Microsoft Word - Tax-Paper

Tax Avoidance and Business Location in a State Border Model

Shawn Rohlin

Department of Economics

The University of Akron

Akron, OH 44325

Phone: (330) 972-7548

Email: [email protected]

Stuart S. Rosenthal

Maxwell Advisory Board Professor of Economics

Department of Economics and

Center for Policy Research

Syracuse University

Syracuse, NY 13244-1020

Phone: (315) 443-3809

Email: [email protected]

http://faculty.maxwell.syr.edu/rosenthal/

and

Amanda Ross

Department of Economics

West Virginia University

Morgantown, WV

Phone: (304) 293-7869

Email: amanda.ross @ mail.wvu.edu



June 6, 2012

Support for this project from the Ewing Marion Kauffman Foundation is gratefully acknowledged. We

thank Jan Brueckner and Gilles Duranton for helpful comments and also seminar participants at the

University of Houston and Rice University. All errors are our own.

Abstract

Previous studies have struggled to demonstrate that higher taxes deter business activity. We revisit this

issue by estimating the effect of cross-border differences in state tax conditions on the tendency for new

establishments to favor one side of a state border over the other. Identification is enhanced by several

features of the research design. We focus only on companies that locate within easy commuting distance

of state borders; we difference activity across adjacent segments of state borders and over time, and we

control for multiple sources of tax revenue. Of special importance given the border design, we highlight

the impact of previously overlooked state reciprocal agreements that require workers to pay income tax to

their state of residence as opposed to their state of employment.

Results indicate that reciprocal agreements have a pronounced effect on companies situated close to a

state border. In locations where reciprocal agreements are in force, higher personal income taxes lure

companies from across the border, while corporate income taxes and sales taxes have the opposite effect.

Where reciprocal agreements are not in place, the results are largely reversed. These patterns are

amplified in heavily developed locations, and differ in systematic and anticipated ways by industry and

corporate/non-corporate status of the establishment. Overall, the results support the view that

entrepreneurs are drawn towards tax sheltered locations, ceteris paribus.

Key Words: Tax deterrence; business location; border models

JEL Codes: H2, H7, R3, R5

The avoidance of taxes is the only intellectual pursuit that carries any reward – John Maynard

Keynes, Attributed, A Dictionary of Scientific Quotations

(1977), Alan L. MacKay, p.140.

I. Introduction

There is truth in Benjamin Franklin’s famous quote, “The only thing certain in life is death and

taxes.” But for local policy makers, the quote above from John Keynes demands more attention. That is

because one way that entrepreneurs and households may seek to avoid taxation is by relocating to more

tax advantaged locations. This idea was not lost on Tiebout (1956) and Hamilton (1976) in their seminal

papers on the possibility that households may vote with their feet. It has also been the focus of numerous

papers that have sought to measure the impact of local tax policy on business location decisions. A

notable feature of those studies, however, is their lack of consensus as to whether local taxes discourage

business, and the absence of a general structure that accounts for the mixed patterns of results.

This

paper revisits these issues. We offer several innovations that help to clarify the mixed results in the

literature, and which confirm the potential for tax deterrent effects.

Throughout, we focus only on companies that locate within easy commuting distance of state

borders. We then difference activity across adjacent segments of state borders and over time. This helps

to control for unobserved factors that might be correlated with our tax measures. Given this border

design, of special importance, we also highlight the impact of previously overlooked state reciprocal

agreements that require workers to pay income tax to their state of residence as opposed to their state of

employment. As will become apparent, these agreements provide a powerful source of identification.

We also control for multiple sources of tax revenue, and show that tax deterrent effects differ depending

on the type of tax and the type of industry in question. Additional models allow for the possibility that

Early literature on the impact of taxes on businesses location typically failed to find evidence of a notable deterrent

effect, and in some cases even found that higher taxes “attract” businesses (Carlton (1979, 1983) and Schmenner

(1978, 1982)). This began to change in the late 1980s and 1990s as studies by Bartik (1985, 1994), Papke (1991),

Hines (1996), and others offered evidence that higher taxes do deter businesses. However, most of the estimates

were noisy or small relative to the effect of other policies (see Wasylenko (1997) for a review).

tax deterrent effects differ in heavily versus lightly developed locations. Finally, in all of our models we

control for the size of state government.

