MIT LIBRARIES
3 9080 02618 0221
DUPL
Digitized by the Internet Archive
in
2011 with funding from
MIT
Libraries
http://www.archive.org/details/biddingforindust0439gree
DEWEY
Massachusetts Institute of Technology
Department of Econonnics
Working Paper Series
Bidding for Industrial Plants: Does Winning a
"Million Dollar Plant" Increase Welfare?
Michael Greenstone
Enrico Moretti
Working Paper 04-39
November, 2004
RoomE52-251
50 Memorial Drive
Cambridge,
02142
MA
This paper can be downloaded without c harge from the
Social Science Research Network
Paper Collection
http://ssrn.com/abstract=6231 22
at
MASSACHUSETTS INSTITUTE
OF TECHNOLOGY
DEC
? 200^1
LIBRARIES
I
Bidding for Industrial Plants:
Does Winning a 'Million Dollar Plant' Increase Welfare?
Michael Greenstone
MIT, American Bar Foundation and
NBER
Enrico Moretti
University of California, Berkeley and
NBER
November 2004
We
thank Alberto Abadie, Michael Ash, Hal Cole, David Card, Gordon Dahl, Thomas Davidoff, Michael
Davidson, Rajeev Dehejia, Stefano Delia Vigna. Mark Duggan, Jinyong Hahn, Robert Haveman, Vernon
Henderson, Ali Hortacsu, Matthew Kahn, Tom Kane, Brian Knight, Alan Krueger, Steve Levitt, Boyan
Jovanovic, David Lee, Therese McGuire, Derek Neal, Matthew Neidell. Aviv Nevo, John Quigley, Karl
Scholz,
Chad Syverson, Duncan Thomas and seminar
Columbia,
Illinois,
Michigan,
NYU, NBER Summer
Wisconsin for very helpful discussions.
Kanter,
Yan
Lee,
Moretti thanks the
Adina Allen, Ben
at
Brown, Chicago,
Wharton, and
Gallagher, Genevieve PhamBerkeley,
Rice. Stanford,
Bolitzer. Justin
UCLA,
Schulhofer-Wohl, Antoine St-Pierre, and William Young provided outstanding
Greenstone acknowledges generous funding from the American Bar Foundation.
Sam
research assistance.
participants
Institute.
UCLA
Senate for a generous grant.
Bidding for Industrial Plants:
Does Winning a 'Million Dollar Plant* Increase Welfare?
Abstract
Increasingly, local governments
locate
within
their jurisdictions.
compete by offering
substantial subsidies to industrial plants to
This paper uses a novel research design to estimate the local
consequences of successfully bidding for an industrial plant, relative to bidding and losing, on labor
Each issue of Ihe corporate real estate journal Site
earnings, public finances, and property values.
Selection includes an article titled "The Million Dollar Plant" that reports the county where a large plant
chose
to locate (i.e., the 'winner'), as well as the
these revealed
one or two runner-up counties
(i.e.,
the 'losers').
We
use
rankings of profit-maximizing firms to form a counterfactual for what would have
happened in the winning counties in the absence of the plant opening. We find that the plant opening
announcement is associated with a 1.5% trend break in labor earnings in the new plant's industry in
winning counties, as well as increased earnings in the same industry in counties that neighbor the winner.
Further, there is modest evidence of increased expenditures for local services, such as public education.
Property values may provide a summary measure of the net change in welfare, because the costs
and benefits of attracting a plant should be capitalized into the price of land. If the winners and losers are
homogeneous, a simple model suggests that any rents should be bid away. We find a positive, relative
trend break of approximately 1.1-1.7% in property values. Since the winners and losers have similar
observables in advance of the opening announcement, the property value results may be explained by
heterogeneity in subsidies from higher levels of government (e.g., states) and/or systematic underbidding.
Overall, the results undermine the popular view that the provision of local subsidies to attract large
industrial plants reduces local residents' welfare.
Michael Greenstone
Enrico Moretti
MIT Department
University of California, Berkeley
of Economics
50 Memorial Drive, E52-359
Cambridge.
02 1 42- 347
MA
and
NBER
mgreenst@,m it.edu
1
Department of Economics
CA 94720-3880
Berkeley,
and
NBER
moretti@econ.berkeley.edu
Introduction
•
Over
the past 30 years, state and local governments have
economic development.
assumed
a greater responsibility for
For example, they frequently offer substantial subsidies to businesses
These incentives can include tax breaks, low-cost or free
within their jurisdictions.
to locate
land, the issuance
of
tax-exempt bonds, training funds, the construction of roads, and other infrastructure investments. These
policies are controversial: local politicians and the subsidized
companies usually extol the benefits of
these deals, while critics complain that they are a waste of public monies.'
competing claims, because there
is little
systematic evidence on the consequences of these policies.^
approach to evaluating policies designed
I'he traditional
plant in
is
it
to calculate the
is
widely cited
for locating in
Vance,
Georgetown, Kentucky was awarded $200 million ($80,000 per job) and
Boeing was given $50 million ($100,000 per job)
it
plants
Mercedes received a $250 million ($165,000 per job) incentive package
Alabama, the Toyota
in
new
For example,
at attracting
number of jobs gained and the cost of the tax breaks awarded to firms.
that
difficult to evaluate these
It is
in
tax abatements to locates
its
corporate headquarters
Chicago (Mitol 2001; Trogen 2002). This "accounting" approach produces eye-catching
but
statistics,
has two important limitations. First, these calculations are done ex-ante and are rarely verified ex-post.
Second, and more fundamentally,
this
approach does not offer a framework
policies increase or decrease welfare of local residents.
the residents of Vance,
On
'
the
results
on
this question, as
one hand, some models suggest that the
positive spillovers and/or increases producer surplus.
enhancing
or, at least, neutral for local residents.
local subsidies
may
is
determining whether the
$165,000 per job
good deal
a
for
Alabama?
Economic theory provides mixed
recent survey.
For example,
for
reflect
government
gain or satisfying Leviathan goals.
On
emphasized by Glaeser (2001)
attraction
in a
of new businesses generates
In these cases, local subsidies
may
be welfare
the other hand, there are models that indicate that
officials' private interests
In these settings, local
by providing for
government
their
own
financial
officials grant subsidies
whose
costs to local residents are larger than the benefits and therefore reduce residents" welfare.
-
This paper empirically assesses the consequences for counties of successfully bidding for large
industrial plants
'
on county-level earnings, property values, and government finances.
For example, a recent front-page
provided generous subsidies to
limited benefits to the local
article in the
attract large
New York Times
manufacturing firms
economy. ("Paying
for Jobs
The empirical
describes the experience of several localities that
in
the 1980s and 1990s and apparently obtained
Doesn't Always Pay Off,
New
York Times,
November
10,
2003).
"
"seem to be a
economic landscape, economists do not yet know why these incentives occur and
desirable". Further a Standard & Poor's publication states, "Economic development wars of
In a recent survey of the literature, Glaeser (2001) concludes that, although location based-incentives
permanent part of
whether they are in
the urban
fact
recent years have not developed a sufficient track record to assess their true cost-benefit ratio" (Standard
1
993 ). Bartik
(
1
99
1
)
provides the most comprehensive evidence.
&
Poor's
challenge
choose to locate where
that plants
is
their
expected profits are highest, which are a function of
A
any subsidies.
their location-specific expected future costs of production and
depend on many
costs of production in a county
plant's expected future
unobservable factors, including the presence of a suitable
skills, and the local regulatory
transportation infrastructure, the availability of workers with particular
environment. These factors are typically
difficult to
measure and
factors, including
number of unmeasured
importance varies across
Similarly, the subsidies are likely to be a
plants based on plants' unobserved production functions.
function of a
their relative
any potential spillovers and the degree of local
politicians' malfeasance.
Heterogeneity
counties
the factors that determine variation in costs of production and subsidies across
in
likely to bias standard estimators. Valid estimates
is
identification of a county that
is
of the impact of a plant opening require the
identical to the county where the plant decided to locate
We
future production costs and the factors that determine subsidies.
ascertain and
As
measure
all
a solution,
we
rely
to
open a large
subsequently narrow the
both expected
in
our ability to
in
the absence of the plant opening. These rankings come
how
which includes a regular
a large plant decided
plant, they typically begin
to roughly 10
list
sites,
where
selection process, but narrowly lost the competition.
firms are considering
by considering dozens of possible
among which
the 'losers').
(i.e..
feature titled the "Million
When
to locate.
(i.e.,
The "Million
the 'winner'), as well as the
The
losers are counties that
Our
identifying assumption
is
have survived a long
is
invalid,
we
suspect that this pairwise sampling approach
is
form a
that the losers
in pre-existing trends.
valid counterfactual for the winners, after adjustment for differences
assumption
They
locations.
2 or 3 finalists are selected.
Dollar Plants" articles report the county that the plant ultimately chose
one or two runner-up counties
to identify a valid
on the revealed rankings of profit-maximizing firms
real estate journal Site Selection,
Dollar Plants" that describes
where
in
these factors.
counterfactual for what would have happened
from the corporate
have
little faith
Even
if this
preferable to the leading
alternative of comparing winners to the other 3,000 U.S. counties.
Consistent with our assumption,
we
find that before the
winning and losing counties have similar trends
employment-population
ratio.
in
wage
bill,
years before the announcement of the winner, the trends
finding
'
suggests
that
Under
not
the winner county,
employment, per capita income and
Several other observable characteristics are also similar.
trends in the winning counties tend to be different from trends in
are virtually identical.
announcement of
the plausible assumption
only
are
in
all
other
US
By
counties. Notably
contrast,
in
the 8
property values in winning and losing counties
that property
markets are forward-looking,
this
before
the
winning and losing counties similar
in
the
years
tax competitions and the
See also the classic Gates (1Q72) and Zodrovv and Mieszkowski (1986) papers on
Wilson 1999) provides another recent review of this literature.
provision of local public goods.
(
annouiiceincnt. but the expected Future changes
values are also ex-ante similar.
economic
in
activity
arc capitalized into property
tliat
Overall, these findings lend credibility to the identifying assumption that
the losers form a valid counterfactual for the winners.
The primai7 employment
by
a statistically significant
opening announcement
1-digit industry in
result
that in the
$16.8 million per year
in
new
wage
bill in
in the
the
new
wage
some of
the
attraction
bill in
first
of a
new
the
I
-digit industry the
winning counties (relative
plant's
I
-digit industry
We
wage
to losing
1.5% of the average wage
is
Taken
year before the announcement.
the plant opening than predicted by pre-existing trends.
the total
plants'
(relative to the period before). This
winning counties
that six years later the
is
increased
bill
ones) after the
bill in
the plant's
literally, this
implies
roughly $100 million higher due to
is
also find evidence of positive trend breaks in
plant's industry in counties that neighbor the winner.'
These
results provide
ex-post estimates of the increase in local economic activity due to the successful
large,
new
plant and indicate that the
new
activity doesn't
crowd out existing
activity
locally.
However, by themselves, these findings
local residents' welfare
subsidies
effect
is
We
unavailable.
of attracting a plant
the increased
economic
taxes—or reduction
We
in
cost.
will be reflected in property values.
(i.e.
This
may
homogeneous and
pay
values. This
from
state
because property values capitalize
(i.e.
the cost of the subsidy).'
derive a simple model that demonstrates that in the presence of bidding for plants by
cause property values to increase, decrease or remain unchanged.
politicians solely
maximize
This
is
is
for part
because counties raise their bids
When
because the county
that has the
less than the
politicians derive private benefits
most
homogenous
result in increased property
attractive characteristics or the largest contribution
second best county, and
from granting subsidies, they
still
win the
will overbid,
',',,,
decrease.
of
until the costs
counties are not
may
of the subsidy, the successful attraction of a plant
government can bid
When
residents' welfare, the successful attraction
equal the benefits and they are indifferent about winning or losing.
if states
is
the benefit of the subsidy) and the increase in property
public services— necessary to pay for the subsidy
a plant will leave property values unchanged.
or
Direct information on the cost of the
follow the stylized Roback (1982) model and argue that the net welfare
activity in the county
counties, a successful bid
counties are
are not informative about whether the subsidies increase
because they do not account for their
.,
,
when
local
and property values
may
plant. Finally
•
,
,
.
See Hanson (1998) for evidence that local shocks affect neighboring jurisdictions and that this effect dissipates
with distance. See Chapter 2 of Bartik (1991) for a review of the effects of state and local economic development
programs on employment and economic growth in local areas. On a related topic, Evans and Topoleski (2002)
examine the social and economics consequences of the opening of an Indian casino.
'
Gyourko and Tracy 1991 demonstrate that the efficient provision of public services is capitalized into land. They
(
)
find that with services held constant, higher taxes are associated with lower land values.
Ultimately, the question of the effect of successfully attracting a plant on property values
empirical one.
Using a unique self-collected data
results in increased property values. Specifically,
break of approximately 1.1-1.7%
model's assumptions are valid,
Even
if
in
file,
we
we
find that the successful attraction
find that in
this finding
can be interpreted as an increase
The
of local subsidies
final part
to attract large industrial plants
of the paper estimates the
is
welfare for local residents.
in
Governments,
we
find that local
governments
in
in police
government finances.
A
Using data from the Annual Survey of
winning counties experienced positive and roughly equal
trend breaks in revenues and expenditures. Notably, there
and no change
local
governments pay for them by cutting
that local
important services, such as education and police protection.''
expenditures. Overall, there
is
is little
a substantial increase in education spending
evidence of deterioration
in the
provision of
public services in counties that win '"Million Dollar" plants.
We
for a county
want
to
emphasize
that the focus
of successfully bidding
for a
new
related policy and/or welfare questions, only
of
this
paper
the credible estimation of the consequences
is
plant, relative to bidding
some of which
analysis does not provide evidence on whether states
competitions between local governments to attract
new
and
losing.
For example, the
business.
Further,
it
whole benefit from
tax
is
not informative about
at
all.
However, since
openings induce tax competitions between jurisdictions, the salient question for a
government interested
in attracting
one of these plants
consequences of successfully bidding. The analysis
The paper proceeds
local subsidies
There are a number of
paper can address.
this
or the nation as a
whether individual counties should pursue industrial development policies
virtually all large plant
It
a relative trend
decreases welfare.
winning a plant on
effect of
widespread concern about the provision of subsidies
local
is
annual property values after the plant opening announcement. If our
the assumptions are not valid, the property value results appear to undermine the popular view
that the provision
vital
winning counties there
an
is
of a plant
as follows.
by local governments
also provides a case study of
describes our research design.
Section
1
is
Section
III
but what are the
and the determinants of plants' location decisions.
new
plant in South Carolina and intuitively
reviews the Roback (1982) model
opening and presents a stylized model of counties' decisions
on property values. Section
ail
structured to inform this very question.
decision to locate a
II
not whether to bid at
reviews the theoretical explanations for the provision of
to attract plants
BMW's
is
to bid for
describes the data sources and presents
Glaeser (2001) emphasizes the importance of answering
what happens when localities are deprived of the marginal
this
in
the context of a plant
a plant and the effect of winning
some summary
statistics.
Section
question empirically, "Economists need to estimate
dollar.
Does
this loss lead to eliminating
very valuable
services or are fairly marginal services cut off?" (pp. il-12).
If
outcomes
in
counties that bid and lose are similar to counties without industrial development policies, then the
analysis sheds light on the broader question of the consequences of whether industrial development policies are
welfare enhancing. There
is
no a
priori reason to believe that this condition
does or does not hold.
IV explains the econometric
iikhIcI
and Section
V
describes the results.
Section VI interprets the results
and VI concludes.
