
(0.197)
-0.083
0.140
-0.031
-0.783*
-0.630**
-1.137**
∆ Ln Property Tax
(0.158)
(0.604)
(0.215)
(0.406)
(0.304)
(0.544)
1.094
1.251
2.064
-4.146*
-1.219
-5.849*
∆ Ln School Spending
(1.420)
(1.404)
(1.595)
(2.434)
(1.615)
(3.220)
The Competitor Variables
0.096
-0.081
0.231
0.068
-0.148
∆ CG_IFT
(0.210)
(0.215)
(0.264)
(0.235)
(0.283)
-0.703*
-0.019
0.049
0.026
0.013
∆ CE_IFT
(0.375)
(0.106)
(0.196)
(0.021)
(0.012)
-0.279
-0.075
0.633*
0.579
0.966**
∆ Ln Property Tax
(0.629)
(0.226)
(0.338)
(0.362)
(0.469)
-0.724
-1.345
4.619**
1.839
6.327**
∆ Ln School Spending
(1.340)
(1.598)
(2.107)
(1.795)
(3.033)
Notes:
1. All regressions include a series of time indicator and community-specific time trend variables.
2. Robust cluster standard errors in parentheses
3. “Property Tax” indicates the non-residential property tax.
4. I fail to find spatial autocorrelation for all specifications.
5. I treat tax abatement variables (G_IFT) as endogenous.
6. * significant at 10%; ** significant at 5%; *** significant at 1%

100

Table 4.27 First Difference Results – Robust Check VI-2
Dependent Variable: ∆ Ln Industrial Property Values
(1)
(3)
(4)
(5)
(2)
Specification
FDIV
FDIV
FDIV
FDIV
FDIV
Include Comp. Variables
No
Yes
Yes
Yes
Yes
The Types of Competitors

(6)
FDIV
Yes

–

Migr.

Dist.

Pop.

Race

Income

-4.574
(3.205)
0.717
(2.007)
-0.198
(2.486)
-1.946
(3.182)
0.430
(0.481)
0.122
(0.366)
-0.028
(0.417)
-0.165
(0.465)
0.667
(0.552)
0.115
(0.507)
-0.246
(0.388)
0.540
(0.437)
4.077
(3.551)
-8.155
(6.645)
-7.032*
(4.169)
3.720
(2.328)

-0.450
(1.085)
-0.202
(0.928)
1.708
(1.438)
1.156
(1.266)
-0.092
(0.182)
0.036
(0.223)
-0.300
(0.246)
-0.271
(0.288)
1.259
(1.166)
-1.542
(1.503)
-0.513
(0.842)
-0.210
(0.830)
0.721
(2.031)
-1.436
(2.211)
0.701
(1.436)
3.388
(2.625)

-1.049
(1.527)
0.483
(1.639)
1.510
(2.896)
1.220
(1.748)
-0.058
(0.293)
-0.111
(0.368)
-0.178
(0.462)
-0.295
(0.365)
0.362
(0.487)
1.479
(1.367)
0.067
(0.395)
0.556
(0.450)
5.750*
(3.397)
-5.112
(4.389)
-0.921
(1.942)
-0.383
(1.467)

-2.093
(1.796)
-0.742
(1.480)
0.080
(1.098)
-0.345
(1.307)
0.059
(0.281)
0.182
(0.238)
0.028
(0.203)
-0.241
(0.277)
1.192
(1.267)
1.314
(1.701)
-0.548
(0.861)
0.998
(0.711)
7.205
(6.559)
-2.805
(4.578)
-8.688
(6.846)
3.276
(2.513)

-3.744
(2.788)
0.998
(1.869)
0.663
(2.237)
-1.994
(2.585)
0.337
(0.385)
0.242
(0.391)
-0.293
(0.381)
-0.096
(0.338)
0.239
(0.899)
-0.064
(0.625)
0.320
(0.664)
0.243
(0.820)
12.628
(9.584)
-9.685
(8.773)
-2.447
(3.559)
-2.849
(4.329)

-3.356*
(1.923)
-0.038
(1.397)
-1.353
(1.585)
-2.232
(2.188)
0.387
(0.347)
0.202
(0.296)
0.092
(0.334)
-0.042
(0.366)
0.346
(0.838)
-1.786
(1.239)
-0.540
(0.876)
1.040
(0.834)
1.019
(3.993)
-8.034
(9.041)
-1.881
(4.102)
5.935
(4.296)

The Own Variables
∆ G_IFT
∆ G_IFT t-1
∆ G_IFT t-2
∆ G_IFT t-3
∆ E_IFT
∆ E_IFT t-1
∆ E_IFT t-2
∆ E_IFT t-3
∆ Ln Property Tax
∆ Ln Property Tax t-1
∆ Ln Property Tax t-2
∆ Ln Property Tax t-3
∆ Ln School Spending
∆ Ln School Spending t-1
∆ Ln School Spending t-2
∆ Ln School Spending t-3

101

Table 4.27 (cont’d)
Specification

(1)
FDIV

(2)
FDIV

(3)
FDIV

(4)
FDIV

(5)
FDIV

(6)
FDIV

1,520

0.332
(0.441)
-0.259
(0.508)
0.566
(0.539)
0.101
(0.498)
1.111
(1.738)
-0.980
(1.642)
-1.346
(1.525)
-0.077
(2.081)
-0.898
(1.010)
1.730
(1.564)
0.175
(0.770)
0.252
(0.837)
0.939
(1.567)
-4.701
(2.974)
-3.624*
(2.166)
-1.742
(2.006)
1,520

-0.107
(0.410)
0.344
(0.416)
-0.666
(0.492)
0.204
(0.405)
0.100
(0.295)
-0.503
(0.452)
0.100
(0.301)
0.010
(0.344)
-0.144
(0.505)
-1.624
(1.454)
-0.350
(0.685)
-0.654
(0.759)
-2.597
(4.058)
0.625
(2.640)
-2.832
(2.719)
1.121
(2.203)
1,520

0.775
(0.618)
0.042
(0.446)
0.288
(0.382)
0.180
(0.527)
-1.257
(0.922)
0.198
(0.608)
0.474
(0.894)
-0.048
(0.635)
-0.537
(0.955)
-1.045
(1.268)
0.320
(0.733)
-0.639
(0.559)
-2.358
(4.142)
-4.829
(4.878)
4.860
(5.126)
-1.480
(2.432)
1,520

1.589
(1.056)
0.354
(0.728)
0.256
(0.740)
0.854
(0.747)
0.007
(0.037)
-0.030
(0.042)
0.042
(0.038)
0.008
(0.030)
0.395
(1.058)
0.220
(0.778)
-0.510
(0.686)
0.210
(0.782)
-7.411
(7.678)
1.941
(5.582)
-2.495
(3.397)
5.241
(4.865)
1,520

-0.341
(0.676)
2.532
(1.635)
0.740
(0.618)
0.704
(0.950)
0.017
(0.029)
0.046
(0.048)
-0.030
(0.041)
0.037
(0.055)
0.138
(0.818)
2.071
(1.392)
0.468
(0.862)
-0.602
(0.735)
2.510
(4.452)
0.108
(7.746)
-3.226
(3.963)
-3.022
(3.846)
1,520

The Competitor Variables
∆ CG_IFT
∆ CG_IFT t-1
∆ CG_IFT t-2
∆ CG_IFT t-3
∆ CE_IFT
∆ CE_IFT t-1
∆ CE_IFT t-2
∆ CE_IFT t-3
∆ Ln Property Tax
∆ Ln Property Tax t-1
∆ Ln Property Tax t-2
∆ Ln Property Tax t-3
∆ Ln School Spending
∆ Ln School Spending t-1
∆ Ln School Spending t-2
∆ Ln School Spending t-3
N

Notes:
1. All regressions include a series of time indicator variables and individual-specific time trend
variables.
2. Robust cluster standard errors in parentheses
3. “Property Tax” indicates the non-residential property tax.
4. I fail to find spatial autocorrelation for all specifications.
5. I treat tax abatement variables (G_IFT, G_IFT t-1, G_IFT t-2, and G_IFT t-3) as endogenous.
6. * significant at 10%; ** significant at 5%; *** significant at 1%
102

4.7 Conclusion and Policy Implications
I estimate the effects of changes in tax rates and school spending on the value growth of
different types of property. Although it is true that changes in tax rates and school spending are
capitalized into property values, I show that the degree of capitalization differs across property
classes. Based on this analysis, I conclude the following:


Property taxation shows significant effects on property value growth for both residential
and business property.