Our emphasis throughout the paper is on the sorting of new business activity into opposite sides

of a state border in response to cross-border differences in state tax conditions. This is consistent with a

focus on tax avoidance as in the quote above. We conduct our analysis using establishment-level data for

roughly 96,000 newly created enterprises in 2002 and 2005. As noted, all of these establishments are

within easy commuting distance of a state border. We then evaluate the impact of cross-border

differences in state tax conditions on the side of the border chosen by the business owner. Identification

is enhanced by breaking state border regions into twenty mile long segments. This allows us to use

border-segment fixed effects to control for unobserved time-invariant local (and state-wide) attributes

associated with a given border segment. It also allows us to control for observable time-varying local

attributes, such as the local level of agglomeration that may amplify or mitigate tax effects. As will

become apparent, our research design yields revealing estimates of the deterrent effects of cross-border

differences in state tax conditions.

As implied above, we decompose tax conditions into two parts. In the first part, we consider the

extent to which the size of local government expenditures (per capita) attracts or deters new business

activity. We do this because local governments must balance their budgets, and for that reason, a

decrease in one tax may require an increase in an alternate tax if government expenditures are to be held

constant.

This suggests that size of government and the mix of tax measures are correlated, and that

estimates of tax deterrent effects may be sensitive to controls for size of government. Nevertheless, most

prior studies of tax deterrent effects do not control for size of government. While in principle that

omission could account for some of the discrepancies in the literature, our results suggest that that may

A cost of our approach is that we take as given the decision of the entrepreneur to create a new establishment in a

given border segment. Our results, therefore, may not generalize to deterrent affects across broad regions.

If state government offers net value to the business community for an additional tax dollar, then larger state

government should attract business; if the reverse is true, then larger state government will discourage business

activity. This is analogous to early arguments by Brueckner (1979, 1981) who considered the impact of local

property taxes on aggregate property values.

not be the case, at least for border-design studies such as ours. In our more simply specified models,

smaller state government does appear to attract business arrivals from the opposing side of the border.

But in models that take location fixed effects into account, these patterns largely disappear and controls

for size of state government have little effect on estimates of tax deterrent effects.

Bearing the above in mind, our primary emphasis is on the manner in which state government

expenditures are financed. For this portion of the analysis we include separate controls for the corporate

income tax, the personal income tax, and the sales tax. These three sources of tax revenue account for an

important share of state government revenues and are often at the center of policy debates about the level

and composition of local taxes.

There is also good reason to expect that industries differ in their

sensitivity to the different types of taxation and related state-specific policies.

When considering the sales tax, it is important to recognize that manufacturing, much of the

service sector, and a notable segment of other non-retail industries are not subject to a sales tax. Holding

constant government expenditures, companies in these industries should be attracted to states that raise

more of their revenue through the sales tax. Among retailers, the influence of cross-border differences in

state sales taxes likely depends on their location relative to the state border. For locations well into the

interior of a state, demand for in-state retail services is likely quite inelastic.

For these locations, higher

state sales taxes will tend to be passed on to consumers with little impact on retail activity. For

companies operating close to a state border, however, the ease of cross-border shopping suggests that

demand for in-state retail services is likely elastic relative to cross-border differences in sales tax rates.

This suggests that close to the border, retail establishments will be drawn to the low sales tax state.

Companies that are currently incorporated or anticipate becoming so are likely more sensitive to

the corporate income tax. As seen in Table 1b, the share of newly established companies that begin as

corporations differs by industry (e.g. manufacturing, wholesale trade, services) and accounts for a bit less

Summary measures that characterize the distribution of revenues and tax rates across states for each of the three

types of taxes are provided In Table 1a and are discussed later in the paper.