1
I.
This section
is
Local Subsidy and Plant Location Decisions
divided into two parts. The
governments use incentives
local
provides a case study of
BMW's
to atti'act
Why do
more
real
Some New
of incentives
to
decisions.
is
localities
models may be unable
that the standard
Here,
in
we
review a recent
line
why
The second subsection
jurisdictions.
the Greenville-Spartanburg
Plants?
firms reduces the welfare of a locality's
because the incentives lead to inefficiently high levels of local production. Yet,
world, large businesses frequently receive subsidies
implication
why
is
tiicir
generally, describes the intuition of our research design.
Local Governments Offer Subsidies to
In standard models, the provision
residents.^ This
reviews the existing theoretical literature on
plants to
decision to locate a car assembly plant
area of northwestern South Carolina and,
A.
first
some
in
exchange for
to capture
of research that provides
provide incentives to newly locating firms.
in the
The
their location decision.
an important feature of these location
a
number of possible explanations
for
some of
the
This review aims to highlight
"structural" forces that are likely to underlie our "reduced- form" estimates of the welfare consequences of
successfully attracting a plant.
The
first set
of theories emphasizes the idea
One example
welfare of their residents.
number
the
in
that local
governments aim
to
maximize only
the
where agglomeration economies are associated with
the case
For example, finns
the location of the firm.
as the
is
same
industry
may experience
productivity increases
or size of geographically concentrated firms increase (Henderson 2003; Garcia-Mila and
McGuire 2001). These
spillovers
workers (Ranch 1993; Glaeser
other firms in the future.
to attract the firms that
may be due
et al.
to the sharing
of information, the number of high skilled
1995; Moretti 2004a and 2004b), or that the
Regardless of their source,
in the
new
firms will attract
presence of these spillovers, localities will bid
produce them and the resulting subsidies allow the firm
to capture the spillovers
that they produce.
Glaeser (2001) suggests that local incentives
that will generate
analysis suggests that there
'
'
may
represent bids by communities to attract firms
producer surplus for the current residents of the community.
may
Conventional welfare
be welfare triangles to be gained by local workers. In particular,
See Wilson (1999) for a review of the theoretical literatures on tax incentives.
). See that paper for further
This review follows a recent paper by Glaeser (2001
details.
if the
labor supply curve
new
The
firm.
upward sloping, inframarginal workers
is
Another explanation for tax incentives
tax
payments (Wilson 1996).
change
its
made
are
size of the bids will reflect the welfare gains generated.
location, because
Gross 2000 and
tax collectors.
Ramey and
Since
that these upfront
of the "sunk" nature of many
Shapiro 2001).
The
by the presence of the
payments are compensation
to operate at a site
it
to credibly
makes firms easy
commit
advance
in
for future
can be costly to
industrial investments (see e.g.,
resulting immobility
governments
difficult for
is
it
is
once a plant begins
In particular,
better off
'"
Goolsbee and
targets for local
to future tax rates, the
upfront tax incentives compensate firms for future expropriation.
All of the previous theories are consistent with local politicians solely acting in the interests of
their citizens, but
an alternative view
desire on the politician's part to
influence, the side-payments
employment
future
bribes will depend
may
is
due
to corruption
and influence or a
government.
In the case
of corruption and
that local incentives are
maximize the
size of
occur through direct payoffs, contributions
to re-election
for politicians or their friends or family (Glaeser 2001).
An
on the probability of detection and the punishment.
derived from the Leviathan view of government (Brennan and Buchanan 1980).
politicians will seek to
maximize the welfare of their
For the purposes of our paper,
are motivated by agglomeration
expropriation--suggest that subsidies
(We
will
second
subsidies
two
show
set
later
it
is
worth noting that the
may be
in
first set
As
is
applied to this case,
of models— where local subsidies
producer surplus or compensation for ex-post
welfare enhancing
or, at least, neutral for local residents.
under what conditions subsidies can be welfare enhancing or welfare neutral.) The
of models— where local subsidies reflect government
may
alternative possibility
of government or the tax base.
citizens and the size
economies or increases
campaigns, or
The magnitude of these
be welfare decreasing.
We
officials' private interests— suggest that
return to this point in Section
II,
where we incorporate these
alternative views in a simple theoretical framework.
B.
A
its
new
Case Study of the
BMW Plant Location Decision and a New Research Design
In this subsection,
plant. In particular,
we
use a concrete example to
we
estate Journal Site Selection to describe
BMW's
1992 decision
Greenville-Spartanburg area of South Carolina. Notably, this
goal of this case study
it is
also possible that the
a particular firm selected a site for
is
to
site
one of the
to highlight the empirical difficulties that arise
is
plant openings on local economies.
if there are local
how
illustrate
use information from the "Million Dollar Plant"" series
new
we
(e.g.,
sports teams). Because
our purposes.
,
;
the corporate real
plants in our sample.
when
we
A
second
estimating the effect of
use this case study to informally explain
firm generates consumer surplus for the residents.
markets for outputs
possibility' is less relevant for
Further,
in
a manufacturing plant in the
The consumer gains
why our
are possible
focus mainly on manufacturing plants, this
may circumvent
research design
Section
these identitlcation problems.
A more
formal analysis
is
After overseeing a worldwide competition and considering 250 potential sites for
BMW
announced
months
conducted
in
II.
1991 that
in
BMW
later.
they had narrowed the
announced
that the
two
of potential candidates
list
in
1992
new
plant,
20 counties.
Six
competition were Greenville-Spartanburg,
finalists in the
South Carolina, and Omaha, Nebraska. Finally,
to
its
BMW announced that they would site the plant in
they would receive a package of incentives worth appioxiniately $1
Greenville-Spartanburg and that
15
million funded by the state and local governments.
Why
profit
did
BMW
choose Greenville-Spartanburg?
seems reasonable
It
assume
to
that firms are
maximizers and choose to locate where their expectation of the present discounted value of the
stream of future profits
is
Two
greatest.
plant's expected future costs
a location, which
in
is
The
selection
it
receives
BMW
factors.
characteristics
at
first is
is
the present discounted value
the site.
case provides a rare opportunity to observe the determinants of these two key
Consider
that
the
a function of the location's expected
supply of inputs and the firm's production technology. The second factor
of the subsidy
The
factors determine their expected future profits.
of production
first
the county's
expected supply of inputs. According
made Greenville-Spartanburg more
attractive
than
the
to
other 250
site-
BMW,
sites
the
initially
considered were: low union density, a supply of qualified workers; the numerous global firms, including
58 German companies,
in
the area; the high quality transportation infrastructure, including
highway, and port access; and access
to
air,
rail,
county characteristics are a
first
key local services.
For our purposes, the important point to note here
potential source of unobserved heterogeneity.
BMW case, they are generally unknown.
that these
is
While these
characteristics are well
If these characteristics also affect the
documented
in the
outcomes of interest
(i.e.,
labor earnings, property values and county finances), a standard regression that compares Greenville-
Spartanburg with the other 3000
A
US
counties will yield biased estimates of the effect of the plant opening.
standard regression will overestimate the effect of plant openings on outcomes, if for example,
counties that have more attractive characteristics
better
outcomes
Now,
(e.g.,
(e.g., better
higher earnings).
consider the second determinant of
Dollar Plant" article explains
willing to provide
transportation infrastructure) tend to have
BMW
why
BMW's
decision, the subsidy
.
The
BMW
"Million
the Greenville-Spartanburg and South Carolina governments
with $115 million
in
subsidies."
According
were
to local officials, the facility's
estimated five-year economic impact on the region was $2 billion (although
this
number
surely does not
As
account for opportunity cost).
a part of this $2 billion, the plant was expected to create 2,000 direct
jobs and lead to another 2.000 jobs
new
indirectly created by the
by the
in related industries
BMW
A, Greenville-Spartanburg's subsidy for
plant.
may
late 1990s.''
As we argued
in
subsection
be rationalized by the 2000 "spillover" jobs
As an example. Magna
International began construction on an
million plant that was to produce roofs, side panels, doors and other major pieces for the
BMW
$80
plant in
1993.'^
It
some
notable that
is
For
this reason, the factors that
represent
subsidy)
heterogeneity
a
second
may
counties
industrial structure, labor force skills,
benefit
unemployment
determine the
a particular plant, depending on their
and
the other factors that affect spillovers.
total size
all
of the spillover (and presumably the size of the
of unobserved
source
potential
more from
rate
heterogeneity.
If
this
unobserved
correlated with outcomes, standard regression equations will be misspecified due to
is
For example,
omitted variables, just as described above.
spillovers (and therefore offer
more generous subsidies)
compares the winners with the other 3000
US
if
counties that have more to gain
in
terms of
also have better outcomes, then a regression that
counties will overestimate the effect of plant openings on
outcomes.
In order to
make
valid inferences in the presence of these
two forms of heterogeneity, knowledge
of the exact form of the selection rule that determines plants' location decisions
As
the
BMW
example demonstrates,
expected future supply of inputs
in
two
the
is
faith in
to
location
decisions
be confounded by differences
—
the
unknown
to
we have
known
they are difficult to measure.
all this
information. Thus, the effect of a plant opening
our ability to ascertain and measure
very likely
generally necessary.
a county and the magnitude of the subsidy— are generally
researchers and in the rare cases where they are
little
factors that determine plant
is
in factors that
In short,
determine the plants" profitability
at the
chosen location.
As
a
solution
maximizing firms
plant opening.
chose
loser
(i.e.,
is
to
this
identification
problem,
to identify a valid counterfactual for
we
rely
on the revealed rankings of
what would have happened
in
profit-
the absence of the
In particular, the "Million Dollar Plants" articles typically report the county that the plant
the 'winner'), as well as
Omaha, Nebraska.
In the
its
2"''
choice
(i.e.,
subsequent analysis
the 'loser'). For example, in the
we assume
that the
BMW
case, the
winning and losing counties are
Ben Haskevv, chairman of the Spartanburg Chamber of Commerce, summarized the local view when he said,
"The addition of the company will further elevate an already top-rated community for Job growth" (Venable, 1992,
p. 630).
'"
Interestingly,
BMW
headquarters from
''
later
decided
New Jersey
to
to
open a second plant
Although the Magna Plant was slated
million
in
incentives.
in the
Greenville-Spartanburg area and relocated
its
U.S.
South Carolina.
Interestingly,
to hire
the
proportional basis) than those received by
300 workers,
incentives
BMW,
state
offered to
and local governments only provided about $1.5
Magna are substantially smaller (even on a
implying that local governments appear to be judicious
concentrating the incentives on plants that are likely to have the largest spillovers.
in
identical
in
assumption
expected
proHts. controlling for dilTcrcnccs
t'utiiie
we
unlikely to hold exactly,
is
regression adjustment to
compare
based on observable variables.''
identifying assumption
may
II.
pre-existing trends.
in
suspect that this pairwise approach
Although
preferable to using
the winners to the other 3,000 U.S. counties or a
matching procedure
In Section
we
V,
present empirical evidence that suggests that this
be valid.
Laud
When
Prices and Wellare
Counties Bids for Plants
This section presents a simple framework that guides the empirical analysis and helps
the resulting estimates.
new
One of the
paper's empirical goals
The
plant affects housing prices.
assumptions under which the change
interpreted as a
change
this
is
in
tlrst
to test
to interpret
whether the successful attraction of a
subsection presents a stylized model that specifies
some
land values induced by the exogenous opening of a plant can be
in residents' welfare.
In the
second subsection,
and specify some assumptions about the bidding process.
what conditions a successful bid
is
we
The goal of
for a plant will result in higher or
allow counties to bid for plants
this
subsection
is
to
show under
lower property values (and therefore
welfare).
A. Land Prices and Welfare
Here,
we
follow the stylized Roback (1982) model, which
We assume that
decisions.
Further, they rent land for
is
often used to model firm location
individuals are perfectly mobile, have identical tastes and a fixed labor supply.
homes
in the
returns to scale technologies and there
county where they work. All firms are assumed to have constant
is
a fixed supply of land in each county.''^
In equilibrium, firm's
unit costs equal the nationally determined product price and individuals' utility cannot be increased by
moving
to a different county.
Now. suppose
but
it
is
that a county
is
exogenouslv assigned a new
a useful starting point for expository purposes.
Importantly,
continue to choose their location to maximize profits and
The opening of the new
The
its
land purchase and by increasing the
'*
Propensity score matching
our approach
observables.
score
is
As
it
is its
is
all
This case
other producers and
unrealistic,
all
new
new
to increase in the county.
plant will directly increase the
demand
if
there
plant will lower the costs of production for other
an alternative approach (Rosenbaum and Rubin 1983).
assumption that the treatment
workers
respectively.
and nominal wages
First, the
may be
number of workers who need housing. Second,
are agglomeration economies, the presence of the
relative to
utility,
plant causes land values
increase in land values occurs for two reasons.
through
plant.
(i.e.,
winner
status)
is
Its
principal shortcoming
"ignorable" conditional on the
should be clear from the example, adjustment for observable variables through the propensity
unlikely to be sufficient.
See Hoehn, Berger, and Blomquist
(
1987) for a model that allows for flexible city boundaries.
In the parlance of
firms in the county.
Roback,
productive amenity.
this is referred to as a
And,
in the
presence of a productive amenity, firms will bid up the price of land to gain access to the spillovers.
In
order to retain workers, firms must pay higher wages to compensate them for the higher rental rate of
land, so that real
wages
are unchanged. This
necessary because
is
in
equilibrium workers'
must be
utility
constant across counties.
With
This
residents.
provides a one-time gain to property owners
this set-up, the increase in land values
county that receives the new plant.
is
This
because the increase
in
is
the only
wages
is
is
B.
A Stylized Model
of Bidding for Plants and
In the previous subsection, the
assumed
that the
new
plant
to
its
we
assume
jurisdiction.
in
exchange
the incentive
package
(e.g.,
in
In
practice,
local
In this subsection,
consider the possibility that the 3,000 U.S. counties compete for the
bids.
mayor
that acts as their agent to bid to attract plants
successful bid involves a trade-off for the county.
The increase
the plant.
attract
for a plant's location decision.
costly to the county, because they involve the provision of services and
tax revenues.
land
in
Land Values
that a county's residents elect a
A
Further, the higher rental rate of
welfare.
county does not have to incur any costs to
by offering subsidies or
We
unchanged.
to rent
opening of a new plant increased property values because we
governments frequently provide subsidies
and the remainder of the paper,
in the
welfare experienced by the county's
Thus, under these assumptions, changes
land counterbalances any spillovers available to tlrms.
in residents'
in
by the higher prices that workers face
offset
land (e.g., for apartments/houses), leaving their utility
values translate one-to-one in changes
change
On
the one hand, subsidies are
may reduce
public services includes the special services for the
the future stream of
new
plant stipulated by
the construction of roads, or other infrastructure investments, tax abatements,
job training funds, provision of low-cost or free land, the issuance of tax-exempt bonds, provision of
cheap electric power,
protection).
is
We
etc.), as
assume
well as the standard public services (e.g., garbage removal and police
that the county's incentive
package
is
financed by property taxes, so that
its
cost
capitalized into land values.'"
On
the other hand, the
which raises the value of
raise the value
land.
new
As
plant directly increases the level of
the
BMW
Annual Survey of Governmenis
activity in the county,
case demonstrates, the presence of the
of land indirectly by generating spillovers,
lowers the costs of production for other plants.
In the
economic
if,
for example,
it
new
plant
may
Thus, property values capitalize both the costs
data, property taxes
account for
49%
of
total
revenues from
For comparison, total sales and gross receipts taxes account for only 4.5% of total revenues.