Residential property values are more sensitive to school spending changes than tax policy
changes.



Commercial and industrial property values are more sensitive to tax policy changes than
school spending changes.



Commercial and industrial property values are more sensitive to changes in property
taxes relative to residential property values.



Tax abatements have a statistically significant positive effect on industrial property value
growth and this effect is relatively larger in the high tax community than in the low tax
community. However, its magnitude is relatively small as compared to the costs to local
government of offering tax abatements; in other words, tax abatements are not costeffective.



Tax abatements create positive spillover effects on residential and commercial property
value growth, but this result does not alter the argument that tax abatements are costineffective.

103



Changes in competitor community property taxes and school spending play an important
role in one’s own property value growth. That is, I present new evidence of fiscal
externalities.
The degree of this fiscal externality is much larger for commercial and industrial property

value growth, indicating that commercial and industrial property is more responsive to
competitor tax policy changes than residential property. In this study, the findings confirm the
two hypotheses I propose regarding property value effects of property tax rates and school
spending. With regard to regional competition, I find strong significant effects of competitor
property tax rates (and/or school spending) on own property value growth. Specifically, the
findings show that policy changes in neighboring communities (the competition) cause fiscal
externalities to one’s own community. Therefore, consideration of one’s relative fiscal position
vis-à-vis competitors within the region is a key issue.
This study contributes to the extensive literature on the property value effects of fiscal
policy changes. The large body of empirical literature on the effects of fiscal policy changes on
property values focuses primarily on residential property rather than commercial/industrial
property. However, it is important to consider the effects of local policies on different property
classes. I show that residents and businesses (commercial/industrial firms) have different policy
needs/preferences and thus there are tradeoffs between property taxes and school spending for
each class of property: 1) Property taxes and school spending are important factors for residential
property value growth; the long run elasticity of residential property values with respect to
property taxes and school spending is - 0.38 and 1.80, respectively; and 2) the long run property
tax elasticities of commercial and industrial property values are –1.77 and -1.73, respectively.
These results imply that if a community raises property tax rates to increase school spending,

104

then business (commercial/industrial) property value growth is likely to be slowed. On the other
hand, this policy is more likely to benefit current and potential new residents because of
increased school spending. If a community lowers property tax rates to increase business
property value growth, then school spending could be curtailed,

39

thereby inhibiting residential

property value growth. More generally, these estimates can be used to guide local policy makers
to meet their economic development objectives, whether it be greater residential property value
growth or further business development. Communities with differing proportions of residential
versus business tax bases may well come to different conclusions about their taxing and spending
balances.
With regard to the IFT abatement program, I draw two policy implications from the
analysis. First, the findings directly support the Bartik (1991, 1994) and Wassmer (2007)
argument that tax abatements may be more effective when they are used in high property tax rate
communities because industrial firms appear to avoid high tax communities. That is, in high tax
communities, tax abatements could be a useful tool to create a competitive relative position that
might generate industrial property value growth in the context of regional competition. Second,
although tax abatements can boost industrial activity, the findings show that there are substantial
costs as well, and that net tax abatements result in a net cost to the local government. This
inefficiency may stem from the lack of targeting. Almost every manufacturing firm is eligible for
property tax abatements in almost every community in Michigan and it is rare for localities to
reject abatement requests (Reese and Sands, 2012). In this case, firms, more likely to stay in a
community without tax abatements, can have an incentive to request tax abatements, thereby

39

In Michigan, although major funding sources for public schools are shifted to state revenues
by Proposal A, the funding from the state is only for school operation, not for capital investments.
Thus, property tax revenues are still an important source for public school.
105

causing inefficiency. Thus, to enhance the efficiency of the property tax abatement program in
Michigan, it is important to improve targeting to limit eligibility for abatements: for example,
developing more effective strategies for targeting manufacturing firms that are less likely to stay
without tax abatements, or limiting to high tax communities that pose potentially unfriendly
business environments. These types of limitation on the use of tax abatements may reduce the
forgone revenues of the abatements as well as help maximize their fiscal benefits.
There are caveats that should be noted. Specifically, this study has not considered
spending for other public services beyond schools. Other work has indicated positive correlations
between spending for local services such as parks and recreation and public safety and local
economic health (Reese and Ye, 2011). Thus, reductions in property taxes may also limit the
ability of local governments to provide a broader array of services important to both residential
and commercial interests. In addition, changes in tax rates and school spending may have
impacts beyond the property value effects I identify, including employment opportunities,
standard of living, and quality of life.

40

All of these factors should be considered as policy

makers ponder the tradeoffs between taxing and spending.

40

Some of these effects are arguably captured in changes in property values, however.
106

CHAPTER 5
Conclusion

This dissertation examines the effects of property taxes, school spending, and tax
abatements on residential, commercial, and industrial property value growth in Southeast
Michigan. First, I examine the effects of changes in property tax rates and school spending on
residential, commercial, and industrial property value growth in Southeast Michigan. I provide
new insight about how the effects of policy changes vary by type of property. To address the
potential endogenous relationship between policy changes and property value growth, I use the
imposition of Proposal A, which resulted in significant changes to both property tax rates and
school spending for all communities in the state. Using an instrumental variables technique, I
find that: 1) residential property values are more sensitive to school spending changes than
property tax rate changes; 2) commercial and industrial property values are more sensitive to tax
rate changes than school spending changes; and 3) commercial and industrial property values are
more sensitive to changes in tax rates than are residential properties. I also find evidence that tax
competition plays an important role in property value growth within the region.
Second, I examine the effectiveness of tax abatements in this same region surrounding
Detroit. In this analysis, I find evidence that tax abatements have a statistically significant
positive effect on industrial property value growth and this effect is larger in high than in low tax
communities. Also, my findings show that tax abatements create spillover effects on residential
and commercial property value growth. However, the benefits of tax abatements are much
smaller than the cost of offering tax abatements and this result is not changed even when I
include the spillover benefits of tax abatements in this analysis. In addition, whereas property tax
rates and competitor property tax rates are significant determinants of the own industrial property
107

value growth, I do not find evidence that competitor tax abatements influence the own industrial
property value growth.
There are several policy implications we can draw from these findings. First, my results
suggest that tax competition plays a vital role in property value growth. In this context, it is
important for a community to keep its tax-service package competitive within the region in order
not to lose property value growth. Second, the findings in this study imply that residents and
businesses have different policy preferences and thus there are tradeoffs between taxing and
spending for each class of property. Finally, these results suggest that tax abatements are costineffective in Southeast Michigan. To improve the efficiency of tax abatements, local
policymakers could improve targeting of abatements to manufacturing firms that are truly less
likely to come (or stay) without abatements. From a state perspective, it may be prudent to limit
use of abatements to high tax communities. Local policymakers do well to understand 1) the
effects of policy changes in the context of regional competition; 2) the tradeoffs in the effects of
policy changes on the value growth of different property classifications; and 3) the effectiveness
of tax abatements considering the cost of granting them.
In addition to these important policy implications, this dissertation offers three main
contributions to the literature. First, I find new evidence of differential responsiveness of
residential and business property value growth to changes in tax rates and school spending.
Second, I find a consistent result that migration flow information appears to better explain
regional competition, at least in the Southeast Michigan context. Lastly, using the spatial
weighting scheme based on migration flow information, proposed by Skidmore, et al. (2012), I
find robust evidence of significant fiscal externalities in the Southeast Michigan region.