Rosenthal and Ross (2010) provides support for this view for the retail sector. They show that at the metropolitan

level, population size is nearly a perfect predictor of retail sector employment, consistent with inelastic demand for

retail services overall.

than half of newly established companies in the United States (the rest are established as sole

proprietorships and partnerships). The differing rates of incorporation across industries may contribute to

differences in deterrent effects arising from state corporate income tax rates and related policies.

All companies rely on labor, including hired workers and the labor effort of the business

owner(s). But states differ markedly in their personal income tax rates. With sufficiently mobile

workers, employers would bear much of the burden of higher personal income taxes which would then be

capitalized into higher nominal wages. Such an outcome would suggest that higher personal income taxes

would deter business activity across a broad range of industries. While that may be true, two

considerations suggest that such a conclusion could be hasty. The first is that workers may not be

sufficiently mobile to push the burden of higher personal income taxes onto employers, at least in the

short run. The second is that a number of pairs of adjacent states have reciprocal agreements that require

workers to pay personal income tax to the state in which they live as opposed to the state in which they

work. A complete listing of states with such agreements is provided in Table 1c. For companies locating

within easy commuting distance of the state border, the presence of a reciprocal agreement should

mitigate the deterrent effect of the state personal income tax since workers can choose to live on the tax

advantaged side of the border while working on the other. Failure to account for reciprocal agreements

has the potential to obscure the deterrent effect of the personal income tax and other forms of taxation as

will be clear later in the paper.

A further feature of our research design is that we allow tax deterrent effects to vary with the

scale of local development. Only a handful of studies have considered this possibility. For at least two

reasons, however, it is possible that state government tax effects could differ with the local level of

development. The first is that agglomeration economies in heavily developed areas (e.g. Duranton et al

(2004), Rosenthal and Strange (2004), Glaeser and Gottlieb (2009), Combes et al (2010)) may swamp any

influence of state government expenditure and tax policies: this would argue for more pronounced

deterrent effects in lightly developed areas. On the other hand, heavily developed areas typically have

higher nominal wages, consistent with a greater degree of labor productivity (e.g. Glaeser and Mare

(2001), Rosenthal and Strange (2008)). With a progressive state income tax code, this would amplify the

deterrent effect of the personal income tax in heavily developed locations. These arguments suggest that

the relationship between agglomeration and tax-deterrent effects is ambiguous, a priori.

Previous studies have also been plagued by concerns that local taxes may be endogenous to local

business activity because of unobserved factors that drive both tax policy and local economic growth.

This could arise, for example, if communities that anticipate more robust growth raise taxes to finance an

anticipated increase in demand for services. Several features of our research design help to mitigate such

concerns. First, we focus on economic activity very close to the state border. To the extent that border

regions are small relative to their states, this will reduce the tendency for states to set policy based on

anticipated economic growth in the border regions. Second, we lag our tax measures by two years so that

they are predetermined relative to when businesses in our data are established. Third, and most

important, we difference activity across adjacent 20-mile long segments on opposite sides of a state

border, and also between two time periods, 2002 and 2005. This reduces the influence of local (border-

segment) time invariant cross-border attributes that might be correlated with changes in cross-border

differences in state tax conditions. Fourth, as suggested above, conceptual arguments suggest clear priors

regarding the influence of reciprocal agreements across state pairs as well as the influence of an

establishment’s industry and corporate/non-corporate status. Those priors provide testable hypotheses

that also help to illuminate tax deterrent effects.

Before proceeding, it will be useful to provide a brief overview of the manner in which

geographic features of the data are specified (further details are provided later in the paper). Throughout,

we use establishment-level data for which location is known down to the zipcode level. We focus

primarily on activity in zipcodes that lie at least partly within ten miles of a state border (results based on

zipcodes that extend to within one mile of the state border are very similar and are mostly not reported for

that reason). State borders are broken into approximately 20-mile long segments or “wedges,” and

zipcodes on opposite side of a state border and which are situated along the same 20-mile wedge are

matched and are said to belong to the same “wedge-pair.” Existing employment and arrivals of new

businesses in the matched zipcodes are measured using 2-digit SIC industry data from the Dun and

Bradstreet Marketplace files for the third quarters of 2002 and 2005. This produces a data file with

roughly 96,000 newly established companies, each of which is treated as a separate observation, and each

is geocoded to a single wedge-pair that identifies its location along a state border.