IiO
also
attracts other plants and/or
own
(i.e.,
the
sources.
increased property taxes and/or reduced services) and beneHts
attracting
(i.e.,
increased economic activity) of
tiie
plant.
tiic
Tiiis subsection's goal
is
when
to theoretically analyze
increase or decrease property values (and therefore welfare)
denote propeity values as
expressed as AP,, =
Vi|
-
P,
and assume
C,j.
Vi,
that the
change
denotes the benefit of
to the county.
We
assume
also
that
known
is
it
We
the presence of county bidding.
property values lor the winning county can be
in
new
plant
The
increase in property values (in the absence of a subsidy).
the successful attraction of a plant will
in
j
for
county
i,
and
size of this benefit
is
to all the other counties bidding
it
eqiiivalent to the
is
exogenous and known
on the
plant.
Cy
is
the
cost to the county of the subsidy provided to the plant.
In the real world,
In
this case the total
contribution.
We
it
often the case that the state bears part of the cost of the incentive package.
is
subsidy received by the plant
assume
that
S
is
exogenous
to the
B,,
is
=
county and
+
Cij
is
where Sy denotes the
Sj,,
state's
provided by the state to account for the
benefits to other counties in the state.
The
we assume
plants are the other side of this two-sided matching problem and
locate in the county
where
their future profits are
their expected future profits in a given county: the subsidy and the expected future costs
that county.
offer,
due
subsidy,
values.
likely that there
It is
heterogeneity
is
in the
maximum
we assume
The
not any collusion
in the
bidding
In the
A
air, rail,
We
highway, and port access.
in
likely that there
It is
is
the expected supply of future inputs.
denote the value to the firm of
higher Zy implies that production costs of firm
exogenous
to the
We
county and
is
known
to
all
assume each county has
also
other counties,
(i.e., Z,j
for
all i)
and
all
j
different county.
If this
assumption
why such overbidding would be
is
B,
I
+
Z,,, is
We
assume
that
Z
is
other counties' bids
incorrect,
it
b\-
(i.e.. B,,
for
all
i).
This assumption prevents
exaggerating the benefits of locating
would cause counties
to
all
"overbid."
It is
in
a
not evident
an equilibrium strategy.
the county-specific cost advantages.
sum,
i.
these factors as
rational expectations about the plant's production costs in
Overall, the total value for a firm of locating
this
are lower in county
all
counties.
finns from increasing the subsidy that a count)' offers
where
counties.
BMW case, recall these factors included the presence of qualified workers, the presence of German
companies and
Z,j.
of independent, or private,
among
firm's choice also depends on the location-specific production costs.
heterogeneity in a plant's valuation of a county due to differences
in
In order to obtain the highest
that the firms conduct an English auction in the presence
is
of production
subsidy that counties are willing to
to differences in counties characteristics (recall Section I-B).
We further assume that there
that they will
maximized. As described above, two factors determine
It
maximized.
in a particular
seems reasonable
to
county
presume
is
the
sum of the
subsidy and
that a plant will select the
county
Homogeneous Counties.
1.
tiie
plants'
homogeneous
assumption
in
V
in
that
and Z:
we
Vi,
Z,j.
= Vq and
Here,
we
= Zq
for
Zjj
we
if this identification
assumption
Case
This case
J.
is
the simplest.
welfare and the state provides no subsidy
between winning and
losing.
We
S
(i.e.,
i.
This would be the case
valid.
our key identifying
if
The homogeneity case
is
important because
We
Z
and
to investigate
how
our
consider four cases.
assume
0).
V
in
losers.
Under homogeneity, the firm simply
not valid.
is
=
and
plant, Vy,
homogeneity assumption when comparing winners and
will retain the
chooses the county that offers the highest subsidy, B.
(!)
all
analyze the more general model that allows for heterogeneity
conclusions differ
new
begin by considering the case where counties are
winners and losers are identical were
our empirical analysis,
Later,
In genera], counties differ in their valuations of a
valuation of the county,
that the county's
The mayor
Formally, the equilibrium bid, B*,
mayor maximizes
raises the bid until she
is
residents'
indifferent
determined by
is
B* = Vo
AP = Vo - B* =
Consequently,
and the successful
attraction
of the plant does not change land prices or
residents' welfare.
Case
We now
2.
assumptions.
is
It
allow states to subsidize the incentives offered
magnitude of spillovers
in
neighboring counties differs across
most generous incentive from the
most generous
state
state will
win the new
does not need to raise the bid
until
she
states.
plant.
is
the county with the second
optimal bid B*
where
2"''
Here, the county that receives the
at the
point that
makes
the
between winning and
mayor of
losing.
The
S™,.,
the incentive provided by the second
most generous
states:
provided by the most generous
most generous
AP = Vq -
state.
(B*-S|„.ix)
=
state.
between the
S^ax
-
Sn.ax-i
state subsidy
>0, where
Importantly, the source of the increase
in
in
the state subsidies and the county's capture of part of the state subsidy.
3.
We now
the
mayor
the incentive
housing values
Case
allow for the possibility that the mayor ma\ have her
provided by the
S,„ax is
heterogeneity
we assume
located in the
indifferent between having the plant and not
state subsidy indifferent
Specifically, land values increase by the difference
most and
for example, the
is
B* = Vo +
S,„;,^.| is
most generous
if,
The mayor of the county
having the plant. She can win the plant by setting the county's bid
(2)
to the plant but retain the other
possible that different states provide different level of incentives
own
goals.
is
the
In particular,
benefits from a higher incentive package because opportunities for graft or
enlarging government are increasing in B.
higher the subsidy provided by the
mayor
We
define this personal benefit as
to the firm, the larger the kickback.
determined probability of detection and punishment, the mayor chooses B
which depends on residents' increase
in
welfare
AP
and her benefit T:
12
to
U = U(AP,
T =
f(B), with
Due
to
maximize her
T);
f >0:
the
an exogenously
where U| >
utility
U,
and U2
>
case where U2 = 0, there
In the
0.
are perfectly aligned.
own
U2 >
0, the
is
We
assume Ut
this case, the
having the
T
ma)or
no principal-agent problem and the mayor's and residents'
we assume
For simplicity,
of the bid
a fixed fraction y
is
(i.e.,
T=
U(AP,T)
that
y B),
where
is
separable
<y<
I
raises her bid to the point that
plant. If all the
mayors behave
in
the
makes her
same way,
indifferent
in its first
and
U = AP + yB.
Thus,
.
but reinstate the assumption that states do not subsidi/.c the bid
;<:
interests
mayor's objective function includes residents' welfare as well as her
private gain from the subsidy.
second argument and
(i.e.
S =
0).
In
between having the plant and not
the mayor's optimal bid
is
B*-Vo/(l-y)
(3)
In this case, the
Vo <
mayor overbids, by choosing B* which
Such overbidding causes land values
residents.
in
larger than the value of the plant to the
is
the winning county to decline relative to the losing
The magnitude of the decline depends on the mayor's weight on her own
county.
y))
If
AP=
welfare:
(y / (1-
-
0.
Cose
4.
When U2 ^
and there
is
a positive state subsidy
(i.e.,
S,|>0), the
cannot be signed. The state subsidy increases land values, as shown above
in
the mayor's personal gain in the objective function results
in
Depending on
the magnitude of these
two
in
effects, land values
a decrease
may
case
change
2,
in
land prices
but the inclusion of
Case
land values as in
3.
increase or decrease in the winner
'^
relative to the loser.
2.
Heterogeneous Counties. Although the key assumption
winning and losing counties are homogeneous,
assumption
is
not valid.
Here,
we
is
our empirical analysis
is
that the
important to understand the consequences
allow for heterogeneity
V, and plants" valuations of counties, Z.
effect
it
in
In section I-B,
in counties' valuations
we
used the
BMW
if this
of attracting the plant,
example
to
argue that the
of the plant opening could be confounded by these two sources of unobserved heterogeneity across
counties.
Specifically,
we
suspect that a naive estimator that ignores the presence of unobserved
heterogeneity will be biased. This subsection formalizes this point.
For simplicity,
the firm chooses
firm
j
where
we
to locate
of choosing county
(Vh) and low
V
(Vl); and
i
is
two
=
and Uj =
V
and Z vary across counties,
based not only on the bid B, but also on Z.
Specifically, the value for
retain the
assumptions that
Bii+Z,,.
Assume
levels of Z, high
S.j
for simplicity that there are only
Z
bid
is
such
the plant:
that the
j
county with low
and
is
two
(Zh) and low Z, (Zl). Consider
and Z are positively correlated, so that the county with high
gain the most from attracting firm
0. If
V
also has high Z.
levels
first
is
indifferent
V
where
V
the case
Thus, one county will
also the least cost production location for firm
Z and low V
of V, high
j.
Its
optimal
between having the plant and not having
B* = Vl-(Zh-Zl).
(4)
county that
In this case, the
the best
is
Land values increase by an amount proportional
Vl) + (Zh
-
Zl) >0.
match enjoys a
rent that
V
to the difference in
Importantly, the increase in land values
capitalized into land values.
is
and the difference
in
AP = (Vh -
Z:
due to the underlying differences
is
V
in
and
Z.
now
Consider
the case
where
V
Z
and
are negatively correlated.
Vh+Zl > Vl + Zh, county
County
while county 2 has high
Z and low
V. If
B*
indifferent
between having the plant and not having the
that
makes county 2
B* =
(5)
in
rent that
the case of positive correlation
Zl)
-
if
(V|i
Vh+Zl < Vl + Zh-
has high
V
and low Z;
plant:
+ (Zh-Zl)
V,.
The winner county enjoys a
applies
1
wins the plant by bidding an amount
1
is
capitalized in land values, although the rent
between Z and V: AP = (Vh
In this case
county 2
Vl) >0. Again, the heterogeneity
our empirical analysis are discussed further
III.
is
is
Vl)
-
the winner and
(Zh
-
its
is
lower than the rent
A
similar conclusion
Zl) >0.
AP =
land prices increase by
the source of the change
in the
-
(Zh
The implications
in prices.'^
-
for
econometrics section.
Summary
Data Sources and
Statistics
A. Data Sources
We
implement the design using data on winning and losing counties. Each issue of the corporate
real estate journal Site Selection includes
large plant decided
where
to locate."
an article
These
titled the
'winner'), and usually report the runner-up county or counties
(i.e.,
study indicated, the winner and losers are usually chosen from an
that in
many
impression
is
cases
number more than
a hundred."'
that they provide a representative
articles usually
indicate the plant's output,
The
that the plant
sample of
all
new
chose
a
the
(i.e.,
BMW
case
sample of ""semi-finalisf"
sites
the "losers").""
initial
As
the
on large
articles tend to focus
which we use
how
"Million Dollar Plants" that describes
always report the county
articles
large plant openings
and our
plants,
in
US.
the
to assign the plant to the relevant
The
-digit
1
industry.
In this case, the optimal bid
is
B* = (Vo +
S^m-i
)/
(l-y).
The change
in
land prices
is
AP=
- (y
/
(1-y))
Vo+
S^iK
-
(i/(i-y))S„,ax-,.
'^
In the unlikely case that
Vh + Zl = Vl +
Zh, the
two counties
are equivalent, and the
winner
is
randomly
determined.
In 1985, the journal Indiistiial
some years
Development changed
its
name
to Site Selection.
was
Henceforth,
Also,
"°
instances the "Million Dollar Plants" articles do not identify the runner-up county.
In
some
in
the feature ""Million Dollar Plants"
titled
did a Lexis/Nexis search for other articles discussing the plant opening and
losing counties.
in Site
"'
The Lexis/Nexis searches were
The names of the
in
semi-finalists are rarely reported.
14
refer to
it
as Site
For these cases,
we
4 cases were able to identify the
also used to identify the plant's industry
Selection.
we
"Location Reports."
Selection.
when
this
was unavailable
These data have two important
winning counties
counties. This
is
is
in
many
limitations.
cases unobserved and the bid
magnitude of the subsidy offered by the
is
almost always unobserved for losing
unfortunate, because an interesting check of the validity of our research design
whether the subsidies offered are equal
to test
First, the
in
the winning and losing counties. Second, in
would be
many
cases
the articles do not report the expected size of the plant.
order to conduct the analysis,
In
we
collected the most detailed and comprehensive county-level
data available on employment, government finances, and property values available for the period from
The employment data comes from
1970 through 1999.
(CBP) data
by
industry level.
many
CBP
The
trends in
Consequently,
we conduct
total
wage
bill at
the county
our analysis at the 1-digit industry by county
data are used to test whether a plant opening
industry, as well as other industries.
its
Census Bureau's County Business Patterns
In order to protect the confidentiality of the respondents, these data are "zeroed out" for
industry by county cells.
level."
the
These annual data report the number of employees and the
file.
employment since the
latter
cannot detect changes
We
is
associated with changes in
focus on the total
in the skill
wage
wage
bill
rather than total
bill
content of labor. Unfortunately, the
CBP
does not report hourly wages.
We
also test whether the successful attraction of a plant affects property values. ^^
unaware of any existing electronic
county by year data
governments
values."''
in
file
files
of annual county-level property value data, so
on property values from two sources.
we
our winner and loser samples directly and requested
These data
contacted
all
we supplemented
all
We
created our
are
own
the state and county
their historical data
because governments determine the value of property
exist
the purpose of assessing property taxes.
Second,
First,
we
on property
in their jurisdictions for
,
these data by
hand entering data from the 1972, 1977, 1982, 1987. and
1992 Census of Governments. Volume 2 Taxable Property Values and Assessment-Sales Price Ratios.
The censuses
reports market values in the year before each census
was unavailable from both
sources,
we
was conducted.
In years
where data
estimated county-level property values by linearly interpolating
the Census of Government data, which likely causes the true variation to be understated.
For the purposes of the analysis, we divide output into the 5 broad "1-digit" industries defined by the
CBP
for
which uncensored wage bill data are available in most years. These industries are: Manufacturing; Transportation
and Public Utilities; Trade (Wholesale plus Retail); Finance, Insurance, and Real Estate; Services. At this level of
aggregation,
^^
Ideally,
1
7% of the cells
we would have
are "zeroed" out.
also
examined the
effect
on property rental
rates, but
we were
unable to find a source for
annual or other high frequency rental data.
^''
We attempted to get this data for the
at least
1
primary sample of 166 counties dating back to the early 1970s.
earlier years.
We
collected
governments did not have data from the
For example, we have nonmissing property value date for 68 counties in 1977, 102 in 1980, 149 in
year of property value data from 153 counties.
In general, these
1990, and 153 in 1998.
15
One
limitation of these data
is
that our
and structures across residential and industrial
that
it
will include the value
but
we
is
the
is
sum of the
values of land
The drawback of this measure of property values
land.
discussed more extensively
in the
is
unavailable for
many of the
counties
mean
value of
new
plants' buildings.
Despite our extensive data collection efforts, the property value data are missing
Our
preferred property value sample
The Data Appendix provides more
losers.
is
details
fiscal
consequences for
and capital outlays).
The
and securities holdings.
held fixed.
of
all
In
The
is
ASG
used to determine the
is
an annual survey of
(e.g.,
education,
intergovernmental transactions, current operations,
aggregate these data to the county
that are
(ASG)
level.
by source, indebtedness, and cash
This aggregation
is
done on the
surveyed continuously from 1970 through 1999 so that the units are
our "Million Dollar Plant" sample of 166 winning and losing counties, 150 counties have
one governmental
unit that reports continuously.
governmental units
expenditures.