108

APPENDICES

109

APPENDIX 4.A Variable Sources
Table 4.A.1 Variable Sources
Variables

Sources

Links

The Number of
Tax Abatements
(G_IFT and E_IFT)

Michigan Department of
Treasury

http://www.michigan.gov/taxes/
0,1607,7-238-43535_43925164513--,00.html

Residential, Commercial, and
Industrial State Equalized
Valuation (SEV)

Michigan Department of
Treasury

http://www.michigan.gov/treasu
ry/0,1607,7-1211751_2228_21957_45818--,00.html

Residential, Commercial, and
Industrial Property Taxes

Michigan Department of
Treasury

http://www.michigan.gov/taxes/
0,1607,7-238-43535_4392557815--,00.html

Uniform Crime Index (UCI)

Federal Bureau of
Investigation
U.S. Department of Justice

General Fund Expenditure per
pupil (GFEP)

Michigan Department of
Education

http://www.fbi.gov/ucr/ucr.htm
http://magic.msu.edu/record=b4
975507~S39a

http://www.michigan.gov/mde/
0,1607,7-140-6530_660521514--,00.html

http://www.census.gov/popest/d
atasets.html
Population

U.S. Census Bureau
Michigan Government

110

http://www.michigan.gov/docu
ments/MCD19601990C_33608_7.pdf

APPENDIX 4.B First Stages IV Results
Table 4.B.1 First Stage IV Results from Column 2 in Table 4.3
Specification
(1) FDIV
(2) FDIV
(3) FDIV
(4) FDIV
∆ Ln
∆ Ln
∆ Ln
∆ Ln
Comp.
Comp.
Dependent Variables
Property
School
Property
School
Tax
Spending
Tax
Spending
Own Community Variables
-0.020
-0.015
-0.024
-0.013
∆ G_IFT
(0.018)
(0.012)
(0.020)
(0.012)
0.013
0.005
0.010
0.006
∆ E_IFT
(0.012)
(0.011)
(0.013)
(0.011)
∆ Ln Changes in Property Tax
0.984***
0.009
-0.043**
0.004
due to Proposal A
(0.011)
(0.009)
(0.017)
(0.009)
0.047
0.067***
-0.007
0.048*
∆ Ln Property Tax t-5
(0.085)
(0.025)
(0.087)
(0.025)
∆ Ln Changes in School Spending
-0.120**
1.049***
-0.113
0.043
due to Proposal A
(0.053)
(0.036)
(0.073)
(0.033)
0.171**
-0.008
-0.012
-0.072
∆ Ln School Spending t-5
(0.078)
(0.048)
(0.083)
(0.050)
0.008
-0.002
0.010
-0.0004
∆ Ln Crime rate
(0.008)
(0.006)
(0.008)
(0.006)
Competitor Community Variable
-0.077***
-0.051***
-0.100***
-0.005
∆ CG_IFT
(0.027)
(0.015)
(0.032)
(0.015)
0.061
-0.051
-0.200
0.047
∆ CE_IFT
(0.126)
(0.069)
(0.148)
(0.072)
∆ Ln Changes in Property Tax due
0.027**
-0.011
1.043***
-0.004
to Proposal A
(0.012)
(0.009)
(0.018)
(0.010)
-0.032
-0.032
0.018
-0.019
∆ Ln Property Tax t-5
(0.080)
(0.036)
(0.080)
(0.036)
∆ Ln Changes in School Spending
0.142***
-0.036
0.143**
0.970***
due to Proposal A
(0.052)
(0.033)
(0.070)
(0.030)
-0.115*
0.043
0.060
0.097*
∆ Ln School Spending t-5
(0.066)
(0.051)
(0.071)
(0.053)
-0.006
-0.001
-0.007
-0.003
∆ Ln Crime rate
(0.007)
(0.006)
(0.008)
(0.006)
F-test
7,637
1,163
2,360
1,485
for excluded instruments
(P=0.000)
(P=0.000)
(P=0.000)
(P=0.000)
N
1,976
1,976
1,976
1,976
Notes:
1. All regressions include a series of time indicator and community-specific time trend variables.
2. Robust cluster standard errors in parentheses
3. Tax abatement variables are treated as exogenous.
4. “Property Tax” indicates the residential property tax.
5. * significant at 10%; ** significant at 5%; *** significant at 1%
111

Table 4.B.2 First Stage IV Results from Column 2 in Tables 4.5 and 4.7
Specification
(1) FDIV
(2) FDIV
(3) FDIV
(4) FDIV
∆ Ln
∆ Ln
∆ Ln
∆ Ln
Comp.
Comp.
Dependent Variables
Property
School
Property
School
Tax
Spending
Tax
Spending
Own Community Variables
0.010
-0.015
0.003
-0.013
∆ G_IFT
(0.011)
(0.012)
(0.011)
(0.012)
0.009
0.005
0.005
0.006
∆ E_IFT
(0.008)
(0.011)
(0.010)
(0.011)
∆ Ln Changes in Property Tax
1.012***
-0.024
-0.124***
-0.025
due to Proposal A
(0.025)
(0.024)
(0.030)
(0.024)
-0.199***
0.206***
0.170**
0.162***
∆ Ln Property Tax t-5
(0.054)
(0.051)
(0.069)
(0.055)
∆ Ln Changes in School Spending
-0.036
1.073***
-0.015
0.062*
due to Proposal A
(0.036)
(0.039)
(0.049)
(0.035)
0.124**
0.009
-0.072
-0.059
∆ Ln School Spending t-5
(0.049)
(0.048)
(0.056)
(0.050)
-0.004
-0.002
-0.003
-0.0003
∆ Ln Crime rate
(0.005)
(0.006)
(0.006)
(0.006)
Competitor Community Variable
-0.054
-0.045
-0.018
0.050
∆ CG_IFT
(0.070)
(0.070)
(0.087)
(0.073)
-0.020
-0.058***
-0.069***
-0.011
∆ CE_IFT
(0.015)
(0.016)
(0.018)
(0.015)
∆ Ln Changes in Property Tax
0.007
0.024
1.131***
0.027
due to Proposal A
(0.024)
(0.023)
(0.029)
(0.024)
0.191***
-0.165***
-0.198***
-0.131**
∆ Ln Property Tax t-5
(0.057)
(0.058)
(0.072)
(0.061)
∆ Ln Changes in School Spending
0.045
-0.057
0.026
0.953***
due to Proposal A
(0.034)
(0.036)
(0.047)
(0.033)
-0.084*
0.028
0.100*
0.085
∆ Ln School Spending t-5
(0.047)
(0.050)
(0.053)
(0.053)
0.005
-0.001
0.005
-0.003
∆ Ln Crime rate
(0.005)
(0.006)
(0.006)
(0.006)
F-test
8,314
1,013
4,883
1,310
for excluded instruments
(P=0.000)
(P=0.000)
(P=0.000)
(P=0.000)
N
1,976
1,976
1,976
1,976
Notes:
1. All regressions include a series of time indicator and community-specific time trend variables.
2. Robust cluster standard errors in parentheses
3. Tax abatement variables are treated as exogenous.
4. “Property Tax” indicates the non-residential property tax.
5. * significant at 10%; ** significant at 5%; *** significant at 1%

112

Table 4.B.3 First Stage IV Results from Column 2 in Table 4.16
Specification
(1) FDIV
(2) FDIV (3) FDIV
(4) FDIV (5) FDIV
∆ Ln
∆ Ln
∆ Ln
∆ Ln
Comp.
Comp.
Dependent Variables
Property
School
∆ G_IFT
Property
School
Tax
Spending
Tax
Spending
Own Community Variables
0.004**
-0.001
-0.002
0.001
-0.002
∆ Election Year t-1
(0.002)
(0.002)
(0.002)
(0.002)
(0.002)
0.107
0.011
0.003
0.007
0.005
∆ E_IFT
(0.079)
(0.012)
(0.011)
(0.013)
(0.011)
∆ Ln Changes in Property
-0.050
0.987***
0.014
-0.044**
0.008
Tax due to Proposal A
(0.096)
(0.012)
(0.009)
(0.018)
(0.009)
-0.096
0.048
0.067***
-0.004
0.047*
∆ Ln Property Tax t-5
(0.068)
(0.086)
(0.025)
(0.087)
(0.025)
∆ Ln Changes in School
-0.480
-0.105*
1.068***
-0.109
0.061*
Spending due to Proposal A
(0.455)
(0.061)
(0.038)
(0.077)
(0.036)
0.143
0.171**
-0.005
-0.019
-0.069
∆ Ln School Spending t-5
(0.106)
(0.079)
(0.047)
(0.083)
(0.050)
0.008
0.008
-0.002
0.010
-0.001
∆ Ln Crime rate
(0.008)
(0.008)
(0.006)
(0.008)
(0.006)
Competitor Community Variable
0.259*
0.059
-0.049
-0.210
0.050
∆ CG_IFT
(0.140)
(0.127)
(0.070)
(0.149)
(0.073)
-0.118
-0.075*** -0.049*** -0.097***
-0.004
∆ CE_IFT
(0.111)
(0.028)
(0.015)
(0.032)
(0.015)
∆ Ln Changes in Property
0.077
0.026**
-0.011
1.041***
-0.004
Tax due to Proposal A
(0.117)
(0.013)
(0.009)
(0.018)
(0.009)
0.092
-0.032
-0.030
0.014
-0.017
∆ Ln Property Tax t-5
(0.078)
(0.080)
(0.036)
(0.081)
(0.036)
∆ Ln Changes in School
0.559
0.127**
-0.053
0.135*
0.954***
Spending due to Proposal A
(0.453)
(0.058)
(0.035)
(0.073)
(0.032)
-0.024
-0.117*
0.038
0.064
0.092*
∆ Ln School Spending t-5
(0.099)
(0.067)
(0.050)
(0.071)
(0.053)
-0.005
-0.006
-0.001
-0.007
-0.002
∆ Ln Crime rate
(0.008)
(0.007)
(0.006)
(0.008)
(0.006)
F-test
1
7,379
975
2,495
1,266
for excluded instruments
(P=0.523) (P=0.000) (P=0.000) (P=0.000) (P=0.000)
N
1,976
1,976
1,976
1,976
1,976
Notes:
1. All regressions include a series of time indicator and community-specific time trend variables.
2. Robust cluster standard errors in parentheses
3. The tax abatement variable (G_IFT) is treated as endogenous.
4. “Property Tax” indicates the residential property tax.
5. * significant at 10%; ** significant at 5%; *** significant at 1%