Three different types of models are estimated using these data. The simplest is estimated by

ordinary least squares and relies on the single differencing of tax conditions across the state border. A

second model adds state-pair fixed effects and identifies tax effects based on changes in cross-border

differences in tax conditions over time. A more robust model uses wedge-pair fixed effects that

difference away time-invariant unobserved factors at the very local level. Both of these latter models

implicitly difference the data twice, across the state border and across time periods.

Our double differencing approach is in the same spirit as recent border studies by Holmes (1998)

and Cunningham (2007). These studies considered the impact of state-level right-to-work laws and local

land use regulation, respectively.

Other even more recent studies by Duranton, Gobillon, and Overman

(2011) and Rathelot and Sillard (2008) have used border models to consider the effect of property taxes

and local corporate income taxes, respectively.

We also build off of three recent papers by Devereux,

Griffith, and Simpson (2007), Jofre-Monseny and Sole-Olle (2008), and Brulhart, Jametti, and

Schmidheiny (2009). These papers have considered the possibility that agglomeration may mitigate the

deterrent effects of business taxes and/or subsidies.

Devereux, Griffith, and Simpson (2007), for

example, found that government subsidies have less impact on a firm’s location decision in more highly

developed areas. Our work is further motivated by recent studies in the agglomeration literature which

show that nearby activity (even within one mile) matters much more to many entrepreneurs than activity

Holmes (1998) was among the first to use border methods to analyze the impact of local of public policies. He

found that states with right-to-work laws in place – which give workers the right to not join the union – enjoyed

notably higher manufacturing employment growth since the 1940s.

Both Duranton, Gobillon, and Overman (2009) and Rathelot and Sillard (2008) instrument for local tax measures

using local political variables such as the share of voters belonging to a more conservative political party. Both

studies find that higher taxes negatively affect growth of existing businesses although Duranton et al (2009) do not

find an effect on the creation of new businesses.

See also Greenstone and Moretti (2004) and Greenstone, Hornbeck and Moretti (2010) for related work.

outside of the immediate area (e.g. Rosenthal and Strange (2003, 2005), Arzaghi and Henderson (2008)).

Our focus on activity very close to the state border echoes this literature.

Taken as a whole, our results provide clear evidence that state-level tax policies do affect the

location decisions of entrepreneurs and new business activity, but not in a way that lends itself to a one-

size-fits-all summary. As a broad characterization, entrepreneurs and new business activity are drawn to

those locations in which they are relatively more sheltered from the cost of financing local government

activities. In addition, previously overlooked reciprocal agreements that govern whether individuals pay

personal income tax to their state of residence or their state of employment have a pronounced effect on

the location patterns of companies situated along state borders.

The following section presents a simple conceptual model that motivates and guides our analysis.

Section 3 lays out the empirical approach. Section 4 describes the data including the geocoding

procedures used to match zipcodes across state borders. Section 5 presents results and Section 6

concludes.

II. Conceptual motivation

2.1 Overview

This section highlights three key conceptual points that are relevant to the empirical work to

follow. The first is that if all establishments were alike, then cross-border differences in tax conditions

would be fully capitalized into land values. In this instance, tax differentials would have no effect on the

side of a state border on which a business would choose to locate. The second is that with heterogeneous

firm-types, cross-border differences in tax conditions are only partially capitalized into land values.

Under those conditions, tax differentials do affect where different types of companies locate. The third is

that even if a given type of firm does not pay a particular tax, its location decision will be sensitive to the

An additional related literature includes tax competition studies that examine the propensity for jurisdictions to

offer tax incentives as a way of attracting new business activity (e.g. Brueckner and Saavedra (2001), Hines (2006),

Chirinko, and Wilson (2008), to name just a few). Such policies are based on the presumption that entrepreneurs are

drawn to tax advantaged locations.

tax if it competes for space with companies that are subject to the tax in question. We highlight each of

these points below.