(i.e.,
data also contains information on revenues
We
sample of governmental units
at least
Statistics Series
on governments' expenditures by function
administration, and public assistance) and type
some
comprised of 30 of the 82 winners and 62 of the 129
governments of new plant openings.
local
that asks detailed questions
for
on the property value data.
The Amntal Survey of Governments: Finance
governments
in
subsequent description of the property value
note here that the estimated increase in propeity values appears to be substantially larger
than any reasonable estimate of the
counties.
is
of the new plants' structures, which mechanically causes measured property
Unfortunately, a series for land values only
values to increase.
our sample. This issue
results,
measure of property values
This
is
in
these counties
The continuous
but account
for
reporters comprise only
approximately
75%
12.5%
of revenues and
because large government units are sampled with certainty, while smaller units are
sampled with "varying probabilities within an area, type of government, and size ordering" (Census
Bureau, 1990,
B.
Summary
p.
1-1)."
Statistics
Table
1
presents
of the analysis. The
and an average of
first
summary
statistics
on the sample of plant location decisions
panel indicates that
1.6 losers per
winner or
in
a total
that
form the basis
our primary sample there are 82 separate plant openings
of 129 losers. There are 166 counties
in this
sample, so
According to the data documentation, the following governments are sampled with certainty: "all county
governments with a population of 50,000 or more, municipal governments with a population of 25,000 or more;
township government in the New England and Middle Atlantic states with a population of 25,000 or more; school
with an enrollment of 5,000 or more; and special
revenue or expenditure of $5 million."
districts
total
districts
16
with long-term debt outstanding of $10 million, or
the average county appears roughly 1.3 times. '"^
of
losers per winner.
We
The
losers.
analysis
is
were
most informative about the consequences of
in
announcement about
1991 and 1992, suggesting that they
of each plant,
identity
57 of the 82 cases there
table also reveals that 63 of the 82 plants
the distribution of the year of the
announced
In
its
reports the distribution of the niuriber
winner and accompanying loser(s) associated with each plant
refer to the
opening announcement as a "case."
The second panel
may
is
in the
and
a single loser
in
attracting industrial plants.
The
final
valuations of counties
be countercyclical. Appendi.x Table 2 reports the
Column
small.
(i.e.,
in
Z,j),
the three years before the
(3) presents the t-statistic and p-value from the
winning and losing counties are similar so
In contrast,
we
or the
announcement of the plant opening. These
columns
(1), (2),
and
test that the entries in (1)
that
any differences
in
and
columns
expect that a comparison of the winners (or losers) with
(4), respectively.^'
(2) are equal.
U.S. counties
all
We
(!) and (2) will be
is
likely
produce larger differences.
to
The
first
two panels reveal
that the
winning and losing counties are similar
pre-announcement trends of the primary outcome
of three outcome variables
total
wage
bill in
the
new
in
same conclusion.
The
third
row of
two
sets
The second panel compares
rates in earnings,
"*
the data.
127 counties appear once
the entire empirical analysis,
in
we exclude
the losers for a given case.
is
winning counties, compared
of counties. The count of full-time employees leads to
at
all
the
US
counties are
much
smaller."^
conventional significance levels.
employment and land values are not
Second,
statistically different in
33 counties appear twice and 6 counties are present three times. Through
the 7 counties with populations exceeding 2 million.
would be difficult to detect the impact of a plant opening.
"^
The losing county entries in column 2 are calculated in
The
in
not statistically meaningful and indicates that
trends in the outcome variables in the three years before the
announcement. The growth
cases so that each case
means
panel indicates that the hypothesis of equivalent aggregate property values across
winning and losing counties cannot be rejected
all
is
In contrast, the corresponding figures for
this
both the levels and
announcement of the plant opening. The mean
approximately $1,127 million
This difference
there were similar levels of activity in the
the
is
in
variables. Specifically, the first panel reports the
the three years before the
plant's industry
to $1,145 million in losing counties.
"^
(i.e., V,j)
Table 2 presents the means across counties of some likely
are reported for winners, losers, and the entire U.S. in
e.xpect that
lists
industry, and the winning and losing counties.
detenninants of these variables
means
Thus, our
panel
22 of the plant openings were
the plant opening.
Although we do not know the determinants of the counties' valuations of plants
plants'
14 there are 2
manufacturing industry.
we
the following manner.
First,
we
In these counties,
it
mean across
mean across all
calculate the
calculate the overall loser average as the unweighted
given equal weight.
figures in the top panel of column 4 are a weighted average for years 1982 to 1993, with weights proportional
number of Million Dollar cases in each year and industry. The figures in panels 2 to 5 of column 4 are a
weighted average for years 1982 to 1993, with weights proportional to the number of Million Dollar cases in each
year (see bottom of Table
for the distribution of cases across years).
to the
I
We
winning and losing counties.
trends in
wage
bill,
return to this issue in Section V, wiiere
employment and land values
we
winning and losing counties. For now,
property values
is
if
graphically
show
that the
note that the finding that both the level and growth rate of
similar in winning and losing counties
our research design,
we
before the announcement are similar in
in the 8 years
is
an especially important
Under
property markets are forward looking.
test
of the validity of
the plausible assumption that
property markets are forward looking, this finding indicates that the expected future changes in the level
of economic activity capitalized into property values are also ex-ante comparable. Overall, the
first
two
findings provide reassuring evidence on the quality of this research design
The
third panel reports
mean
conventional
criteria.
This
is
and changes of socioeconomic and demographic variables. In
levels
general, the null hypothesis of equal
means
in
opening announcement and the percent growth
important exceptions where the p-value
two
graduates. However, for these
mean
across
all
U.S. counties in
statistically different in
all
winning and losing counties cannot be rejected with
true both for the levels of the variables in the 3 years preceding the plant
is
in
those years for selected variables. There are two
.05 or less: level of per capita
column
(4).
income and
fraction
of college
mean than
is
the
Furthermore, for per capita income, the percent growth
is
not
variables, the losers"
mean
closer to the winners'
is
winners and losers counties. Interestingly, there are important differences between
U.S. counties and the winning counties.
larger population, a higher
In particular, the
employment-population
winners are richer, have a substantially
ratio, a better
educated population, and fewer people
over the age of 65.
The
fourth and fifth panels
show
the geographic distribution of plants across the four regions of
the U.S. and the industry distribution of the labor force within the categories.
winning and losing counties are not well balanced.
compared
to
39%
of the losers.
unobserved differences
(e.g.,
This
is
For example,
66%
Here, the means across the
of the winners are
potentially problematic, since
it
in the
suggests that there
South
may
be
union density) across the winning and losing counties.
This finding underscores the value of our identification strategy's reliance on comparisons of
changes
in
outcomes between winners and
losers, rather than cross sectional
comparisons. In this setting
cross-sectional differences will only bias the results if levels of these variables determine future changes.
This would be the case if for example, union density predicts growth
As
the next section discusses,
we
that
is
limited to cases
employment and property
values.
estimate models that include region by year fixed effects to account for
the uneven distribution of winner and losers across regions.
sample
in
where the winner and
We
also estimate
models on a restricted
loser counties are both from the South.
To preview
the results, our findings are unaffected by these specification checks, implying that the geographic
imbalance does not explain our
results.
IV. Econometric
In light
of the firm's selection
Model
rule, the goal is to estimate the causal effect
of winning a plant on
county-level outcomes. This section discusses the 2-step econometric model used to estimate this effect.
In the first step,
we
the following equation:
fit
17
17
= aic+
Yj
Y„c,
(6)
+ Yj
'^wtW.ic,
17
Y,„ =
(6')
where
but
i
is
it
^li
Lijci
+
Hi,
+
iiut+£,.ic.,
17
ac+ Y.
X
":vvtWjcx+
references industry,
j
J^lt Ljcx
+
Ht
+
%+
^.ic„
indicates a case, c denotes county, and
normalized so that for each case the year the plant opening
variable
in
equation
a,c
|i„ ([!,) is
a
is
(ttc)
in
full
set
outcome
variables.
The use of
-8
and
denotes year,
The outcome
0."'
county-level measure
that adjust for
same
in the
is
permanent
for all fixed county characteristics.
we
others
in
the smaller sample
that the industry x year (year) effects are the
Hut
in (6'), Y,^,, is a
x also
=
are the respective stochastic error terms.
These account
sample includes the entire U.S., while
Plant" sample of 166 counties.
>=
^j^,
is t
of indicators that nonparametrically controls for industry x year (year)
specifications the
(t1it) is
and
announced
of industry x county (county) fixed effects
the intercept of the
a vector
^ijci
indexes year,
t
is
wages. The outcome variable
(6), Y.jct, is total
of property' values or a government finance variable,
differences
or
t=-i9
T=-:9
restrict
it
effects.
In
some
our "Million Dollar
to
equivalent to imposing the restriction
166 counties and the remainder of the country.
a set of separate fixed effects for each of the cases interacted with an indicator for whether i
t
<=
We
5.
restrict attention to these
values of
i,
because the sample
is
balanced over
this
range.
W,,eT
industr>' in
and
are indicator variables. Wjjct (Wj^O equals
Wj^^^^
winning counties (the new plant's county)
analogously for losing counties.
1
on the new plant's
for observations
for a given value
of
i.
L,|ct
and
Lj^t
are defined
In the cases with multiple losers, this indicator variable will equal
1
for
observations from multiple counties within a case.
The
specific
vectors
means of
(or until) the plant
I.
and Kl are the parameters of interest
in
these equations.
They measure
the dependent variables in winning and losing counties, respectively,
are conditional on
with
ttv,
all
the indicator variables described above.
opening announcement.
For example,
tiwt
when
i
=
3
is
The period
is
where the means
determined by the years since
Thus, the effect of winner or loser status
the conditional
the period-
mean of the outcome
in
is
allowed to vary
winning counties 3 years
announcement and TiLiWhen
after the
=
i
-3
mean
the conditional
is
in the losing counties 3 years before
the announcement.
A
few
of the
details about the identification
the case fixed effects
(i.e., r|,„
>=
winner-loser pair for x
differencing in a
-8
and
<=
t
First,
and most importantly,
guarantee that the n's are identified from comparisons within a
r\^^)
and
bear highlighting.
ti's
5
and are a way
Second,
regression framework.
to retain
the intuitive appeal of pairwise
possible to separately identify the
is
it
Ji's
and the
industry x vear (vear) effects because the plant opening announcements occun'ed in multiple years.
Third,
are winner and/or losers multiple times and any observation from these counties will
some counties
simultaneously
Fourth, the specification does not control
multiple n's.
identify
covariates, such as the variables listed in Table 2. This
per capita income and population)
is
because many,
if
time-varying
for
not all of these variables (e.g.,
be affected by the plant opening so are likely endogenous.
may
Table 2 suggested that there might be important unmeasured region-specific determinants of the
outcome variables
in
(e.g.,
and county level
(6)
(6'),
in
nonparametrically adjust for
we
Since the analysis
differences in union density).
all
it
is
industry x county level
at the
possible to include region x year fixed effects
is
unobservables that vary across regions over time. As a
test
in
order to
of robustness,
present results from specifications that include these fixed effects.
The
implement
2"''
step provides a
to
summarize the
To
to infer the effect of attracting a plant.
jt"s
stack the vectors Jiw and n^ for -8<=
we
this step,
method
i
<=5
into a
28 x
1
vector.
This vector
is
the dependent variable in the following equation:
(7)
jisT
=
+5
1
(Winner) +
+ y (trend *
+ p (trend
where
*
y
trend
+X
(trend * l(Winner))
I(t>=0))
1
(Winner)
*
1
(x
>=
0))
indexes winner status and x remains the year relative to the plant opening announcement.
s
equation allows for a differential intercept for the
trend,
+ Vs„
\|/.
The parameter k measures whether
the trend differs after the
winner parameters.
It
also includes a
common
This
time
the time trend differs for winners, while y captures whether
announcement of the
plant opening
(i.e.,
when
x
>=
"
0).
when the plant begins production is unknown, so we use the year of the announcement of the winner
= 0. In most cases, the construction of the new plant starts immediately after the announcement.
'°
We include the winner fixed effect in (7), because (6) and (6') include all periods (i.e., fi-om -19<= t <=17), while
the second step only uses observations where -8<= x <=5. For this reason, the winner fixed effect in (7) is not
''
The
date
county as
T
collinear with the county by industry (county) fixed effects (6) and (6").
include the interaction of winner status and a
the trend break.
" Since
The
equal to one
if x
>=
We
also experimented with
to allow for a
mean
models
that
shift in addition to
findings from this approach are qualitatively similar.
the estimated
identical.
dummy
ir's
for winners (losers) are obtained
Consequently, the
fact that the Tt's are
from a balanced panel,
estimated
20
is
their standard errors are virtually
unlikely to be a source of heteroskedasticity
in the
P
the parameter of interest.
is
It
measures the difference
announcement of the
(relative to losers) after the
plant's
opening
in the
time trend specific to winners
0)),
Uwt]
=
It is
is
its
expected production costs
outcome variables
variables predict the
assumption
*
l(Winner)
*
1(t
>=
valid.
instructive to consider this assumption in the context of Section Il-B.
that Bj,
(i.e., Z,j)
in
(6)
are greatest.
and
It
is
we
In that section,
where the sum of the subsidy
that plants will plants will choose to locate in the county
negative of
announcement).
(relative to before the
Formally, the consistency of this paraineter requires the assumption that cov[(trend
likely that these
(i.e., B;,)
state
and the
same unobserved
Thus, consistent estimation requires the
(6').
and Zy are individually equal between the winner and loser counties, after adjustment
for the covariates in (6) or (6")
and
(7).
Notice, this
is
stronger than assuming that the
sum of B,| and Zn
is
equal after adjustment.
The
differential
growth
rates in the
South and Rust Belt
this period
in
and the
fact that the
winners are disproportionately from the South and the losers from the Rust Belt provide a basis for
concern that the identifying assumption may not hold. Specifically,
^Winner
^
gLoser
^
j}-"'",
but thc wiuucrs havc an advantage
winning a plant more and therefore make
losers value
source of bias
if
Z
and/or
B
it is
possible that on average B^^'"""
production costs
in
larger bids (B^°^"
growth and
are a proxy for a county's future
>
(i.e., z^^''"""
>
g*'"""^')
and the
jhis would be a
and
(6) or (6')
Z"-"'"')
+
(7) fail to fully
adjust for these differences.
Consequently,
estimation.
by year fixed
and
T
>=
is
it
especially important to clarify
Specifically, the identifying assumption
effects, case fixed effects, region
is
how we have
that the industry
guarded against inconsistent
by county fixed
by year fixed effects and the detrending by winner status
As
a
also report the results from the fitting of (6) or (6') and (7) on a subsample that
is
together condition out any differences in By and Zy between winners and losers.
robustness check,
we
restricted to cases
where the winner and
loser are both
from the South. ^" This sample
the possibility that the results are due to differential growth rates across regions
South). Notably, the paper's findings are similar
It is
This
is
noteworthy that
if
in
the
full
the identifying assumption
sum
think of cases where the
<
if
is
where
positive and cases
counties with bad future outcomes
ones that have the most
estimation of equation (7).
to gain
by attracting a
removes
the Rust Belt and
and restricted samples.
is
it
(i.e.,
new
restriction
(e.g.,
invalid, the bias cannot be signed a priori.
because the sign of the bias depends on whether cov(By,t>WT) + cov(Zy,UwT)
cov(Bij,uwt)
effects, industry
is
<
negative. For example,
declining
plant.
is
wage
>
or
it
is
0.