113

Table 4.B.4 First Stage IV Results from Column 2 in Tables 4.18 and 4.20
Specification
(1) FDIV
(2) FDIV (3) FDIV
(4) FDIV (5) FDIV
∆ Ln
∆ Ln
∆ Ln
∆ Ln
Comp.
Comp.
Dependent Variables
Property
School
∆ G_IFT
Property
School
Tax
Spending
Tax
Spending
Own Community Variables
0.005**
-0.001
-0.002
0.0002
-0.002
∆ Election Year t-1
(0.003)
(0.001)
(0.002)
(0.001)
(0.002)
0.106
0.010
0.003
0.006
0.005
∆ E_IFT
(0.079)
(0.008)
(0.011)
(0.0098)
(0.011)
∆ Ln Changes in Property
-0.024
1.018***
-0.013
-0.125***
-0.014
Tax due to Proposal A
(0.218)
(0.025)
(0.026)
(0.032)
(0.026)
-0.157
-0.199*** 0.212***
0.169**
0.169***
∆ Ln Property Tax t-5
(0.124)
(0.054)
(0.051)
(0.068)
(0.055)
∆ Ln Changes in School
-0.490
-0.033
1.093***
-0.018
0.082**
Spending due to Proposal A
(0.458)
(0.038)
(0.041)
(0.050)
(0.038)
0.133
0.129***
0.013
-0.073
-0.054
∆ Ln School Spending t-5
(0.107)
(0.049)
(0.047)
(0.056)
(0.050)
0.008
-0.004
-0.002
-0.003
-0.001
∆ Ln Crime rate
(0.008)
(0.005)
(0.006)
(0.006)
(0.006)
Competitor Community Variable
-0.123
-0.021
-0.057*** -0.069***
-0.010
∆ CG_IFT
(0.123)
(0.015)
(0.015)
(0.017)
(0.015)
0.224*
-0.049
-0.043
-0.018
0.051
∆ CE_IFT
(0.130)
(0.070)
(0.071)
(0.087)
(0.074)
∆ Ln Changes in Property
0.019
0.004
0.018
1.132***
0.021
Tax due to Proposal A
(0.221)
(0.024)
(0.024)
(0.030)
(0.024)
0.158
0.190*** -0.170*** -0.197*** -0.137**
∆ Ln Property Tax t-5
(0.122)
(0.057)
(0.057)
(0.071)
(0.061)
∆ Ln Changes in School
0.545
0.046
-0.072*
0.029
0.939***
Spending due to Proposal A
(0.441)
(0.035)
(0.037)
(0.047)
(0.035)
-0.014
-0.087*
0.022
0.101*
0.078
∆ Ln School Spending t-5
(0.104)
(0.047)
(0.050)
(0.053)
(0.053)
-0.005
0.005
-0.001
0.005
-0.002
∆ Ln Crime rate
(0.008)
(0.005)
(0.006)
(0.006)
(0.006)
F-test
1.21
7,704
884
4,304
1,173
for excluded instruments
(P=0.523) (P=0.523) (P=0.523) (P=0.523) (P=0.523)
N
1,976
1,976
1,976
1,976
1,976
Notes:
1. All regressions include a series of time indicator and community-specific time trend variables.
2. Robust cluster standard errors in parentheses
3. The tax abatement variable (G_IFT) is treated as endogenous.
4. “Property Tax” indicates the non-residential property tax.
5. * significant at 10%; ** significant at 5%; *** significant at 1%

114

Table 4.B.5 First Stage IV Results from Column 2 in Table 4.22
Specification
(1) FDIV
(2) FDIV (3) FDIV
(4) FDIV (5) FDIV
∆ Ln
∆ Ln
∆ Ln
∆ Ln
Comp.
Comp.
Dependent Variables
Property
School
∆ G_IFT
Property
School
Tax
Spending
Tax
Spending
Own Community Variables
0.004**
-0.001
-0.001
-0.002
-0.001
∆ Election Year t-1
(0.002)
(0.002)
(0.002)
(0.002)
(0.002)
0.107
0.010
-0.001
0.016
-0.0001
∆ E_IFT
(0.078)
(0.012)
(0.011)
(0.011)
(0.011)
∆ Ln Changes in Property
0.257
1.021***
-0.102
0.045***
0.086
Tax due to Proposal A
(0.156)
(0.014)
(0.070)
(0.014)
(0.076)
-0.142**
0.062
0.068**
0.078
0.043
∆ Ln Property Tax t-5
(0.062)
(0.083)
(0.026)
(0.083)
(0.029)
∆ Ln Changes in School
2.087*
0.194*
-0.807**
0.191*
0.430
Finding due to Proposal A
(1.184)
(0.105)
(0.313)
(0.105)
(0.364)
0.129*
0.146***
-0.005
-0.049
-0.030
∆ Ln School Spending t-5
(0.076)
(0.047)
(0.028)
(0.047)
(0.030)
Competitor Community Variable
0.422***
-0.055
-0.102
-0.026
0.002
∆ CG_IFT
(0.152)
(0.111)
(0.071)
(0.112)
(0.075)
-0.337
-0.074***
0.066
-0.029
-0.035
∆ CE_IFT
(0.221)
(0.023)
(0.053)
(0.023)
(0.053)
∆ Ln Changes in Property
-0.202
-0.004
0.117
0.972***
-0.080
Tax due to Proposal A
(0.161)
(0.016)
(0.078)
(0.016)
(0.081)
0.143**
-0.045
-0.037
-0.062
-0.020
∆ Ln Property Tax t-5
(0.072)
(0.078)
(0.036)
(0.078)
(0.038)
∆ Ln Changes in School
-2.005*
-0.189*
1.010***
-0.186*
-0.201
Funding due to Proposal A
(1.176)
(0.105)
(0.314)
(0.106)
(0.364)
-0.013
-0.092**
0.038
0.102***
0.054*
∆ Ln School Spending t-5
(0.057)
(0.036)
(0.029)
(0.036)
(0.031)
F-test
1
8,708
4
7,366
4
for excluded instruments
(P=0.379) (P=0.000) (P=0.000) (P=0.000) (P=0.000)
N
1,976
1,976
1,976
1,976
1,976
Notes:
1. All regressions include a series of time indicator and community-specific time trend variables.
2. Robust cluster standard errors in parentheses
3. The tax abatement variable (G_IFT) is treated as endogenous.
4. “Property Tax” indicates the residential property tax.
5. * significant at 10%; ** significant at 5%; *** significant at 1%