2.2 One sector bidding for land

Suppose first that there is only one bidder for land which we will refer to as the business sector,

land markets are competitive, and all firms are identical. Firms are price takers and sell their product for

P. Output is produced using one unit of land and public goods that are provided by the state government

(S), including roads, infrastructure, and other services. All land is owned by absentee investors.

If firms are not taxed they still receive services from S given its public good nature, and the firm’s

profit function is given by,

(, ) (, ) (, ) (, )ud P Aud Sud Rud





   (2.1)

where d = {1,2} for side 1 or side 2 of the border, u is the distance to the border, A(u,d) are attributes at a

given location, and R(u,d) is the cost of land. More valuable attributes enhance productivity (θ

> 0), but

we impose no restriction regarding the manner in which A changes with distance and direction from the

border. Local government services S are state-specific and change in a discrete fashion upon crossing the

state border. For that reason, S is sensitive to d. With competitive markets, profits are driven to zero and

the firm’s bid-rent is given by,

(, ) (, ) (, )

ud P Aud Sud



  (2.2)

Suppose now that the state on side 1 of the border imposes a tax on firms per unit output (T),

which is used to help finance the given level of local government services. We assume that all companies

located on side 1 of the border are subject to the tax but that companies on side 2 do not pay the tax.

Maintaining the zero profit condition, the bid-rents on sides 1 and 2 of the border are given as below,

123

( ,1) ( ,1) ( ,1) (1)

uPAu Su T





   (2.3a)

( , 2) ( , 2) ( , 2)

uPAu Su



  (2.3b)

In Figure 1, we display the bid-rents for land on either side of the state border before and after

side 1 imposes its tax. To simplify the exposition, we assume that productive attributes are increasing

along a trend line as one moves towards the interior of State 2. For that reason, bid-rent is drawn as

upward sloping, but we emphasize that the key points to follow are not sensitive to that assumption.

With T

set equal to zero, bid-rent is given by segment ab . Following the imposition of the side-

1 tax, the bid-rent on side 1 shifts down by an amount equal to T

and the bid-rent function is given by

cdeb . Implicitly, we are assuming that land is inelastically supplied to firms. Side-1 landowners

therefore absorb the entire burden of the tax (T) and the equilibrium land rent function is given by

cdeb .

Importantly, in this very simple model, side 1 taxes do not affect the spatial distribution of business

activity on either side of the border – landowners absorb the full cost of the tax.

2.3 Two sectors bidding for land

Suppose now that there are two types of companies bidding for land, type-I and type-II. Because

the two sectors have different production functions, their valuation of local attributes differs, and this

causes their bid rent functions to differ as well. The bid-rent functions for the two sectors are drawn in

Figure 2. In the absence of a tax, type-II companies outbid type-I firms for space to the right of point j

while type-I firms are the high bidders to the left of point j. With competitive markets, land goes to the

highest bidder and equilibrium land rents are given by the upper envelope of the bid-rent functions,

agk .

In this example, type-II companies occupy land to the right of

j while type-I companies are found to the

left of

Suppose now that a tax

T is imposed on side 1 of the border, but the tax applies only to type-I

firms. With the downward shift in type-I bid-rent on side 1 of the border, the equilibrium land rent

function is given by the new upper envelope of the bid-rent functions,

chnegk . As drawn in the figure,

type-II firms outbid type-I firms in all locations to the right of point

j as before, but also for land between

points m and i which was formerly occupied by type-I firms. Type-I firms occupy land to the left of point

i (as before), and also a segment of side-2 between points m and j.

This simple model has several implications for the effect of state-level taxes on conditions close

to the state border. First, competition for space between multiple sectors bidding for land mitigates the

degree to which taxes are capitalized into lower equilibrium land rents. This is illustrated by the fact that

segment

hd lies below segment hn : competition between sectors reduces the extent to which taxes are

capitalized into lower equilibrium land rents in locations where tax sheltered industries emerge as the

high bidder for space.