One
can
possible that
bills or land prices) are the
At the same time,
it
is
plausible that
Regardless, the subsequent results are insensitive to weighting by the square root of the
inverse of the standard errors of the n"s from equation (6).
cov(Zij,UwT)
outcomes
>
since counties with ciiaracteristics that firms find desirable
0,
(i.e.,
increasing
wage
bills or
may
also have
good future
land prices).
V. Results
This section
is
divided into three subsections. The
workers' earnings within the winning county and
for the
new
plant's
1
industry and for
-digit
its
reports the estimates of winner status on
surrounding counties.
The
II).
Separate results aie reported
The second subsection
other industries.
all
reports the
Under some assumptions, property values may
association between winner status and property values.
provide a measure of overall welfare (Section
first
third subsection
examines the association between
winner status and a series of local government revenue and expenditures categories.
Employment Outcomes
A.
Columns
and (2) of Table
(1)
the fitting of equation (6)
is
normalized so that
estimated Ttw and
total
=>
I
wage
>=
effects.
bill in
-8.
til
i
^
on
wage
total
is
for a given value of
i.
(3) reports the difference
on the same scale, the winner's
This same nonnalization
difference in the estimated tiw and
is
it
evident that
line
used
is
and
ttl's
their standard errors
is
announced. Each row reports the
Consequently, the point estimates are annual measures of the
in
the
1
-digit industry
of the new plant for 5
industry, industry by time, and case fixed
between the estimates of ttw and
til
in the
have almost identical trends from
i
=
is
in all
til
shifted up so that the difference
within each row.
is
-digit industry
-8
through
t
=
between
In order to get the
i.
the lines
1
a graphical version of column (3) of Table
of the new
-I
A
.
plant, the
statistical test
when
is
The bottom panel of Figure
the subsequent figures.
against i and
1
from
counties from 1970-1998. Recall, i
separately plots the estimates of tiw and Kl against
1
lines
bill in
CBP for all
millions of dollars by winner/loser status
=
wage
data from the
These estimates are conditioned on county by
Column
these graphs,
bill
estimated ttw's and
the year that the plant location decision
The top panel of Figure
-1.
3 report the
i
plots the
3.
From
winning and losing counties
confirms that the trends
in the
the winning and losing counties are statistically indistinguishable in the 8 years before the
plant opening
annoimcement
Importantly, this
(e.g.,
the
t-statistic
associated with the test of equal
trends
is
0.37.)
consistent with our identifying assumption that losing counties provide a valid
is
counterfactual.
However beginning with
the trend in the
'
wage
Specifically, a case
losers, the
sample
is
is
bill in
the
1
the year that the plant opening
-digit industry
of the new plant
required to have a winner and
at least
1
restricted to the losers located in the South.
22
is
in
announced, there
is
a sharp increase in
the winning counties.
loser located in the South.
In contrast, the
For cases with multiple
losing counties' trend
is
largely unchanged.
announcement with
the effect of the plant opening
mean
in differences
is
it
appropriate to model
more
a trend-break, rather than the
typical difference
shift.
7'able 4 reports estimates of
column
figure also demonstrates that
'I'iie
(1) (a), the JXw's
and
wage
digit industry, the
used
ttl's
1.''
and the top panel of Figure
and
(i
its
standard error that result from fitting equation (7).
were presented
to obtain these estimates are those that
The column
parameter estimate suggests that
(1) (a)
increased by a statistically significant $16.8 million per year
bill
Table 3
in
new
in the
In
plant's I-
winning
in
counties (relative to losing ones) after the announcement of the plant's opening (relative to the period
before the announcement)."' This
Taken
literally, this
digit industry
is
is
1
.5% of the average wage
implies that by the end of the period
bill in
(i.e., x
=
winning counties
5) the
wage
bill in
in the
year
new
the
=
x
-1.^'
plant's 1-
roughly $100 million (9%) higher due to the plant opening than predicted by pre-existing
trends.
It is
openings.
possible that this finding
Belt counties that are on a
bid a
lot to
downward
However,
it is
losers (recall Figure
Nevertheless, this
is
may
largely be
all
among winners and
in the
wage
an important issue and
restriction reduces the
to cases
In all the
When
and standard error
(in
subsequent figures, specification
In the
to 54.
parentheses)
column
(a) is
standard errors, the results are virtually identical.
''
column
This
after adjustment).
documented
in
Table
2.
opening announcement
specification checks to explore
to equation (6), so the
(b).
nw's and
Notably, the point estimate
In the
effects.
The
are
31.7
(1) (a) result
is
second check,
is
is
we
it.
ttl's
essentially
restrict the
This
loser are located in the South.
results are not
and
(7.0),
shown
in the table,
respectively.
but the
These two
not due to unobserved regional shocks.
used to obtain the ttw's and
ttl's.
the 1" step standard errors are clustered by county and year, and the
(1) (a) specification
losers
prior to the plant
where both the winner and
number of cases from 82
specification checks demonstrate that the
"'"'
than the plant
time varying region-specific factors. The results from the estirnation of the version of
sample of winners and losers
"
bill
we implement two
unchanged by the inclusion of the region by year fixed
estimate
rates, rather
comparisons of losing Rust
1).
equation (7) that uses these n's are presented in column (1)
point
to
on an upward trend (even
that are
check involves the addition of region by year fixed effects
are adjusted for
due
trend (even after adjustment for pre-existing trends) and willing to
undermined by the similar trends
among winners and
growth
to differences in regional
(a) estimate
supported by the geographic imbalance
is
first
due
winning Southern counties
possibility
The
is
For example, the column (1)
2"^*
step
is
weighted by the
f
step
For example, the p coefficient (standard error) from the column
16.6 (5.3).
(1) (a) specification
when
the dependent variable
estimate and standard error are 0.0079 (0.0037).
23
is
the In
of the
total
wage
bill,
the parameter
Column
the restricted
when
(2) reports the results
The estimates
are used to estimate P from equation (7).
bill
the estimated n's are obtained from fitting equation (6) on
sample of 166 winning and losing counties (rather than the
and
in (a)
As
entire U.S.).
before, the
(b) suggest that the trend in the
Tt's
wage
increased by $13.2 and $12.6 million, respectively. In the context of the standard errors, these
The
estimates are essentially identical to those in column (1).
cases where both the winner and the loser are from the South
is
point estimate (standard error) based on
shown
not
in the table,
but
it
is
28.0 (6.0).
Figure 2 presents the column (2) (a) results graphically.
As mentioned above, we focus on
cannot detect changes
to the
total
wage
bill
employment, rather than the
to
Table
4,
wage
Appendix Figure
ones. For example.
Returning
the total
bill
employment
rather than total
since the latter
content of labor. However, the employment results are generally similar
in the skill
total
column
wage
presents graphs that are analogous to Figure 2 for
1
bill.^"
(3) reports a "naVve" estimator that
obtained by using a standard
is
regression to test for a trend break in the winners after the announcement relative to before, controlling
for observable characteristics.
it is
It is
labeled "naive" because the losers are not used as a counterfactual, so
not conditioned on case fixed effects. This
the estimator that
is
would
typically be used in the absence
of the Million Dollar Plant research design. Interestingly, the point estimates here are roughly
than those
in
column
(1)
and almost double the column
50%
larger
This highlights the value of our
(2) estimates.'
research design.
Column
(1) of
Table 5 and Figure 3
associated with changes in
columns
is
(I) (a)
employment
whether the announcement of the new plant opening
is
winning county. The point estimates
in
and (1) (b) suggest that the trend
approximately 0.9-1.1% of the
i
=
-1
level.
significant by conventional criteria, but the
The remainder of Table
new
test
in other industries in the
5
plant affects the trend in the
in the
bill
in
bill
The column
column
one
(b)
and Figures 4 and
wage
wage
5
is
increased by $42.0-32.1 million, which
(a) estimate
would be judged
examines whether the successful
is
attraction
of the
Here, a neighbor
counties that neighbor the winner.
defined as a county that physically abuts a county or
statistically
imprecise.
We
connected by a bridge or boat.
is
compare the
neighbors of winners to the neighbors of losers. This part of the analysis offers an opportunity to assess
whether the wage
bill
gains "spillover" to neighboring counties or
In Figure 4, there appears to be a trend break in the total
plant.
The
estimates
in
column
The corresponding
is
million, or approximately
coefficient (standard error) from the
467.3 (176.2).
The
zero
sum
locally.
the
bill in
in
column
24
.2-1
.5%.
employment regression on
coefficient from a regression that includes
Estimates are based on data from the entire U.S. as
1
(
1
).
all
same
1-digit as the
They suggest
2 of Table 5 confirm this visual impression.
announcement trend increased by $36.5-40.4
Plant counties
is
wage
the
new
that the post-
This estimate borders
sample of Million Dollar
U.S. counties
is
287.8 (192.0).
on
statistical significance as
judged by conventional
industries from the (a) specification.
the
announcement of
criteria. F-'igure 5 plots the results for all other l-digit
The bottom panel
reveals that the relative
the plant opening in winning counties
is
column
reflected in the large point estimate ($33.8-161 .7 million) and standard errors in
These
Put another way, the effect of the plant opening announcement
neighbors.
fact,
winning counties are not the
results suggest that the gains in the
appears
to spillover to
The poor
neighboring counties."
downward
trend before
This
halted, but the data are noisy.
result
not zero
is
(3)
of Table
of losses
sum
is
5.
in their
locally and, in
precision of the ""other industries" estiinates
temper any conclusions about spillovers outside the new plant's
industry.'
B. Property Values
we
Mere,
assess whether winner status
is
associated with changes
property values. In a model
in
with rational, perfect-foresight agents, the net effect of winning the plant should be capitalized into
property values immediately after the announcement.
immediate mean
shift in
reappraised every few years in
gradually
in
As
property values.
many
our sample. Consequently,
states,
we
In this case,
it
would be appropriate
to test for an
the Data Appendi.x describes, property values are only
so that changes
in
property values will emerges only
continue to use the trend break model specified in equation
(7).
The property value
6, the
<=
T
sample
<=
sample
5.
fhis restriction
losers,
in
column
California to
( 1 )
of Table
in
2%
is
This
is
As
(relative to the period before the
announcement).
zero by conventional criteria.
also explored whether there are effects
in
a test of robustness,
we
included a
time-varying state factors. This
state.
(2)
of Table
-8
adds the
=
on the wage
is
full set
of
state
annual increases in
-1.
the plant's
These estimates would not be judged
Nevertheless, this
The column
bill at
little
property values increased
is
the state level.
The
to
be
approximately 0.6-0.8% of
(2) specifications
add California by
point estimates are large and
empirical content.
by year fixed effects
in
equation (6)
in
order to control for
very demanding of the data due to the small number of observations from each
5.
See the Data Appendix
(3)
announcement of
Nevertheless, the estimated trend breaks are generally of a larger magnitude across
Tables 4 and
and
Column
restricts
in the
(relative to losing ones) after the
positive but the standard errors are so large that they have
all
(1 )
the preferred sample.
imply that the trend
6, the point estimates
winning counties
statistically different than
imposed, because Proposition 13
annually.'"'
the average property value in winning counties at x
^'
columns
observations from California are dropped, which reduces the sample by 2
restriction that all
by $103-140 million
We
In
limited to the "Million Dollar Plant" counties with nonmissing property value data for
is
property values
opening
6.
This sample includes 32 of the 82 winners and 69 of the 129 losers.
winners and 7
In
Table 6 and Figure
results are presented in
for a further discussion
of the
difficulties with California data.
25
all
the
dependent variables
year fixed effects to (6')
in
modestly larger than those
Figure 6
two
that the
is
in
(1) but overall indicate a similar conclusion.
based on the preferred sample that excludes California counties.
of counties had remarkably similar trends
sets
announcement of the plant opening. This
and
due
2.
however
to the
economic
may
it
result
is
provide an even more meaningful
are
similar
it
wage
Figures
bill in
1
of our identifying assumption. Specifically,
test
suggests that the expectations about future
winning and losing counties
in
This figure shows
aggregate property values prior to the
in
similar to the findings for the
forwaid-looking nature of property markets,
activity
advance of the plant opening
in
Figure 6 also reveals a dramatic upward trend break
announcement.
These estimates are
order to account for the effect of Proposition 13.
column
property values
in
in
winners that
coincides with the announcement of the plant opening.
The corresponding
point estimates in
column
(3) of Table 6 indicate that the trend break in
winning counties ranges from $176-278 million. Conventional
statistically different
from zero. They are approximately 1.1-1.7% of the
They suggest
counties.
by the end of the examined period
that
winning counties had increased by $1.0-1.7
in
To
mean
put this in context, the
announcement
As
a
is
robustness check,
sample
It
is
billion (1983$), or 6.6-1
property values in winning
-1
6 years
(i.e.,
six-year change in property values
is
we
also estimated
after x
property values
-1),
relative to losing cotmties.
0.2''/'o,
in
=
the years before the opening
41 7.5 (104.4),
which
natural to
entirely reflect the
new
is
to the
sample that
a
is
restricted to cases
error)
from
this
larger than the estimates in Table 6.
measure of property values
wonder whether
of the new plant's building
models based on
from the South. The point estimate (standard
loser are both
Recall, a shortcoming of our
structures.
=
i
to be
57.2%.'"
where the winner and the
restricted
would judge these estimates
criteria
is
that
it
includes the value of land and
the property value results are mechanically due to the addition
assessment
where the property value
In the extreine
roles.
plant's structures, the estimates in this subsection
would not
reflect
results
an increase
in
local residents' wealth.
Some back of the
to Site Selection, the
(1983$).
This figure
sample. "*"
Site Selection
"
The
total
investment
this is not a first-order
it
seems
on average the plants
total
total
likely that they include the costs
in
1
-digit
wage
bill is
15.0
investment of the plants
lists for
in
our
of land, structures, equipment, and
13.6) for this
(
According
was $390 million
investment numbers, but based on
sample with specification (a).
list. For example, our
our sample are smaller than the ones on the Top 10
sample includes 12 plants with 1991 announcement dates but only 2 of them qualified
unable to locate top 10
concern.
the top 10 plant openings in 1991
vague about what variables enter the
is
point estimate (standard error) on the
We suspect that
among
almost certainly larger than the mean
is
our reading of the articles
^'
envelope calculations suggest that
average
other years.
,,
26
.
,
for the top 10
list.
We
were
perhaps job training tor
new
employees.'"
for a small portion of the total increase
million investment
is
it
Overall, the property value results
The
tliat
the
new
property values, even
in
correct for our sample and
plant reduces property values.'''
evident
is
fail
in
plants' structures can only account
the unrealistic case
where the
!t)390
only describes the cost of structures.
it
to provide evidence that the successful attraction of a big
implication
that the county's share
is
of the tax incentives
is
smaller
than the local benefits of the plant opening.'^
C. Local Government Finance Outcomes
A
with cuts
issue.
widespread concern about the provision of these subsidies
in
that local
from the estimation of
In this table, the tt's that are
equation (6')
used as the dependent variables
in (7)
on the sample of 150 "Million Dollar Plant" counties
one governmental unit reporting consecutively from 1970 through 1999.^* Column
mean of
the budget variables
estimate in column
The
table
(
1
)
to this
winning counties
in
at i
=
and column
-I
this
important to
bear
in
approximately 0.8%
residents.
The Table
first
mind
in
that
ASG
according to the
winning counties
after the
have
at
of the point
(3) lists the ratio
reports the effects in absolute terms.
in
column
In
viewing these
data the trend in population
announcement.
It is
panel reports the expenditure results.