115

Table 4.B.6 First Stage IV Results from Column 2 in Tables 4.24 and 4.26
Specification
(1) FDIV
(2) FDIV (3) FDIV
(4) FDIV (5) FDIV
∆ Ln
∆ Ln
∆ Ln
∆ Ln
Comp.
Comp.
Dependent Variables
Property
School
∆ G_IFT
Property
School
Tax
Spending
Tax
Spending
Own Community Variables
0.005**
-0.001
-0.0004
-0.0002
-0.001
∆ Election Year t-1
(0.002)
(0.001)
(0.002)
(0.001)
(0.002)
0.105
0.009
0.001
0.018**
0.001
∆ E_IFT
(0.078)
(0.008)
(0.011)
(0.009)
(0.011)
∆ Ln Changes in Property
0.245
1.029***
-0.180
0.052**
0.181
Tax due to Proposal A
(0.186)
(0.018)
(0.117)
(0.022)
(0.115)
-0.192**
-0.081*
0.154***
-0.107*
0.133***
∆ Ln Property Tax t-5
(0.090)
(0.045)
(0.045)
(0.057)
(0.048)
∆ Ln Changes in School
1.403*
0.026
-0.651*** 0.232***
0.225
Funding due to Proposal A
(0.829)
(0.052)
(0.224)
(0.065)
(0.222)
0.112
0.089***
0.007
-0.047
-0.015
∆ Ln School Spending t-5
(0.076)
(0.032)
(0.028)
(0.032)
(0.030)
Competitor Community Variable
0.330**
-0.029
-0.066
0.202***
0.007
∆ CG_IFT
(0.137)
(0.064)
(0.065)
(0.068)
(0.066)
-0.251
-0.002
0.030
-0.019
-0.018
∆ CE_IFT
(0.186)
(0.013)
(0.047)
(0.015)
(0.044)
∆ Ln Changes in Property
-0.211
-0.009
0.149
0.967***
-0.218*
Tax due to Proposal A
(0.186)
(0.018)
(0.117)
(0.022)
(0.116)
0.191**
0.067
-0.117**
0.122**
-0.102**
∆ Ln Property Tax t-5
(0.086)
(0.048)
(0.046)
(0.061)
(0.050)
∆ Ln Changes in School
-1.321
-0.0284
0.827*** -0.233***
-0.030
Funding due to Proposal A
(0.8344)
(0.0523)
(0.221)
(0.065)
(0.217)
0.003
-0.0512*
0.029
0.089***
0.044
∆ Ln School Spending t-5
(0.060)
(0.0301)
(0.029)
(0.030)
(0.031)
F-test
2
8,719
5
7,208
6
for excluded instruments
(P=0.047) (P=0.000) (P=0.000) (P=0.000) (P=0.000)
N
1,976
1,976
1,976
1,976
1,976
Notes:
1. All regressions include a series of time indicator and community-specific time trend variables.
2. Robust cluster standard errors in parentheses
3. The tax abatement variable (G_IFT) is treated as endogenous.
4. “Property Tax” indicates the non-residential property tax.
5. * significant at 10%; ** significant at 5%; *** significant at 1%

116

41

APPENDIX 4.C
Determining the Weighted Average of
“Competitor” Characteristics – Migration Competitors
To generate the “competitor” variables, there are five steps. In the following equations,
each i , j , and

t represents community, county and year, respectively42.

‫݅݉− ܫ‬
݊
݃‫݅ ݐ݊ܽݎ‬
‫݆ݏ‬
‫݅ ݅ ݈ݑ݌݋݌‬
ܽ‫ݐ݊݋ ݐ‬
݆

Step 1: For controlling community size, in-migrants and out-migrants per capita are calculated.
In-migrants‫= ݐ‬
݅
݆

Out-migrants‫= ݐ‬
݅
݆

(4.H.1)

ܱ‫݅ ݉ −ݐݑ‬
݃‫݅ ݐ݊ܽݎ‬
‫݆ݏ‬
‫݅ ݅ ݈ݑ݌݋݌‬
ܽ‫ݐ݊݋ ݐ‬
݆

(4.H.2)

Step 2: Obtain the in-migrants ratio with county in-migrants data.
In-migrants Ratio‫= ݐ‬
݅
݆

In-migrants݅
݆

∑௜
(In-migrants௧ )
௜
௝

(4.H.3)

Step 3: Using out-migrants at the county level and in-migrants ratio (4.H.3), the out-migrants at
the community level are calculated. Since data on out-migrants who moved from one community
to another within the same county is not available from Census sources, I use a proxy for the
number of out-migrants moving from one community to another within the same county. From
national Census data, 24.9 percent of the migrant population aged 5 years and older in 1995

41

This appendix is taken from Skidmore, et al. (2012)

42

The data is based on all 152 communities in the five county region surrounding Detroit
(Macomb, Monroe, Oakland, Washtenaw, and Wayne counties) over the 1973-2007 period.
117

moved to another community within same county. Following Schachter et al. (2003), I use the
݉
Out-migrants‫( = ݐ‬Out-Migrants݆ × (In-Migrants Ratio݅ )
)
݆

24.9 percent figure as an estimate for within county migration activity.

(H.4)

Step 4: With out-migrants (C.4), calculate the ratio between out-migrants (4.H.4) and totalmigrants in all “competitor” counties.
Out-migrants Ratio‫= ݐ‬
݅
݆

Out-migrants݅
݆

∑݆∑݅
(Out-migrants‫) ݐ‬
݅
݆

(4.H. 5)

Step 5: By multiplying the estimated out-migrants ratio by key values and then summing up
‫ݐ‬
Competitor Property Taxes‫݅∑݆∑ = ݐ‬
[(Property Taxes‫( × ) ݐ‬Out-migrants Ratio݅ )]
݅
݆
݅
݆
݆
weighted key values, the “competitor” variables are generated.

(4.H.6)

‫ݐ‬
Competitor School Spending‫݅∑݆∑ = ݐ‬
[(School Spending‫( × ) ݐ‬Out-migrants Ratio݅ )] (4.H.7)
݅
݆
݅
݆
݆
‫ݐ‬
Competitor Crime rates‫݅∑݆∑ = ݐ‬
[(Crime rates‫( × ) ݐ‬Out-migrants Ratio݅ )]
݅
݆
݅
݆
݆

118

(4.H.8)

APPENDIX 4.D
Instrument for School Spending:
Changes in the Total Funding for Public Schools

43

Prior to the implementation of Michigan’s Proposal A, the per pupil total funding for
public schools relied on three factors: the guaranteed tax base per pupil, local property values per
pupil measured by State Equalized Valuation per pupil (SEV),

44

and district millage rates. The

state determined the guaranteed tax base and allowed districts to set their own property tax rates.
The pre-reform state aid formula per pupil is calculated as:
∗
∗
‫ܸ݅ ≤ ݅ ݂ )ݐ ߬ × ) ܸ݅ − ݐܸ( + ݐܩ‬
݅
‫ݐ‬
݅  ܸ‫ݐ‬
‫ݐ‬
‫ = ݅ ݅ܣ‬ቊ
݀‫ݐ‬
∗
0݅  ݅ > ܸ݅
 ݂ ‫ݐ‬
    ܸ
‫ݐ‬

(4.I.1)

where G represents the flat grant per pupil in year ‫ ܸ݅ ݐ‬represents the state-determined
, ∗
‫ݐ‬

guaranteed tax base per pupil for a district ݅in year ‫ ܸ݅ ݐ‬represents local property values per
, ‫ݐ‬
pupil for a district ݅in year ‫ ݐ‬and ߬ ‫ ݐ‬is property tax millage rates for a district ݅in year ‫ݐ‬
,
.
݅

This pre-reform formula shows that districts with per pupil tax bases in excess of the

guaranteed tax base per pupil were ineligible for state aid. In these districts, local property tax
revenues were a primary public school funding source. In 1993, the year before the reform, the
guaranteed tax base was $96,270 and about 50 percent of districts were off-formula in Southeast
Michigan.

43

I use the pre- and post-reform state aid formula provided by the Michigan House and Senate
Fiscal Agencies (1994)
44

SEV is 50 percent of the estimated true market value.
119

Based on this pre-reform formula, the pre-reform per pupil total revenue for public
schools is calculated as:

∗
∗
‫ܸ݅ ≤ ݅ ݂ ݐ ߬ × ݐܸ +ݐܩ‬
  ݅
݅   ܸ ‫ݐ‬
‫ݐ‬
ܴܶ݅ = ቊ
‫ݐ‬
∗
ܸ݅ × ߬ ‫ܸ݅ > ݅ ݂ݐ‬
   ݅ ܸ
‫ݐ      ݅ ݐ‬
‫ݐ‬
(4.I.2)

where ܴܶ݅ represents the total revenue for a district ݅in year ‫ݐ‬
.
‫ݐ‬

After the imposition of Proposal A, the key component for public school funding is the

pupil that a district can receive. The state determines the foundation allowance per pupil in year ‫ݐ‬
so-called foundation allowance. The foundation allowance per pupil is the maximum revenue per
based on the amount of the eligible revenue that a district received in year ‫ .1 −ݐ‬The postreform state aid formula per pupil is calculated as:
‫݅ܨ ≤ ݅ܨ ݅ ,)810.0 × ݅ ܷ( −ݐ ܨ‬
݂ ‫ݐ‬
‫ݐ‬
‫ݐ‬
‫ = ݅ ݅ܣ‬ቊ ݅
݀‫ݐ‬
‫݅ܨ > ݅  ݅ ,)810.0 × ݅ ܷ( − ݅ܨ‬
݂ ‫ݐ‬
‫ܨ‬
‫ݐ‬
‫ݐ‬
‫ݐ‬

(4.I.3)

where ‫ ݅ܨ‬represents the foundation allowance for a district ݅in year ‫ ݅ ܷ ݐ‬represents
, ‫ݐ‬
‫ݐ‬

nonhomestead property values per pupil for a district ݅in year ‫ݐ‬and ‫ ݅ܨ‬is the cap of the
‫ݐ‬
foundation allowance.