Second, cross-border differences in tax conditions affect the equilibrium locations for both the

industry subject to the tax and the tax sheltered industry. This arises from the sorting equilibrium when

land is allocated to the highest bidder. It also suggests that in the empirical work to follow, it is important

to consider the impact of different types of tax measures on the location patterns of all industries, even

when an industry is not subject to a given tax (as with manufacturing and the sales tax, for example).

III. Empirical Model

As emphasized earlier, the primary goal in the empirical work is to estimate the impact of cross-

state border differences in tax conditions on the tendency of newly established companies to favor one

side of the border over the other. Throughout, we work with establishment-level records and restrict our

sample to newly created enterprises in 2002 and 2005 that are situated within easy commuting distance of

a state border.

We begin with the following expression for the likelihood that a given entrepreneur would choose

side 2 of a state border over side 1:

11 2 21 2 3 1, 2,

I( )( )( )

it t t t t w t w t

SS TT



 . (3.1)

In this expression, I

equals 1 if entrepreneur i chooses side 2 of the state border and 0 if side 1. The

terms

– S

and T

– T

are the cross-border differences in state-level government expenditures and

tax conditions corresponding to the border along which establishment i is located. The term

1, 2,wt wt



represents the cross-border difference in all other attributes that might affect the profitability

of choosing side 1 versus side 2 of the border. The subscript

w denotes the segment (or “wedge” as

described in the Introduction) along the border in which a company is located. The subscript

t represents

the time period in which the company makes its location choice (third quarter of 2002 or third quarter

2005).

For a given set of measures for

– S

and T

– T

(which are described in the following

section), our primary challenge is to adequately control for the elements of

1, 2,wt wt





so as to ensure

that we obtain consistent estimates of the primary parameters of interest, θ

and θ

. We proceed by

splitting

31, 2,

()

wt wt



 into time-invariant (Ω

) and time-varying components (e

). Substituting into

(3.1) we obtain,

11 2 21 2

I( )( )

tt ttit w wt

SS TT e







 . (3.2)

Given the specification in (3.2), it is apparent that we can control for Ω

using border-segment (“wedge-

pair”) fixed effects given that we have two periods of data. In principle, the remaining term, e

, could be

correlated with the tax measures that are the focus of the study, but for reasons described in the

Introduction we believe that such concerns are modest. We estimate (3.2) using a linear probability

model.

IV. Data and Summary Statistics

4.1 State tax and expenditure conditions

The model outlined above requires measures of state-level tax conditions in each period, t. We

address this as follows. State expenditure and population data were obtained from the U.S. Census

Bureau website.

These data were used to compute state-level government expenditures per capita for

each of the sample years, 2002 and 2005, and for each of the states in the continental U.S. Data on state

See http://www.census.gov/govs/state/historical_data.html for links to the data.

tax rates were obtained from the Tax Foundation website while data for state revenues raised through a

given tax were obtained from the U.S. Census Bureau website.

As noted earlier, we focus on three prominent sources of state tax revenue: the corporate income

tax, the personal income tax, and the sales tax. Panel A of Table 1a displays the share of state

government expenditures financed through these three sources of revenue for both 2002 and 2005. It is

noteworthy that the personal income tax and the sales tax both account for roughly 13 to 14 percent of

state budgets while the corporate income tax contributes a much smaller share, just 2.5 percent in 2005.

Together, these three sources of revenue comprise roughly 30 percent of state budgets.

Panel B of Table 1a also demonstrates that there is considerable variation in tax rates across

states. For 2000 and 2003, for the sales tax rate, the maximum personal income tax rate, and the

maximum corporate income tax rate, the panel reports the median and standard deviation across states, in

addition to the number of states for which the tax is not imposed. Notice that for each tax measure, the

standard deviation is roughly half the size of the median, indicating that there is quite a bit of variation

across states. In addition, for each tax, between four and six states do not levy the tax in question.