The Bureau of Economic
in
is
it
by
Thus, these counties were gaining
straightforward to obtain per-capita effects
The 'Total" row suggests
=
-1
mean.
that the trend in total
The category can be divided
current expenditures, capital outlays, and payments to other governmental units.
was roughly equivalent
results,
increased
3.
expenditures increased by $17.2 million or 1.8% of the i
increase
that
(2) reports the
mean.
by subtracting 0.008 from the entries
The
are obtained
divided into separate panels for expenditures, revenues, and indebtedness and within
is
these broad categories separate results for subcategories are reported.
'"
governments fund them
For a series of local government tlnance variables, column (1) reports the estimated P and standard
error from equation (7).
least
is
important services, such as education and police protection. Table 7 empirically explores
In
into
percentage terms, the
the current and capital categories.
Analysis' National
Economic Accounts
reports the total stock
of private fixed assets
in
the manufacturing sector. This figure includes the value of structures, equipment, and software but excludes land.
Notably, structures only account for
42%
of this
total.
This
is
further evidence that the
$390 million
figure greatly
overstates the share of investment devoted to structures.
^ We
also tried to estimate (6') with a full set of state by year fixed effects.
demanding of the
This specification
is
excessively
and 92 counties, but only
7 of these states have multiple counties. The resulting estimates were so imprecise that they lacked meaning.
'*'
See Chapter 5 of Bartik (1991) for a review of the limited literature on the effect of economic development
policies on local property values. Also see Dye and Merriman (2000) for the effects of tax incremental financing
schemes that fund economic development projects on property values.
data,
because the property value data contain observations from 29
1
**
The region by year
fixed effects are not included in this version of (6').
27
states
The
education expenditures was 4.8% of the
in capital
money was
"selected sub-categories" results reveal where the extra
-1
spending per pupil may have increased, although
this
The
trend break
accounted for roughly
55% of the
this calculation is speculative
the trend break in housing and
Interestingly,
expenditures was larger than the overall trend break
publicly operated utilities
mean and
This finding and the current education expenditure results suggests that
increase in capital expenditures.
assumptions to be valid. ^^
=
i
targeted.
expenditures.
in
The
and requires a number of
community development
increase
in
consistent with the expansion of the manufacturing sector.
is
spending on
Notably, the
spending on police protection was unchanged but on a per capita basis
results suggest that
it
declined
modestly.
The second panel
1/3
of
reveals that the trend in revenues increased by $18.5 million.
was due
this increase
important source for the subsidies used to attract
tax revenues
compared
and
utility
comparison of the
revenues. This
in
required to balance their budgets annually.
in
of this finding,
it
The
break
nonguaranteed debt, which
final entry in the table
the government.
This type of debt
status, but
obligation to repay
the
new
new
It is
is
results also reveal trend breaks in local
larger trend break on utility expenditures
plants
may
receive discounts on utility fees.
that the trend break in expenditures
is
be expected since local governments are generally
to
notable though that the balance
initially surprising that the
is
million.
tax-exempt
The
was not achieved through
valuable public services.
In light
in
The
plants.
and second panels reveals
first
roughly equal to the trend break
cutbacks
new
charges of roughly $4 million.
to the charges trend break implies that the
A
Approximately
higher transfers from state governments, underscoring that states are an
to
it
it.**^
is
It
demonstrates that
a form of debt that
is
is
this
is
is
trend in long-term debt increased by $16.3
largely explained by the $12.8 million trend
not guaranteed by the full faith and credit of
generally issued under the authority of a government body to gain
used for private purposes and the government generally does not have any
seems sensible
to
conclude that
this is
one of the types of incentives given
to
plants.
VI. Interpretation
The
results
suggest that the successful attraction of a "Million Dollar" plant
associated with an increase
in
in propert>'
values.
Under the assumptions detailed
property values
may
be interpreted as an increase
For example, the
ASG
population variable
the
number of students
According
to the
in
ASG
is
in
in
Section
is
II,
on average
this increase
welfare for residents of the winner county.
a measure of
all
residents.
It is
not possible to separately determine
public schools.
manual,
this
type of bond has been used to fund industrial and commercial development,
pollution control, private hospital facilities, sports stadiums, and shopping malls (Census 1999, Section 9.3).
28
I'licre
tliat
are at least three dilTereiil explanations Ibr this result.
winning and losing counties aie ex-ante homogeneous
is
First if the identifying
and the mode!
valid
in
Section
assumption
II
is
correct,
then the results indicate that the effect of state subsidies dominates any malfeasance by politicians.
implication
to
is
that the increase in winners' property values
due
is
to the transfer
winning counties.
It is
wage
also possible that the
columns
bill results in
may be
justified
new
plant affects property values in the neighbors of winners (recall the
(2) and (3) of Table 5) and/or the entire
from an efficiency standpoint.
state.
has no empirical basis, then
A
we
second explanation
speculate that winning counties gain
that both
is
In this case, the counties
causes property values to increase
in
explanation
is
at
is
unavailable.
If this possibility
the expense of the rest of the state.
winning and losing counties systematically bid
benefit of attracting these plants.
this
In this case, the state subsidies
Unfortunately, this possibility cannot be tested directly,
because property value data on the neighbors of winners and entire states
of
The
of resources from states
winning counties,
do not compete away
relative to losing ones.
all
less than the
the rents and this
The unappealing
feature
is
not a readily apparent friction that would explain such systematic
that
winning and losing counties are not ex-ante homogeneous, which
that there
mistakes.
A
third explanation
is
would invalidate our identifying assumption.
unobserved heterogeneity
and
Z
in
Section
II).
in counties'
However, the
may
combined
effect
of
valuations of plants and plants' valuations of counties (the terms
V
In this case the findings
pre-announcement
similarity of the
reflect the
levels
and trends
in the
wage
bill,
property values, and observable characteristics of winning and losing counties, as well as the robustness
of the results
of region-year effects and limiting the sample to cases where both the
to the inclusion
winner and loser are located
in the
South, suggests that our research design
trends in property values in the 8 years before the
announcement
is
is
valid.
The
similarity in the
especially reassuring.
Under the
plausible assumption that property markets capitalize expected future economic activity, this finding
indicates that there
losing counties in
were not differences
in
expectations about future economic outcomes in winning and
advance of the announcement of the plant opening.
Another central issue of interpretation
and residents' welfare.
validity
assumes
As Section
II
is
whether there
is
a connection between property values
of the welfare interpretation
highlighted, the validity
rests
on the
of the Roback model's assumptions, some of which may not be justified. For example, the model
that there
is
no involuntary unemployment and moving
property values would solely reflect the value of land.
is
costless.
Further, the ideal measure of
Since the available measure of property values
include both the value of structures and land, the property value results overestimate the increase in
welfare for landowners. This
is
because they
reflect the value
ones, in addition to the change in the price of land.
29
of new structures, or improvements to old
Even
if
the
Roback assumptions
fail to
hold, this paper's results
still
provide
new
estimates of the
impact of successfully bidding for a large industrial plant on workers' earnings, property values and
government finances.
policy's efficacy and local
There are
fails to
shed any
"Minion Dollar"
widely believed that
It is
at least
three of these
all
outcomes are important measures of a
economic well-being.
four limitations to the paper's analysis that bear highlighting. First, the analysis
on the precise mechanism or channel by which counties benefit fi'om attracting a
light
plant.
For example,
we
cannot ascertain whether
tlie
increase in
employment
increases in demand, agglomeration economies, or other type of externalities. In future work,
this
same research design
to directly test the agglomeration theory by exploring
"Million Dollar Plant" lowers the production costs of other plants
in
the
is
we
due
to
will use
whether the opening of a
same
industry in winning
counties.
A
second limitation
is
that the external validity
plants in our sample likely differ
from the average
of the results
industrial plant in
suspect that these results do not generalize to the opening of any plant.
which the "Million Dollar' plants
differ
the exact selection rule that determines
is
from the average plant
which plants
is in
is
unknown.
many
In particular, the
82
important dimensions so
we
The most important dimension
their size.
in
Although we do not know
are featured in the "Million Dollar Plant" articles,
it
quite clear that large plants are the focus.
We
suspect that the 82 plants in our sample are a reasonably representative sample of
US.
large plant openings in the
that are
of interest to the public and policymakers, and that
our sample of large plants
However, the
A
results
is
new
to study in this paper.
In this respect,
the correct one for exploring the positive effects of winning a plant.
third important limitation
is
we seek
cannot inform the nonnative question of which
is
that in general
we
offered either by the winners or losers. This implies that
subsidies
all
precisely these large plants that generate the type of bidding wars
It is
t\'pes
of plants should be subsidized.
don't observe the magnitude of the subsidies
we cannot measure how much
an extra dollar of
worth. Moreover, the paper's results do not provide guidance on whether a local government
should increase or decrease
its
bid for a particular plant.
Fourth, the focus of this paper
is
on local communities. As we mentioned from the outset, the
question of whether states or the nation as a whole benefit from tax competitions between
governments
to attract
new
business
is
local
beyond the scope of this paper.
VI. Conclusions
This paper makes two contributions. Substantively,
it
provides credible estimates of the effect of
successfully bidding for an industrial plant, relative to bidding and losing, on local economies.
that the
announcement of a "Million Dollar"
plant opening
30
is
We
find
associated with a 1.5% increase in the trend
ill
earnings
announcement
wages
new
plant's
These
less precise.
due
crowd out
industry
winning counties
in
We
(relative to the period before).
other industries in the
in
activity
the
in
same county and
results provide
some of the
to the successful attraction
of a
to
ones)
losing
the
after
also tlnd evidence of positive trend breaks in total
in
first
large,
(relative
neighboring counties, although these estimates are
ex-post estimates of the increase
new
plant and indicate that the
in local
new
economic
activity doesn't
existing activity locally.
Under some assumptions, property values may provide a summary measure of
change
in
welfare, because the costs and benefits of attracting a plant should be capitalized into the price of land.
If
the winners and losers are
most
homogeneous, a simple model suggests
reliable data suggest that there
is
the net
any rents should be bid away. The
that
a positive, relative trend break of l.i-l.7% in property values.
Since winners and losers appear to have similar observable characteristics and similar expected future
may be
property values
in
advance of the opening announcement, the property value
explained by heterogeneity
in
subsidies from higher levels of government (e.g., states) and/or systematic
changes
in
underbidding.
results
Overall, the results undermine the popular view that the provision of local subsidies to
attract large industrial plants reduces local residents' welfare.
The
analysis further reveals that local
government finances
winning counties are not adversely
in
Local governments
affected by the successful bidding for a plant.
in
positive and roughly equal trend breaks in revenues and expenditures.
increase
in
winning counties experienced
Notably, there
is
a substantial
education spending.
The second contribution of
the paper
is
methodological.
The paper demonstrates
that
it
is
possible to use the revealed rankings of profit-maximizing agents to identify' a credible counterfactual. In
our particular case,
we show
that losing counties
but narrowly lost the competition
absence of the plant opening
(e.g.,
auctions) to
make
in
—may form a
—counties
that
have survived a long selection process,
valid counterfactual for what
would have happened
winning counties. The same methodology may be useful
causal inferences.
,
31
.
in
in the
other contexts
References
Timothy
Bartik,
Upjohn
Black,
J.
Institute for
1991. Who Benefits from State and Local Economic Development
Employment Research: Kalamazoo, Michigan.
Dan A. and William
H. Hoyt.
1989. "Bidding for Finns." The American
Policies?
Economic Review
.
W.E.
79(5):
1249-1256.
Brennan, Geoffrey, and James Buchanan.
Constitution
.
Cambridge University
Press:
1980.
New
The Power
to Tax: Analytical
Foundations of a Fiscal
York.
Dye, Richard F. and David F. Merriman. 2000. "The Effects of Tax Incremental Financing on Economic
Development." Journal of Urban Economics 47: 306-328.
Evans, William N. and Julie H. Topoleski.
2002.
"The Social and Economics Impact of Native
American Casinos." Working Paper No. 9198. Cambridge, MA: NBER.
Garcia-Mila, Teresa and Therese
J.
McGuire. 2001. "Tax Incentives and the City." Brookings-Wharton
J. Rothenberg Pack, Washington, DC: Brookings
Papers on Urban Affairs, edited by W. G. Gale and
Institution Press, 2002, pp. 95-1 14.
Glaeser,
Edward
L.
1932. Cambridge,
2001.
MA:
"The Economics of Location-Based Tax
Institute of Economic Research.
Incentives."
Discussion Paper No.
Harvard
Glaeser, Edward, Jose Scheinkman, and
Andre
of Cities. "Tbwnjfl/ oj Monetary Economics 36:
Schleifer.
1
1995.
"Economic Growth
Glaeser, Edward, Hedi Kallal, Jose Scheinkman and Andre Schleifer. 1992. "Growth
of Political Economy 100: 1126-1152.
Goolsbee, Austan and David B. Gross.
2000.
in a
Cross-Section
17-43.
"Estimating the
in Cities."
Journal
Form of Adjustment Costs with Data on
Heterogeneous Capital Goods." Mimeograph, University of Chicago.
Gordon, Roger H., and John D. Wilson. 1998. "Tax Structure and Government Behavior:
Agent Model of Government." Mimeograph. Michigan State University.
A
Principal-
Gyourko, Joseph and Joseph Tracy. 1991. "The Structure of Local Public Finance and the Quality of
Life." Journal of Political Economy 99(4): 774-806.
Hanson, Gordon H.
1998.
"Market Potential, Increasing Returns, and Geographic Concentration."
Working Paper No. 6429. Cambridge, MA: NBER.
Henderson, Vernon.
1982.
Journal of Urban Economics
"Evaluating Consumer Amenities and Interregional Welfare Differences."
1
1:
32-59.
Henderson, Vernon. 2003. "Marshall's Scale Economies." Journal of Urban Economics 53: 1-28.
Henderson. Vernon, Ari Kuncoro. and Matt Turner.
of Political Economy 103(5): 1067-1090.
32
1995.
"Industrial
Development
in Cities.""
Journal
Hochn, .loliii P., Mark C. Berger, and (llcnn C. Blomquisl. 1987. "A llccloiiic Model of
Wages, Rents and Amenity Values." Joiinial of Regional Science 27; 605-20.
Hsieh, Chang-Tai and Enrico Moretti.
Social
Waste
Mitol,
2001.
A.
Jenniter
"Can Free Entry be
2003.
Real Estate Industry." Joiirncil of Political
in the
"Boeing
is
Wind
Under
(forthcoming).
Wings."
Chicago's
http://more.abcnevvs.go.com/sections/business/dailynews/boeingO 05
1
Commissions and
Fnefficient? Fixed
Economy
II
Interregional
Available
at:
.html.
"Estimating the Social Return to Migher Education: Evidence from Longiludiiiai
and Repeated Cross-Sectional Data." Journal of Econometrics.
Moretti, Enrico. 2004a.
2004b. "Workers' Education, Spillovers and Productivity: Evidence from Plant-Level
Moretti, Enrico.
Production Functions", American Economic Review.
Oates, Wallace E.
1972.
Fiscal Federalism
.
Harcourt Brace Jovanovich:
Ramey, Valerie A. and Matthew D. Shapiro. 2001. "Displaced
Closings." Journal of Political Economy 109:958-992.
New
Capital:
A
York.
Study of Aerospace Plant
Rauch, James E. 1993. "Productivity Gains from Geographic Concentration of Fluman Capital: Evidence
from Cities." Journal of I hhan Economics 34(3): 380-400.
Roback, Jennifer 1982. "Wages, Rents and the Quality of Life", Journal of
1257-1278.
Political
Economy
1974.
"Hedonic Prices and Implicit Markets: Product Differentiation
Rosen, Sherwin.
Competition." Journal of Political Economy 82: 34-55.