45

This formula implies that a district with a foundation allowance above the cap will not
receive the full amount of the state-determined foundation allowance. However, it is possible for
these districts to levy an additional millage, which is referred to as a “hold harmless” millage, on
homestead property as a complimentary revenue source.
45

Nonhomestead property is considered to be business property, rental housing/house, and
vacation homes.
120

Based on the post-reform formula, the post-reform per pupil total revenue for public
schools is calculated as:

46

‫݅ܨ ≤ ݅ ݂ ݐ ܨ‬
 ‫ݐ ܨ‬
݅
‫ݐ‬
ܴܶ݅ = ቊ ݅
‫ݐ‬
‫݅ܨ > ݅ ݂  ݅ܨ‬
݅ ‫ܨ‬
‫ݐ   ݐ‬
‫ݐ‬

(4.I.4)

For the instrument for school spending, I focus on the small window before and after the

implementation of Proposal A. Since Proposal A was partially adopted in 1994 but not fully
implemented until 1995, the key instrumental variable for school spending is based on changes
in the total funding (ܴܶ݅ ) between 1993 and 1995.
‫ݐ‬

47

district ݅before and after the implementation of Proposal A and its changes as the key instrument
Tables 4.I.1-4.I.3 provide an example of how I calculate the total revenues per pupil for a

for school spending.

46

In this calculation, since the local decision to impose a hold harmless millage on homestead
property may be endogenous to tax base, I omit the per pupil local revenue that comes from hold
harmless taxes on homestead property.
47

Since the key policy variable is in the natural logarithm, I use the changes in the natural
logarithm of per pupil total funding for education resulting from Proposal A as the key
identifying instrument for school spending.
121

Table 4.D.1
The total revenue per pupil as funding for public schools : FY1992-1993
State
The
District
Gross
Local
Formula
Local
Total
Guaranteed
District Millage
Allowance
SEV
Aid
Revenue Revenue
Tax Base
Rate
Per Pupil
Per pupil
Per
Per Pupil Per Pupil
per pupil
Pupil
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(1)×0.001
(1)×0.00
(3)-(6)
(5)+(6)
1
1×(4)
×(2)+342

South
30.03
$96,720
$3,233
$195,489
$0
$5,870
$5,870
Lake
Van
45.25
$96,720
4,698
$80,627
$3,468
$1,050
$4,698
Buren
Total
Sample
37.52
$96,720
$3,954
$117,546
$694
$4,181
$4,874
Average
Note:
1. I assume that a district qualified for all of the incentive payments. Under this assumption, the
maximum flat grant per pupil in 1993 was $342.

122

Table 4.D.2
The Total Revenue Per Pupil as Funding for Public Schools: FY1994-1995
Local
Foundation
Nonhomstead State Formula
Total
Revenue on
District
Allowance
SEV
Aid
Revenue Per
18 Mills
Per Pupil
Per pupil2
Per Pupil
Pupil
Per Pupil
(1)
(2)
(3)
(4)
(5)
(1)-(4)
(2)×0.018
(3)+(4)
South Lake
$6,766
$82,649
$5,012
$1,488
$6,500
Van Buren
$5,717
$43,461
$4,935
$782
$5,717
Total
Sample
$5,996
$45,718
$5,018
$1,119
$5,700
Average
Note:
1. The cap of the foundation allowance is $6,500 in 1995 and the minimum of the foundation
allowance is $4,200.
2. For this calculation, I use the estimated nonhomestead SEV per pupil in 1993 by multiplying
the aggregate SEV per pupil in 1993 by the ratio of the nonhomstead SEV per pupil in 1995 to
the aggregate SEV per pupil in 1995. This is useful to avoid the possibility that the
instruments could include an endogenous response to the reform.
† In the case that a foundation allowance per pupil is larger than $6,500, because of the cap of
foundation allowance, state formula aid per pupil is calculated as: $6,500-(4)

123

Table 4.D.3
Changes in the Total Revenue Per Pupil for Public Schools
as the Key Instrument of School Spending

South Lake
Van Buren
Total Sample
Average

Total Revenue Per
Pupil
FY1994-1995

Changes in
Total Revenue
Per Pupil

(1)

District

Total Revenue Per
Pupil
FY1992-1993

(2)

$5,870
$4,698

$6,500
$5,717

(3)
(2)-(1)
$630
$1,019

$4,874

$5,700

$826

124

APPENDIX 4.E Estimated Time Slopes
Table 4.E.1 The Number of Granted Tax Abatements (G_IFT) and
of Granted Competitor Tax Abatements (CG_IFT)
Estimated Time Trend
Estimated Time Trend
Slopes in The Number of
Slopes in The Number
Community
Granted Competitor Tax
of Granted Tax
1
Abatements (G_IFT)
Abatements (CG_IFT)
Fraser
-0.272***
0.032***
Memphis
0.084***
0.015***
Mt. Clemens
0.044***
0.029***
New Baltimore
0.044***
0.030***
Richmond
0.035***
0.030***
St. Clair Shores
0.031***
0.029***
Sterling Heights
0.033***
0.029***
Utica
0.031***
0.030***
Warren
0.038***
0.029***
Armada Township
0.105***
0.028***
Bruce Township
0.031***
0.030***
Chesterfield Township
0.081***
0.030***
Clinton Township
0.078***
0.029***
Harrison Township
0.033***
0.029***
Lake Township
0.031***
0.033***
Lenox Township
0.031***
0.029***
Macomb Township
0.017***
0.030***
Ray Township
0.031***
0.030***
Richmond Township
0.003***
0.031***
Shelby Township
0.037***
0.029***
Washington Township
0.028***
0.030***
Luna Pier
0.031***
0.031***
Monroe
0.039***
0.030***
Petersburg
0.031***
0.032***
Ash Township
0.068***
0.029***
Bedford Township
0.033***
0.030***
Berlin Township
0.037***
0.031***
Dundee Township
0.042***
0.030***
Erie Township
0.040***
0.030***
Exeter Township
0.031***
0.031***
Frenchtown Township
0.057***
0.030***
Ida Township
0.031***
0.031***
La Salle Township
0.031***
0.031***
London Township
0.031***
0.032***
Milan Township
0.058***
0.024***
Monroe Township
0.037***
0.030***
125

Table 4.E.1 (cont’d)
Community
Raisinville Township
Summerfield Township
Whiteford Township
Berkley
Birmingham
Bloomfield Hills
Clawson
Farmington
Farmington Hills
Ferndale
Hazel Park
Huntington Woods
Keego Harbor
Lathrup Village
Madison Heights
Novi
Oak Park
Orchard Lake Village
Pleasant Ridge
Pontiac
Rochester
Royal Oak
South Lyon
Southfield
Sylvan Lake
Troy
Walled Lake
Wixom
Addison Township
Bloomfield Township
Brandon Township
Commerce Township
Groveland Township
Highland Township
Holly Township
Independence Township

Estimated Time Trend
Slopes in The Number
of Granted Tax
Abatements (G_IFT)
0.031***
0.049***
0.035***
0.031***
0.031***
0.031***
0.031***
0.026***
0.031***
0.003***
0.027***
0.031***
0.031***
0.031***
0.030***
0.031***
0.037***
0.031***
0.031***
0.037***
-0.130***
0.033***
0.016***
0.036***
0.031***
0.027***
0.042***
0.006***
0.031***
0.031***
0.031***
0.026***
0.031***
0.034***
0.028***
0.031***

126

Estimated Time Trend
Slopes in The Number of
Granted Competitor Tax
1
Abatements (CG_IFT)
0.031***
0.030***
0.030***
0.012***
0.012***
0.012***
0.012***
0.012***
0.012***
0.012***
0.012***
0.012***
0.012***
0.012***
0.012***
0.012***
0.012***
0.012***
0.012***
0.012***
0.018***
0.012***
0.013***
0.012***
0.012***
0.012***
0.012***
0.014***
0.012***
0.012***
0.012***
0.012***
0.012***
0.012***
0.012***
0.012***