We measure tax conditions using the state tax rates in panel B of Table 1a. Specifically, we use

each state’s maximum corporate income tax rate, the maximum personal income tax rate, and the sales tax

rate.

Identification of tax deterrent effects then requires that changes in cross-border differences in tax

rates are exogenous to the decision of an entrepreneur to choose one side of the state border over the

other. While we cannot provide direct empirical evidence to support that assumption, the various features

of our research design discussed earlier help to mitigate concerns about endogeneity. These include

For tax rates see, http://www.taxfoundation.org/taxdata/show/230.html. For tax revenues see: U.S. Census Bureau,

State & Local Government Finance, Historical Data: 2005, the following URLs:

http://www.census.gov/govs/estimate/historical_data_2005.html; http://www.census.gov/govs/state/historical_data.html.

Most of the remaining portion of state budgets is derived from grants from the federal government (roughly 25

percent), all other forms of state taxes and licensing fees (roughly 20 percent), and insurance trust revenue for

government retirement and social insurance programs including contributions by state government workers and net

earnings on fund investments (roughly 25 percent). See: U.S. Census Bureau, State & Local Government Finance,

Historical Data: 2005, http://www.census.gov/govs/estimate/historical_data_2005.html

Most states specify progressive tax schedules for the corporate and personal income tax rates, in addition to a

variety of exemptions, deductions, and other allowances that affect the effective tax rate faced by a given business

owner. We use the maximum rates and in doing so, implicitly assume that the maximum rates are correlated with

the rates that business owners expect to pay allowing for all other features of a state’s tax code.

lagging the tax measures in Panel B of Table 1a by two years relative to which companies in our data are

established, the focus on activity very close to the border, and the double differencing strategy in which

activity is differenced both across the border and over time.

A second condition required for identification is that there must be sufficient numbers of adjacent

states for which one or both of the adjacent states changed their tax rates between 2000 and 2003. Table

1d provides evidence on this point. Observe that between 2000 and 2003, six states changed their

maximum corporate income tax rate, 11 states changed their maximum personal income tax rate, and five

states changed their sales tax rate. Because most states border on multiple states, the number of instances

in which cross-border differences in tax rates changed is much higher: 24 for the corporate income tax

rate, 35 for the personal income tax rate, and 23 for the sales tax rate. Evidence presented later in the

paper suggests that this level of variation is sufficient to identify tax deterrent effects.

A third condition necessary for identification is that the tax rates used in the analysis must be

strong correlates with the unobserved effective tax rates that entrepreneur expect to pay. Those effective

tax rates depend on the statutory tax rate, including the full progressive schedule of income tax rates,

corporate income apportionment formulas (e.g. Goolsbee and Maydew (2000)), and various other state

and industry-specific provisions that reduce a company’s exposure to a given type of tax (e.g. Slemrod

(2004), Grahm (1996)). In this respect, we recognize that the tax rates specified are approximations of the

effective tax rates that entrepreneurs expect to pay.

We also estimated all of our models using tax revenue shares to characterize state tax conditions. This

was done by dividing gross revenue from each of the three tax instruments (corporate income, personal

income, and sales tax) by the level of state government expenditures. This has the appeal of allowing for

the full schedule of tax rates and related state-specific provisions that affect the effective tax rates faced

by business owners. Several studies in the tax literature have used revenue shares for these reasons,

including studies of the impact of cross-country differences in tax conditions for which specification of

the full tax code would be difficult (e.g. Slemrod (2004)). However, for other reasons we favored the tax

rate approach highlighted above. First, it is the tax rate – not the tax revenue share – that enters into a

company’s cost function. Second, coefficients on the revenue share variables must be interpreted relative

to the omitted revenue category, which in our case is all sources of state revenue apart from the three

highlighted tax measures. Third, interpretation of the revenue share coefficients also depends on the

position of a state along its Laffer curve. For governments operating on the upward sloping portion of

their Laffer curve (as is likely the norm (e.g. Uhlig and Trabandt (2011)), rising tax revenue shares are

indicative of increasing effective tax rates, but the reverse is true for governments on the downward

4.2

Matching business activity across state borders

A key feature of our empirical strategy is to match business activity along opposite sides of a

state border, restricting our focus to just those locations within easy commuting distance of the border.