Rosenbaum, Paul and Donald Rubin. 1983. "The Central Role of the Propensity Score
Studies for Causal Effects." Biometnka 70: 41-55.
S;7e5'e/ec?/o/;,
in
",
in
90(6),
Pure
Observational
Various Issues, 1982-93.
Standard and Poor's. October 25, 1993. Credit Week Municipal: The Authority on Credit Quality.
Tiebout, Charles
M.
1956. ""A Pure Theory of Local E.\penditures."' Journal of Political
Economy
64(5):
416-24.
Trogen, Paul.
2002.
"Which Economic Development
Policies
Work: Determinants of
State per Capita
Income." Mimeograph. East Tennessee State University.
Census Bureau.
1999. "Government
Mimeograph, Department of Commerce.
U.S.
U.S. Census Bureau.
and
Employment
Classification
Manual."
2003. "User Guide to Historical Data Base on Individual Government Finances."
Electronic File, Department of
Venable, Tim. 1992.
August: 630-1.
Finance
"BMW
Commerce.
Drives into South Carolina with $300 Million Auto Plant." Site Selection
33
Wilson John D. 1996. "The Tax Treatment of Imperfectly Mobile Firms: Rent-Seeking, Rent-Protection,
and Rent-Destruction." in The Political Economy of Trade Policy: Papers in Honor of Jagdish Bhagwati
edited by Robert C. Feenstra, Gene M. Grossman, and Douglas A. Irwin. MIT Press: Cambridge, MA.
,
Wilson, John D. 1999. "Theories of Tax Competition." National Tax Journal 52: 269-304.
1986.
"Pigou, Tiebout, Property Taxation, and the
Zodrow, George R. and Peter Mieszkovvski.
Underprovision of Local Public Goods." Journal of Urban Economics 19: 356-70.
34
DATA APPENDIX: Property Viiluc Data
There are a number of issues about the property value data that may affect the interpretation of
the paper's results. First, the linear interpolation procedure causes the true variability in the data to be
understated and biases the standard errors in the regressions downwards. Further, it may obscure the
timing of changes in property values associated with plant openings.
Second, the data measure the aggregate market value of
and
industrial land.'*'
an assessment rate
In
is
many
land and structures across residential
all
value differs from the market value but,
states, the assessed
available that allows for the conversion to market values.
It
in
these cases,
was not possible
to
A
obtain separate measures of the value of the land, excluding the structures, for our entire sample.
drawback of this measure of property values is that it will include the value of the new plant's structure,
which mechanically causes measured property values to increase.
This issue is discussed more
extensively in the subsequent description of the property value results, but
we
note here that the estimated
mean
increase in property values appears to be substantially larger than any reasonable estimate of the
value of new plants' buildings.
Third, there are important differences across states
important difference
it
is
that the frequency
is
in the statutes that
govern assessments.
of appraisals varies dramatically across
important to note that because the property assessment cycles are mandated by the
One
For our purposes
states.
state,
it
unlikely
is
of assessment is systematically correlated with plant openings. Appendix Table
reports
the distribution of state mandated property assessment cycles for the 31 states represented in the preferred
property value sample. 14, or almost half of the states, require that all properties are reassessed or
that the timing
1
revalued every year, but 12 states reassess less frequently than once every 3 years.
Connecticut allows 10 years between revaluations.
In
many of the
states that allow for
between reappraisals, 1/x of the properties are reassessed each year, where x
is
the
Interestingly,
more than a year
number of years
between mandated revaluations.
The relevance
our analysis
for
property values each year. This
is
is
that these data are not
designed to give accurate measures of
relevant because a model with rational, perfect-foresight agents
would
winning the plant should be capitalized into property values immediately
Thus, it would be appropriate to estimate this effect with a 1-time change in
light of the infrequent reappraisals in many states, a trend break model is more
predict that the net effect of
after the
mean
announcement.
property values.
In
appropriate and this model
A
related issue
is
the property value data.
is
discussed further
in the
It
is
difficult
years after
change
it is
Due
jumps
in
a spurious correlation between plant
Further, the longer the period between reappraisals, the
changes in property values associated with a plant opening in the first few
Consequently, we drop all observations from counties with at least one annual
to detect
opening.
its
in property
jumps could induce
possible that these
opening announcements and property values.
more
econometrics section.
that since the reappraisal dates are unobserved, there are likely to be
values greater than
to the passage
70%
or less than
of Proposition 13
in
-70%.
1978, the California data present a set of unique issues.
Proposition 13 limits the increase in the assessed value of a home to 2% per year, unless there is a transfer
of ownership or substantial improvements are made. As a consequence, it is difficult to detect positive
changes in property values in California. To the best of OLir knowledge, no other states place a cap on the
increase in assessed property values.^" In our preferred sample, we drop all California counties from the
sample.
We
also demonstrate that the results are robust to including these counties in the sample, with or
without California by year fixed effects.
States
have different names for the market value of land and
market value",
^^
Appendix
"full
D
The most frequently used ones
of the 1992 Census of Governments, Volume 2 Taxable Property Values, Number
Valuations for Local General Property Taxation
property.
structures.
are "fair
cash value", or "true cash value".
According to
this
lists
Appendix, California
is
1
Assessed
the state-specific requirements for periodic valuation
the only state that places a
values.
35
cap on the increase
in
of
real
property
For a different set of reasons, another state for which
property value data fully reflect market value
is
Indiana.
We
we
are not completely certain that the
re-estimated
all
Due
to these data challenges,
we
its
meaningfulness.
The
propeity. For our purposes, one limitation
is
OFHEO
price index
that the index
is
match with our pioperty value data is not
match, the correlation in the annual percentage change is 0.54.
counties, so that the
Finally
we
note that
when we
which the property value data are
that are qualitatively similar,
For example, the model
The
OFHEO
in
price index
is
column 3a
is
Table
MSAs
(OFHEO)
in
based on repeated sales of the same
available for metropolitan areas, but not for
perfect.
For those observations
tiiat
we can
estimate our property value models dropping the set of counties for
entirely
although
in
correlated our measure of property values for counties in
with an annual MSA-level produced by the Office of Federal Housing Enterprise Oversight
order to ascertain
models
the property value
excluding observations from Indiana counties and obtained estimates that are similar to the ones
from the Annual Survey of Governments,
less precise than the estimates reported in
yields an estimate of 283.0 (98.0).
available from vvww.ofheo.gov.
36
we
Table
6.
obtain estimates
Table
1:
"The Million Dollar Plant" Sample
(1)
Number of Observations
Winners
82
Losers
129
Distribution of the
Number of Loser
Counties per Winner
1
57
2
14
3
7
2
4
7
1
.
8
1
Distribution of Cases Across Industries
63
Manufacturing
Transportation and Public Utilities
8
Trade
4
Finance. Insurance and Real Estate
1
6
Services
Distribution of Year of
Announcement of Plant Location
1982
1983
5
3
,
1984
3
1985
6
1986
6
1987
8
1988
9
1989
7
1990
6
1991
12
1992
10
1993
7
Notes: The '"Million Dollar Plant" sample
is
derived
from various issues of Site Selection. See the
for
more
details.
text
Table
2:
County Characteristics by Winner Status
Winning
Losing
(l)-(2)
Counties
Counties
t-stat
(!)
(2)
Ail
US
Counties
[p-value]
( 1 )
Total
Wage
Bill in
New
Employment
in
New
Plant's IndustTy
Aggregate Property Values
(Smillions)
Trends
in
1127
1145
0.17
175
(1455)
[0.87]
(760)
40635
41568
0.13
6846
(54143)
17084
(49986)
19099
[0.88J
(26990)
(8773)
(10630)
[0.30]
in
total
Wage
Bill in
New
-
Percent Growth in
Employment
in
New
.023
.016
1.04
.018
(.089)
(.084)
[.30]
(.282)
Plant's
Industry-
.022
.011
1.57
.006
(.092)
(.089)
[.12]
(.283)
-
Plant's
Industry
Percent Growth in Property Values
.050
.055
-0.25
(.128)
(.092)
[0.80]
Socio-Economic and Democraphic Characteristics
Per Capita Income
Percent Growth
in
Per capita Income
Per Capita Total earnings
Percent Growth
in
Per Capita Total Earnings
Employment-Population Ratio
Percent Growth
in
Employment-Population Ratio
Per Capita Transfers
Population
Fraction of Population with High- School Degree
Fraction of Population with College Degree
Fraction of
Fraction of
(4)
-1.00
Outcome Variables
Percent Growtli
(3)
(4)
(1480)
Plant's Industry
(Smillions)
(2)
(3)
Levels of Outcome Variables
Pop
1
7 or
Younger
Pop 65 or Older
13660
15223
-2.70
11416
(3211)
(4250)
[0.008]
(2636)
0.014
0.011
-1.02
0.011
(0.010)
(0.019)
[0.31]
(0.057)
8993
I0I02
-1.78
7236
(4359)
0.013
(3730)
[0.08]
(2253)
0.015
-0.89
0.011
(0.028)
(0.025)
[0.37]
(0.124)
0.541
0.573
-1.40
0.468
(0.170)
(0.132)
[0.17]
(0.126)
O.OIO
0.011
-0.51
0.009
(0.017)
(0.015)
[0.60]
(0.042)
1770
1930
-1.76
2333
(628)
(554)
[0.080]
(1480)
342876
449280
-1.79
90139
(424939)
0.729
(346988)
0.762
[0.076]
-1.63
(400341)
0.695
(0.088)
(0.092)
[0.10]
(0.103)
0.197
0.238
-2.22
0.134
(0.074)
(0.089)
[0.02]
(0.063)
0.257
0.246
1.41
0.269
(0.037)
(0.027)
[0.16]
(0.035)
0.125
0.123
0.13
0.149
(0.051)
(0.029)
[0.89]
(0.043)
Geographic Distribution
Northeast
Midwest
0.093
0.250
-2.86
0.069
(0.292)
(0.392)
[0.005]
(0.254)
0.129
0.203
-1.43
0.340
(0.320)
(0.375)
[0.16]
(0.473)
South
West
(5) Industi'v Distribution
ol'tlit:
Maiuifacturiiig
I
Itililies
Wholesale
Retail
Finance, insurance. Real Estate
Services
Notes: The figure
plant opening.
for years
1982
Dollar cases
are a
in
in
The
to
0,391
3.88
0,449
(0.460)
fO.OOO]
(0,097)
0.127
0,152
-0.77
0,140
(0.323)
(0.348)
[0.44J
(0,347)
Labor Force
Construction
Trnnsportntion,
0.660
(0.474)
columns
1
0.067
0.059
1,79
0,050
(0.036)
(0.019)
[0,075]
(0,043)
0.268
0.222
2,30
0,236
(0.156)
(0.107)
f0,02]
(0,181)
0.052
0.059
-1.56
0.053
(0.031)
(0.022)
[0,12]
(0,044)
0,068
0.068
-0,09
0,065
(0.046)
(0.022)
[0,92]
(0,054)
0.217
0.216
0.17
0,263
(0.053)
(0,045)
[0,87]
(0,104)
0.059
0.074
-3.54
0.051
(0.026)
(0.030)
[0.001]
(0.035)
0.248
0.284
-2.99
0.246
(0.088)
(0.070)
[0,003]
(0.101)
and 2 are averages for the three years before the
figures in the top panel of
column 4
are a weighted average
number of Million
1993, with weights proportional to the
each year and industry. The figures
in
panels 2 to 5 of column 4
weighted average for years 1982 to 1993, with weights proportional to the
in each year (see bottom of Table
for the
number of Million Dollar cases
distribution
of cases across
years). All
1
monetary values are
in
1983
dollars.
.
Table
3:
The
Effect of Plant
Openings on the Wage
Status Relative to the Date of the Plant Location
Time
T
=
T
I
=
=
=
Losers
(1)
(2)
(3)
-320.4
-75.5
-244.8
(159.0)
[118.1)
-313.1
-38.5
(158.9)
118.0)
-8
-6
-253.3
-8.2
(158.8)
,118.0)
-5
28.5
118.0)
-191.7
I
i
=
=
=
-3
121.9
:iI8.1)
-109.0
156.7
(158.9)
^118.1)
-92.7
180.1
-l
=
-39.0
1=1
=
(
=
(
243.6
5
and
their standard errors
Business Patterns data.
column
(1)
and
in
(
-169.0
ttw;
(column
118.2)
(column 2) coefficients
(6) on the County
reports the difference between the
1)
and
tti t
from the estimation of equation
Column
(2) entries.
coefficients are plotted
-186.7
118.1)
412.6
(158.7)
Notes: The Table reports the
-196.0
118.1)
368.1
181.3
(158.6)
1
118.1)
295.1
99.1.
=4
-215.0
269.4
(
(158.7)
1
-215.0
118.1)
54.3
3
-242.1
222.2
(158.7)
1
-272.9
118.3)
7.7
2
-265.8
203.1
(158.7)
=
-258.4
118.6)
(159.0)
1
18.0)
-136.4
(158.7)
T
-258.1
66.3
J
(158.9)
-2
-274.6
-253.7
-225.1
-4
Difference
-245.0
(158.9)
(158.9)
T
by Year and Winner
Winners
T=-7
T
Bill,
Announcement
(3)
See the text for more
the top panel of Figure
1
details.
The normalized
—
'
«
y
pi
c ~
p
2 S
o
'-^
•,"1
It;
OJ
"-
^ (D ^
p > "0
Uw
00
D 5>
—
< Z
?
>s
-—
Cj
_C
OJ
53
z
p
-S
tAl
—
!>
CD
a>
^
:5
P C
"13
C C
n j-j
IJ
-o
>
c >
S
h
00
>-
rs OS
—
.
00
z
a3
<^
0-1
-P
Si
^
(U
S
o o
ON
>-
u
_
to
OJ
b =
<
r- 00
C3
— ^
E
Q. 5.
—
a
^4-
^
? p
<D
c
H
O
u
—
c
(D
(U
cb
OJ
Q.
E E
K3
.—
D^
5
H
c/]
CQ
-^ J^ T3
-O
C
H
S^SO
[—
OJ
eS
^A
(D
^^
>f-
CQ
OJ
e2
(D
_o
ID
UU
<^
'-^
"O
c
cr
—
o o p
—
oo
ro i-ri
r^ r^
i
~
c> v-i
VO ^o
rn
O
CJ
^^
-X
(U
p
CO
_aj
Xi
5
c/)
1)
ID
2
-z:
:-
o
2
,
lO
t^
ir^
s
t-~-
(^i
t^ lo
ON IT)
^O "O
iO r^
—
CN
ON CO
O
o
,L_
33
oo
22
p^
^
k:
^
OJ
<u
=
oo
lO
_-
—
CO
r-i
u 2
E 5i)
,
O
.2 ^>
t-.
<=>
OO
o
_
>-
;
E
O
dj
,
^
.
,
flj
t?
:^ _0
o ~
e E
o
U
•
— o
c
o
H
(U
P3
Di
_
3
c o
™
u >
CO
—
1)
—
j_.
ct:
U
<u
c
Cxi
o
X)
'~
ac:
c
<D
H
c c
rt
o
OU
P3
S "C!
>
< C
'.^
C^
c
Ki
cr Q.
(U
C Wl
o CO
c c
so a.
.£ 5
c
:~
:^
t/l
'-+—
fOJ
C>0
<u
Cd
>,
OJ
T3 "^
s C s
t^
<u
c
J;:J
<u
c
J=
>^
J^
C3
'
CO
C^
—
CD
r-l
>>-
5
aj
—
o
o
B
E
ts
^ o
o
p
r<i
Z
c
-a
_o
CO
^
m
,o
£
c
"ra
a
_aj
<u
Co
C/1
o
(D
C
ID
H~
_0J
—
-o
c
>-
ol oo
^
I?