Table 4.E.1 (cont’d)
Community
Lyon Township
Milford Township
Novi Township
Oakland Township
Orion Township
Oxford Township
Rose Township
Royal Oak Township
Southfield Township
Springfield Township
Waterford Township
West Bloomfield Township
White Lake Township
Ann Arbor
Milan
Saline
Ypsilanti
Ann Arbor Township
Augusta Township
Bridgewater Township
Dexter Township
Freedom Township
Lima Township
Lodi Township
Lyndon Township
Manchester Township
Northfield Township
Salem Township
Saline Township
Scio Township
Sharon Township
Superior Township
Sylvan Township
Webster Township
York Township
Ypsilanti Township

Estimated Time Trend
Slopes in The Number
of Granted Tax
Abatements (G_IFT)
0.031***
0.008***
0.031***
0.031***
0.030***
0.010***
0.031***
0.027***
0.031***
0.013***
0.027***
0.031***
0.031***
0.030***
0.008***
0.004***
0.046***
0.042***
0.031***
0.031***
0.031***
0.031***
0.025***
0.031***
0.031***
0.031***
0.043***
0.031***
-0.031***
-0.204***
0.031***
0.031***
0.013***
0.038***
0.031***
0.021***

127

Estimated Time Trend
Slopes in The Number of
Granted Competitor Tax
1
Abatements (CG_IFT)
0.012***
0.012***
0.007***
0.012***
0.012***
0.012***
0.012***
0.011***
0.012***
0.012***
0.012***
0.012***
0.012***
0.016***
0.016***
0.016***
0.016***
0.016***
0.016***
0.015***
0.016***
0.016***
0.016***
0.016***
0.015***
0.016***
0.016***
0.014***
0.019***
0.019***
0.016***
0.016***
0.016***
0.015***
0.016***
0.016***

Table 4.E.1 (cont’d)
Community
Allen Park
Belleville
Dearborn
Dearborn Heights
Detroit
Ecorse
Flat Rock
Garden City
Gibraltar
Grosse Pointe
Grosse Pointe Farms
Grosse Point Park
Grosse Pointe Woods
Hamtramck
Harper Woods
Highland Park
Inkster
Lincoln Park
Livonia
Melvindale
Northville
Plymouth
River Rouge
Riverview
Rockwood
Romulus
Southgate
Taylor
Trenton
Wayne
Westland
Woodhaven
Wyandotte
Brownstown Township
Canton Township
Grosse Ile Township
Grosse Pointe Township
Huron Township

Estimated Time Trend
Slopes in The Number
of Granted Tax
Abatements (G_IFT)
0.037***
0.047***
0.035***
0.031***
0.028***
0.021***
0.045***
0.029***
0.046***
0.031***
0.031***
0.031***
0.031***
0.026***
0.031***
0.034***
0.030***
0.032***
0.045***
0.007***
0.031***
0.071***
0.031***
0.037***
0.042***
0.060***
0.023***
0.056***
0.037***
0.052***
0.029***
0.049***
0.042***
0.043***
0.036***
0.036***
0.031***
0.038***
128

Estimated Time Trend
Slopes in The Number of
Granted Competitor Tax
1
Abatements (CG_IFT)
0.025***
0.023***
0.025***
0.025***
0.025***
0.025***
0.025***
0.025***
0.025***
0.025***
0.025***
0.025***
0.025***
0.025***
0.025***
0.025***
0.025***
0.025***
0.025***
0.025***
0.025***
0.023***
0.025***
0.025***
0.025***
0.025***
0.025***
0.025***
0.025***
0.025***
0.025***
0.025***
0.025***
0.025***
0.025***
0.025***
0.026***
0.025***

Table 4.E.1 (cont’d)
Community
Northville Township
Plymouth Township
Redford Township
Sumpter Township
Van Burden Township

Estimated Time Trend
Slopes in The Number
of Granted Tax
Abatements (G_IFT)
0.029***
0.092***
0.037***
0.033***
0.031***

Estimated Time Trend
Slopes in The Number of
Granted Competitor Tax
1
Abatements (CG_IFT)
0.025***
0.024***
0.025***
0.025***
0.025***

Note:
1. Competitor tax abatement variable is based on “migration competitors”
2. * significant at 10%; ** significant at 5%; *** significant at 1%

129

REFERENCES

130

REFERENCES

Anderson, J.E., Wassmer, R.W., 1995. The decision to “bid for business”: municipal behavior in
granting property tax abatements. Regional Science and Urban Economics 25, 739–757.
Anderson, J.E., Wassmer, R.W., 2000. Bidding for Business: The Efficacy of Local Economic
Development Incentives in a Metropolitan Area. W.E. Upjohn Institute for Employment
Research, Kalamazoo, Mich.
Anselin, L., 1988. Spatial econometrics: methods and models. Springer.
Bartik, T.J., 1989. Small business start-ups in the United States : estimates of the effects of
characteristics of states. Southern Economic Journal 55, 1004–1018.
Bartik, T.J., 1991. Who benefits from state and local economic development policies? W.E.
Upjohn Institute for Employment Research, Kalamazoo, Mich.
Bartik, T.J., 1992. The effects of state and local taxes on economic development: a review of
recent research. Economic Development Quarterly 6, 102–111.
Bartik, T.J., 1994. Jobs, productivity and local economic development: what implications does
economic research have for the role of government? National Tax Journal 47, 847–861.
Bartik, T.J., Eisinger, P., Erickcek, G., 2003. Economic Development Policy in Michigan, in:
Ballard, C.L., Courant, P.N., Drake, D.C., Fisher, R.C. (Eds.), Michigan at the Millennium :
A Benchmark and Analysis of Its Fiscal and Economic Structure. Michigan State University
Press, East Lansing, Mich., pp. 279 – 297.
Baum, C.F., Schaffer, M.E., Stillman, S., 2007. Enhanced routines for instrumental
variables/GMM estimation and testing. Stata Journal 7, 465–506.
Bertrand, M., Duflo, E., Mullainathan, S., 2000. How much should we trust differences-indifferences estimates? The Quarterly Journal of Economics 119, 249–275.
Besley, T., Case, A., 1995. Incumbent Behavior: Vote-Seeking, Tax-Setting, and Yardstick
Competition. The American Economic Review 85, 25–45.
Block, R.N., Belman, D., 2003. Automotive and Other Manufacturing Industries in Michigan:
Output, Employment, Earnings, and Collective Bargaining, 1980–2001, in: Ballard, C.L.,
Courant, P.N., Drake, D.C., Fisher, R.C. (Eds.), Michigan at the Millennium : A Benchmark
and Analysis of Its Fiscal and Economic Structure. Michigan State University Press, East
Lansing, Mich.

131

Bradbury, K.L., Mayer, C.J., Case, K.E., 2001. Property tax limits, local fiscal behavior, and
property values: evidence from Massachusetts under Proposition. Journal of Public
Economics 80, 287–311.
Brett, C., Pinkse, J., 2000. The determinants of municipal tax rates in British Columbia.
Canadian Journal of Economics 33, 695–714.
Brueckner, J.K., 1979. Property values, local public expenditure and economic efficiency.
Journal of Public Economics 11, 223–245.
Brueckner, J.K., 1982. A test for allocative efficiency in the local public sector. Journal of Public
Economics 19, 311–331.
Brueckner, J.K., 2003. Strategic interaction among governments: an overview of empirical
studies. International Regional Science Review 26, 175–188.
Buettner, T., 2003. Tax base effects and fiscal externalities of local capital taxation: evidence
from a panel of German jurisdictions. Journal of Urban Economics 54, 110–128.
Case, A.C., Rosen, H.S., Hines Jr., J.R., 1993. Budget spillovers and fiscal policy
interdependence: evidence from the states. Journal of Public Economics 52, 258–307.
Charney, A.H., 1983. Intraurban manufacturing location decisions and local tax differentials.
Journal of Urban Economics 14, 184–205.
Citizens Research Council of Michigan, 2007. Survey of Economic Development Program in
Michigan Report No. 347.
Cragg, J.G., 1971. Some statistical models for limited dependent variables with application to the
demand for durable goods. Econometrica 39, 829–844.
Crane, L.P., Manion, M.M., Spiecker, K.F., 1996. A Cost of Community Services Study of Scio
Township. Washtenaw Land Trust, Ann Arbor, Michigan.
Feldman, N.E., Courant, P.N., Drake, D.C., 2003. The Property Tax in Michigan, in: Ballard,
C.L., Courant, P.N., Drake, D.C., Fisher, R.C. (Eds.), Michigan at the Millennium : A
Benchmark and Analysis of Its Fiscal and Economic Structure. Michigan State University
Press, East Lansing, Mich., pp. 577–602.
Fischel, W.A., 1975. Fiscal and Environmental Considerations in The Location of Firms in
Suburban Communities, in: Mills, E.S., Oates, W.E. (Eds.), Fiscal Zoning and Land Use
Controls : The Economic Issues. Lexington Books, Lexington, Mass.
Fischel, W.A., 1987. The Economics of Zoning Laws : A Property Rights Approach to American
Land Use Controls. Johns Hopkins University Press, Baltimore.

132

Fischel, W.A., 1995. Regulatory Takings : Law, Economics, and Politics. Harvard University
Press, Cambridge, Mass.
Freedgood, J., 2004. Cost of Community Services Studies: Making The Case for Conservation.
Farmland Information Center, Washington, D.C.
Gérard, M., Jayet, H., Paty, S., 2010. Tax interactions among Belgian municipalities: Do
interregional differences matter? Regional Science and Urban Economics 40, 336–342.
Goss, E.P., Phillips, J.M., 1999. Do business tax incentives contribute to a divergence in
economic growth. Economic Development Quarterly 13, 217–228.
Guilfoyle, J.P., 1998. The effect of property taxes and school spending on house prices :
evidence from Michigan’s Proposal A. Michigan Dept. of Treasury, Office of Revenue and
Tax Analysis, Lansing, Mich.
Hamilton, B., 1975. Zoning and property taxation in a system of local governments. Urban
Studies 12, 205–211.
Hamilton, B.W., 1976. Capitalization of intrajurisdictional differences in local tax prices. The
American Economic Review 66, 743–753.
Haughwout, A., Inman, R., Craig, S., Luce, T., 2004. Local revenue hills: evidence from four
U.S. cities. The Review of Economics and Statistics 86, 570–585.
Hoyt, W.H., 1991a. Property taxation, Nash equilibrium, and market power. Journal of Urban
Economics 30, 123–131.
Hoyt, W.H., 1991b. Competitive jurisdictions , congestion , the Henry George Theorem: When
should property and. Regional Science and Urban Economics 21, 351–370.
Kezdi, G., 2004. Robust standard-error estimations in fixed-effect panel models. Hungarian
Statistical Review 9, 95–116.
Krelove, R., 1993. The persistence and inefficiency of property tax finance of local public
expenditures. Journal of Public Economics 51, 415–435.
Ladd, H.F., 1975. Local education expenditures, fiscal capacity, and the composition of the
property tax base. National Tax Journal 28, 145–158.
Lang, K., Jian, T., 2004. Property taxes and property values: evidence from Proposition 21/2.
Journal of Urban Economics 55, 439–457.
LeSage, J.P., Pace, R.K., 2009. Introduction to Spatial Econometrics. CRC Press, Boca Raton.

133

Luce, T.F., 1994. Local taxes, public services, and the intrametropolitan location of firms and
households. Public Finance Review 22, 139–167.
Mcguire, T.J., 1985. Are local property taxes important in the intrametropolitan location
decisions of firms? An empirical analysis of the Metropolitan area. Journal of Urban
Economics 18, 226–234.
Michigan House and Senate Fiscal Agencies, 1994. Appendix A. The Michigan School Aid Act
compiled and Appendices.
Mieszkowski, P., 1972. The property tax: an excise tax or a profits tax. Journal of Public
Economics 1, 73–96.
Mieszkowski, P., Zodrow, G.R., 1989. Taxation and the Tiebout model: the differential effects of
head taxes, taxes on land rents, and property taxes. Journal of Economic Literature 27,
1098–1146.
Oates, W.E., 1969. The effects of property taxes and local public spending on property values: an
empirical study of tax capitalization and the Tiebout hypothesis. Journal of Political
Economy 77, 957–971.
Oates, W.E., 1972. Fiscal Federalism. Harcourt Brace Jovanovich, New York.
Oates, W.E., 1973. The effects of property taxes and local public spending on property values : a
reply and yet further results. Journal of Public Economics 81, 1004–1008.
Palmon, O., Smith, B.A., 1998. New evidence on property tax capitalization. Journal of Political
Economy 106, 1099–1111.
Papke, L.E., 1991. Interstate business tax differentials and new firm location: Evidence from
panel data. Journal of Public Economics 45, 47–68.
Pollakowski, H., 1973. The effects of property taxes and local public spending on property
values : a comment and further results. Journal of Political Economy 81, 994–1003.
Reese, L.A., Larnell, T.B., Sands, G., 2009. Patterns of tax abatement policy: lessons from the
outliers? The American Review of Public Administration 1, 1–23.
Reese, L.A., Sands, G., 2012. Money for nothing : industrial tax abatements and economic
development. Lexington Books, Lanham, Md.
Reese, L.A., Ye, M., 2011. Policy versus place luck: Achieving local economic prosperity.
Economic Development Quarterly 25, 221–236.

134

Ross, S., Yinger, J., 1999. Chapter 47 Sorting and voting: A review of the literature on urban
public finance, in: Cheshire, P., Mills, E.S. (Eds.), Applied Urban Economics. Elsevier, pp.
2001–2060.
Samuelson, P.A., 1954. The pure theory of public expenditure. The Review of Economics and
Statistics 36, 387–389.
Schachter, J.P., Franklin, R.S., Perry, M.J., 2003. Migration and Geographic Mobility in
Metropolitan and Nonmetropolitan America: 1995 to 2000. Washington, D.C.
Skidmore, M., Reese, L., Kang, S.H., 2012. Regional analysis of property taxation, education
finance reform, and property value growth. Regional Science and Urban Economics 42,
351–363.
Staiger, D., Stock, J.H., 1997. Instrumental Variables Regression with Weak Instruments.
Econometrica 65, 557–586.
Stephen, M., Mcguire, T.J., Papke, L.E., 1994. The influence of tax on employment and
population growth : evidence from the Washington, D. C. Metropolitan area. National Tax
Journal 53, 105–123.
Tiebout, C.M., 1956. A pure theory of local expenditures. Journal of Political Economy 64, 416–
424.
Wassmer, R., 1994. Can Local Incentives Alter a Metropolitan City’s Economic Development.
Urban Studies 31, 1251–1278.
Wassmer, R.W., 2007. The increasing use of property tax abatement as a means of promoting
sub-national economic activity in the United States.
Wassmer, R.W., Anderson, J.E., 2001. Bidding for Business: New Evidence on the Effect of
Locally Offered Economic Development Incentives in a Metropolitan Area. Economic
Development Quarterly 15, 132–148.
Wasylenko, M.J., 1980. Evidence of fiscal differentials and intrametropolitan firm relocation.
Land Economics 56, 339–349.
Wellisch, D., 2000. Theory of Public Finance in a Federal State. Cambridge University Press,
Cambridge, U.K.
White, M.J., 1975. Firm Location in A Zoned Metropolitan Area, in: Mills, E.S., Oates, W.E.
(Eds.), Fiscal Zoning and Land Use Controls : The Economic Issues. Lexington Books,
Lexington, Mass.
Wildasin, D.E., 1986. Urban Public Finance. Harwood Academic Publishers, Chur, Switzerland.

135

Wildasin, D.E., 1989. Interjurisdictional capital mobility: Fiscal externality and a corrective
subsidy. Journal of Urban Economics 25, 193–212.
Wilson, J.D., 1999. Theories of tax competition. National tax journal 52, 269–304.
Wilson, J.D., Wildasin, D.E., 2004. Capital tax competition: bane or boon. Journal of Public
Economics 88, 1065–1091.
Wolkoff, M., 1985. Chasing a dream: the use of tax abatements to spur urban economic
development. Urban Studies 22, 305–315.
Wooldridge, J.M., 2010. Econometric analysis of cross section and panel data. MIT Press,
Cambridge, Mass.
Yinger, J., Bloom, H.S., Börsch-Supan, A., Ladd, H.F., 1988. Property Taxes and House Values :
The Theory and Estimation of Intrajurisdictional Property Tax Capitalization. Academic
Press, Boston.
Zodrow, G.R., 2000. The property tax as a capital tax : a room with three views. National Tax
Journal 54, 139–156.
Zodrow, G.R., Mieszkowski, P., 1983. The Incidence of The Property Tax: The Benefit View
Versus The New View, in: Local Provision of Public Services : The Tiebout Model After
Twenty-Five Years. Academic Press, New York.
Zodrow, G.R., Mieszkowski, P., 1986. Pigou, Tiebout, property taxation, and the underprovision
of local public goods. Journal of Urban Economics 19, 356–370.

136