*-'
(U
ca
h-
-5 •03
;>
P
•5
o
=
n
>
bo
-^
kS
" ~ o
^ C) -^
O
—
>
U
•^^
?
u
Q-
o
i-
AJ
p
lU
O
f"
CO
X
ra
>
5
r-
i^i
-a
E-
b-
"H
J
t—
u
-
c
ca
cc
'^
i^
> —c ^
O-
ta
—
bX)
O
eXiO
<
^
5
o
:^
J
Di
o
o
O
dI
x:
o
UU
>
<
ra
.s
Di
In
^ Z
—
u
>-
^
en
D
i-
Di
Bo
o
O
-a
CI.
::
<
UO
U-
-<-»
c3
:-
W)
- o
iji:
'5
^ H
Table
The Impact of Winner
7:
Status on
Government Expenditures, Revenues, and Indebtedness
Dependent Variables
(l)/(2)
P
(std. error)
Mean
(1)
(2)
(3)
17.2
966.8
0.018
823.5
0.018
143.3
0.019
35.0
0.029
31.0
0.048
290.7
0.015
50.4
-0.002
25.8
0.047
70.7
0.076
953.6
0.019
273.6
0.020
211.3
0.027
680.0
0.019
392.2
0.010
56.3
0.071
904.3
0.018
543.4
0.024
EXPENDITURES
Total
(4.7)
Current
14.5
(4.3)
Capital
2.7
(2.1)
Intergovernmental
1.0
(0.5)
Selected Sub-Cateaories
Education (Capital)
1.5
(0.5)
Education.
(Current)
Police Protection
4.3
(1.3)
-0.1
(0.3)
Housing
& Comm
Dev.
1.2
(0.6)
5.4
Utilities
(i.l)
REVENUES
Total
18.5
(5.0)
Intergovernmental
5.6
(2.0)
From State
Government
Total
Own
5.7
(1.6)
12.8
(3.5)
All Taxes
4.1
(1.7)
Utility
Charges
4.0
(0.8)
INDEBTEDNESS
Total Debt
16.3
(8.3)
Nonguaranteed Debt
12.8
(6.7)
Notes: The entries
from the
fitting
millions of dollars.
column
in
column
of equation
Column
N
The rows
of p and their heteroskedastic consistent standard errors
all of which are measured in
identify the dependent variables,
mean of the dependent variable in t-1 in winner counties and
of the point estimates from (1) to the mean. The data source is the Annual
= 4500. See the previous tables and the text for more details.
(3) presents the ratio
Survey of Governments.
(1) are estimates
(7).
(2) reports the
Appendix Table
Cycles by Slates
Number
Distribution of Real Property Assessment
I:
in
1991 and Siibsegiient Years.
Number of Slates
of Years
Between Revaluations
(1)
(2)
1
14
2
2
3
1
4
8
5
1
6
2
7
8
9
10
1
Unknown
2
Noles; The table reports the number of years between revaluations
for the
states
in
The
the
preferred
sample
for
the housing price
from this regression are reported in
columns (3a) and (3b) of Table 6. In some states, the assessment
regressions.
results
cycle varies within the
In these cases, the table reports the
state.
The "Unknown" categoi^ is
was not possible to determine the
shortest of the revaluation periods.
reserved for states where
assessment cycle.
D
it
The source
for the information in this table
is
of the 1992 Census of Governments, Volume 2
Taxable Property Values. Number 1 Assessed Valuations for
Local General Property Taxation.
Appendix
>
2 <
o'
E
X ^o
H
U -
^r
>.
<' _l
>-
u >
—
3=:
"r;
J 2
o o
'-c
-a
o
'-T
o
-r
c5
OT
ra
^^
^
^ E >
U
-J
u
.
c
—
_^'
<
=
o 6
^
Sj£mQ££3:uSS>
^
_i
o
D
< U
-
<
<a
;?_)_)
Is
<
_^
f= I ^
o
-I
53
"
£I
=
ra
o^-^
J
Ji-
-5
n
'S
;_,
o u 5
j= ^ = P
fcDiH-lHc^XUOQOi/i^oQU^eiiH
< o
>
X
< o
^^^^^
—
.iii
_l>_lES_lfS_l_l>_IJ
•5
-5
CQ
_"
<
—
bp'°
o
m
>
-5
Z I UQS
o o
>3^:
a.
o
U <
<
aJ <
u
>;
s
"I
=
o~^ 0--
^
o o-^
0000
CO
X
c
X
"-^
re"
?
s
f-
H
g
in
0)
z XX
re'
-a
lU
E c
o
re
a>
"re
"H
3
X > u a z
U
t
^
i K
^^
s c
_) —1
7:
«->
V,
-;-
5:
c/5
^
i u
1^ ^'^
in
«
^
^5
If ° 2
re
re
_1 _J
"u
J"
5
>
OQ
u
a>
u p
(U
|^
_1
:s
re
-^
^
c
u
§
i
p
on
.5
re
-;
-1 _J rs -J
C
re
C/)
f"
>
re
re
oj
<5
._
,_
^;5
-o
P.-\
u
PA
Westchester,
re
"3
?:
C/0
jj
^
c
c
i>
^
5 D
oj
S
:^ Di
,_
(L>
c
Montgonier\',
re
c
hp =
c
c
^
-3
^
-a
(^
re
H
s
^
^
c
=
c w s " c
"
&^5
i2?
_i
::S
"E.
g:
OJ
^
-p
a.
,3
'5
c
re
S
=
[3
5p
5p
<
>
2
CQ
5
CQ
t^
a-
^
"p
¥
1
"5
<
pa
Di
01
rn
>r,
w^
U-,
-a
"re
re
H
>
c
rPf*
5h
aft
1
I
Z
U
=
~re
[S
5
%
,0ft
(L)
_^
1
_
s
§•
CO
a.
P
3
^
=
re
Q
o
-^
1=^
1)
J
@c
X
5
1
>,
5
f-
s
'u
"3})
c;
"^
c
SO
^
fe
Si
cy
B
c
Z <
H 3 5 Q
UUQ
t.
U
^
_)
a.
(£P
c
C/1
y. •^
H
Delaware,
«
p K i " P
(D
Z
C/1
IE
Jj
re
Di
u <
c
X
^
t.
<u
N'S'
cS
>-
bb
c
-p
"0
X
CQ
,0
K s
3
to
U Z
vo
M^
^
*
•!l-
0^
^
<
<
<
-P
It;
"
z
s
g '>
=
re
c
=
!3 a!
^
re
re
re
< r X a
^
^
^
i
^^K
-H -a
i ^ p S c
^,3
•—
^
— c
"1
B
s
5~
'^
1-
Jr;
2n
|J
c
OJ
r:
J
So
c
re
"re
(L>
Q
^
1-
j-
J
—1
" -
re"
i
oj
re
S S
2
g^'?
t- k:
t.
^zxf
re
g
= ^
Lo 00
< X
i_
^
^
CQ
S
^
Che
i K 5
c „ c
^^ IJ "^,3'^ J ?
S'
^
V-
1-
t-
OJ
'G
c/1
c3
re
E
c
'-c
-J
^
F
22
-0
c
<
1
T3
i
-a
're
u
s
5
H
i)
_E
0.
a.
E
Z
U 2
On
Q
1
1 @
re
B
re
-D
^
LU
<
n
r^
m
"S.
-M
E
3
<
5
<
^
m
t^
r^i
-f
X3
—
X3
a
(2
CO
r^
c^
m
^
^
"-
IJ
c
^
,'3ft
(S
c
X
s
^
,_
^
y - c ^ C
c
^
c
[-L_
^
re
c
c
ir:
_)
C K
1
»
s.
c
1^ "^
<
bb
^
'-c:
— J J >
1
M
c
re
m2>5
^
_
-5
<
a
<
>
i_
2^
-J
1 <
5
^ £P
re
;§
" -2c •^y 3
QS
_1
<
<
re
VO
"
'S1-0
n
'*
-
J
<
u
g
<
> o < ^
O
Z
u
>-
U
>-
^"f_
ca
c
Q.
1
c
o
5
-C
P-,
bfl
T3
o
U
r~
5P
U
O
U
o
c5
Q ^ <
j;
b
"_>
O
o
i>
1-3
lU
CO
Qa
s
^ D ^
(-;
^
_1
1 ^
.22
c/)
>
fe
"
?:
OJ
J^
J
(U
CQ
op
o
u
Z U U
—
p o 7^
^^
E
o
J
^J
33
CrC
jj
'^
M 2^
^ 1 c
~
X
'S
tS
O
O
-
m
-^
•^
^ U ~
ss
a.
S
a3
o3
H P
P
o
^ Jo ~
;S J
Q
i>
J^ J
I)
1)
o
if
E
V,
1^ H
5
~
ugg^.ls
c/2
O
1 &
en'
O
00
(U
o E o
CQ
u < U Q Q <
53
fU
^
00 T3
^4
&i
e^-
:=
a
53
1>
IJ IJ
a o i> u
o o o o
(U
a>
;r
i>
JJ^J
=y
3
^
til)
4-
H
o
cH?
^
.5
CO
S5
1
o
=^
^
Z
5
r"^
_.
rl
r-
t~-
S
p
o
Q
1
_3
z
a.
c
'/"I
\o
r-
r~-
C/5
^
fe
1 ^
^
j_
<u
o
O
a>
^
-
_
^
._
(L)
^
—
o
oo
r^
GN
r-
>- _J
P J
O O
r^
P
^
UHJ
^ J
1;^
°r
Si.
1
h=
o
o
—
n
oo
E
:^'
X
IX
^
1-
a>
J
S
—
EP
,ab
C/5
^
^
—
'^
^
i-
oj
^
O
c
1
o
<
<
LO
D
6
ra
i i
Z
<
*~
<
<
5
C^
o
o
^
ly,
'1
^
.—
ii
r-~
t
;_
^ :_
~
^ i S
S
«
VO
o
>
O
«3
U
^
r^
;=
Oi
t
S
3 ^ <Tc 1>
>.U
u
° i
11 ^ o
t_
CO
53
<—
c/5
c/i
^
"5
CQ SI fc
i_
^
^ 1— ^ I— I- t=
= " S " " C
t<L>
5:
J
5^
S
00
c
a
,3
s
S
_
1
5
^
s
t<
u
M
a
oo
M3
VO
>
CO
5
<
-c
'5.
Q-
o
<
CQ
3
C/1
u:
en
^
o
VO
VO
t-^
VO
O
k:
«
PL-
It;
Jo
'^3
iZ
'~
o
_1
<y
H
,"2)
J
D
i
«
--0
2
ra
,M
u
S
_L
.2
-3
a-
1^ '^^ •l5 5^
2
-a
^
ra
ro
o
-a
~
r3
O
2
6. u
Q-
^5
P
H
< <U
z
>-
"o 'C
omX S
=<!
c
&0
S
i>
-a
s
<
>
>
<L>
>
Cj
5?
iz.
< <
< 5
>
4^12 g
V^ I- ^ t. i_ 1- ^ i- ^
QJ
(U
p
H" '^ ?:
CO
VI
o o o O O O ° > ° ^
_1
<L1
•S"
H
r•^
Q uo 2 d £ m
c/5
^
>;
1
> X
<O c o u < ^ o
> U .SP coZZ
E "1 -d
X
.°
g ^1 1
o
c75
"3
O
5^
'e
o
U
-=3
id
CQ
U
z
J
£
c/0
^
8
sp
>
m
c
o
53
r-
s
."SB
1
-a
ro
s
>
E
O -5
2 o
•S
1
U
o
1
a,
<:
<
M
ID
5
>
tC
3
,M>
o
@
00
^
"S
i
c^
Figure
1:
The
Effect of Plant
Opening on 1-Digit Industry Wage
Bill in
Winner and Loser
Counties
—
Winners
Losers
600
400
200
-200 -10
-5
year
-100 -
-200
-300
-10
-5
year
Notes:
Top
Panel: Conditional average wage
comparison, the scale of the average wage
bill in
bill in
winner and
loser counties.
To
facilitate the
winner counties has been adjusted to equal the
wage bill in loser counties at time t-1.
Bottom panel: difference in conditional average wage
bill in
winner and loser counties.
Figure
2:
Counties
-
The Effect of Plant Opening on
Winner and Loser Sample
1-Digit Industry
Winners
-e
Wage
Bill in
Winner and Loser
Losers
400 -
200
-200
-10
-100
-10
year
Notes:
Top
Panel: Conditional average
wage
comparison, the scale of the average wage
winner and loser counties. To facilitate the
winner counties has been adjusted to equal the
bill in
bill in
wage bill in loser counties at time t-L
Bottom panel: difference in conditional average wage
bill in
winner and loser counties.
Figure
3:
Industries,
The Effect of Plant Opening on Wage Bill
Same County - Winner and Loser Sample
in
Winner and Loser Counties
Winners
-
Other
Losers
1500
500
-500
-1500
-10
-5
year
1000
500
-500
-1000
-10
-5
year
Notes:
Top
Panel: Conditional average wage
comparison, the scale of the average wage
bill in
bill in
winner and loser counties. To
facilitate the
winner counties has been adjusted to equal the
wage bill in loser counties at time t-1.
Bottom panel: difference in conditional average wage
bill in
winner and loser counties.
Figure 4: The Effect of Plant Opening on Wage
Winner and Loser Sample
Bill
Winners
-
Same
Industry, .\eighl;oriiig Counties
—e-
Losers
1500 -
1000
500
-500
-10
year
400
200
-200
-400
-10
year
Notes:
Top
Panel: Conditional average wage
average wage
time
bill in
bill.
To
facilitate the
comparison, the scale of the
winner counties has been adjusted to equal the wage
bill in loser
counties at
t-1.
Bottom
panel: difference in conditional average
wage
bill
between winner and loser counties.
Figure 5: The Effect of Plant Opening on Wage
Winner and Loser sample
Bill
-
Other Industres, Neighljoring Counties
Winners
Losers
0-
-1000
-2000
-3000
-10
-5
year
4000-
3000
2000
1000
-10
-5
year
Notes:
Top
time
wage bill. To facilitate the comparison, the scale of the
winner counties has been adjusted to equal the wage bill in loser counties at
Panel: Conditional average
average wage
bill in
t-1.
Bottom
panel: difference in conditional average
wage
bill
between winner and loser counties.
,")
•'
. /''(-;,'
Figure
6:
The
Effect of Plant
1
Opening on Property Values
Winners
a-
in
Winner and Loser Counties
Losers
H
-5000
-10000 -
-15000
2000-
1000 -
-
-1000
-2000
year
Notes:
Top
Panel: Conditional average property value in winner and loser counties.
To
facilitate
the comparison, the scale of the conditional average property value in winner counties has been
adjusted to equal the conditional average property value in loser counties at time
Bottom
panel: difference in conditional average property value in winner
and
t-1.
loser counties.
Appendix Figure
1:
The
Effect of Plant Opening on 1-Digit Indtistry Employment
and Loser Counties - Winner and Loser Sample
—
Winners
in
Winner
Losers
25000
20000 -
15000
10000
5000
-2000
-4000
-6000
-8000
-10000
-12000 ^
-10
-5
year
Notes:
Top
Panel: Conditional average
employment
in
winner and loser counties. To
facilitate
the comparison, the scale of the conditional average employment in winner counties has been
adjusted to equal the conditional average employment in loser counties at time
Bottom
panel: difference in conditional average
employment
in
t-1.
winner and loser counties.
Date Due
3 9080 02618 0221