As discussed in the following subsection, the geographic location of new establishments in our data is

reported down to the zipcode level. Both zipcodes and state borders are of irregular shape, and this

complicates efforts to match adjacent business activity on opposite sides of a state border. To facilitate,

we first use geographic information system (GIS) software to create 1 and 10 mile buffer zones on each

side of all of the state borders in the United States. In addition, state borders are broken into segments by

laying down a 20 by 20 mile grid across the continental United States. Only grid squares that intersect

state borders are retained (see Figure 3), and each is divided into pieces by the intersecting state borders.

Each piece is then referred to as a “wedge” and pairs of wedges on opposite sides of a state border that

belong to the same grid square are referred to as a “wedge-pair.” Figure 4 illustrates using a snapshot of

the border region of Nebraska, Kansas, Missouri, and Iowa.

We next overlay a zipcode map for the United States on top of the grid squares, and retain only

those zipcodes that intersect or lie entirely within one of the border wedges. Each zipcode is then

assigned to the wedge that it most overlaps. Note that multiple zipcodes could be assigned to a single

wedge. Business activity associated with each border wedge is determined based on all zipcodes assigned

to that wedge using activity throughout the entirety of each of the assigned zipcodes. Using this

procedure we match business activity in zipcodes on opposite sides of a state border that are situated

roughly along a common border segment that is 20 miles in straight-line length.

sloping portion of their Laffer curves. Fourth, changes in cross-border differences in tax revenue shares

are sensitive to changes in state-wide economic activity (e.g. aggregate sales and aggregate income) that

may mask the impact of changes in the underlying tax rates. For these reasons, we feel that the tax rate

rather than tax revenue approach is more robust. Accordingly, we emphasize the former in the text, but

present estimates from the revenue share models in the appendix for review.

4.3 Dun and Bradstreet data and summary measures

Data on business activity for the analysis were obtained from the Dun and Bradstreet (D&B)

Marketplace files for the third quarters of 2002 and 2005. The data provide information on different types

of establishments aggregated to the zipcode level. Using these data, we measure counts of existing and

newly created (in the previous 12 months) establishments and their corresponding employment for

different 2-digit SIC (Standard Industrial Classification) industries.

Although we obtain our data aggregated to the zipcode level, we are able to convert the data into

establishment level observations. That is because all of the key control measures in our model are state-

or wedge-pair level measures for a given 2-digit SIC industry. We know how many newly established

companies are present for a given 2-digit SIC/zipcode, denoted here as x. We then create x observations

for that 2-digit SIC/zipcode, each of which is associated with the same set of location-specific control

variables (e.g. state tax measures). Using these data, we estimate linear probability models as described

in Section 3 that evaluate the likelihood that an establishment locates on given side of the state border as a

function of wedge-pair and state-level control measures.

Table 2 reports summary measures for newly established businesses in our data. We do this for

two different levels of geography. The first is based on zipcodes that lie at least partially within 10 miles

of a state border, while the second is based on zipcodes that lie at least partially within 1 mile of a state

border. In addition, for each wedge-pair, the side that is situated in the state that appears earlier in

alphabetical order is always labeled as side 1 while the other is labeled as being on side 2. This labeling

convention is adopted throughout the remainder of the paper.

Observe first that there are fewer newly created establishments in our sample in 2005 than in

2002. In the regression models to follow we address this by including a dummy variable for 2005 in all

of the regressions. Also apparent, there are more arrivals on side 2 than on side 1. Given the random

assignment of state-pair side-1 and side-2 designations, this difference either reflects a tendency for grid

squares to be positioned more on side 2 of the state borders, or for side-2 designated areas to be more

heavily developed. To address these possibilities, in some of the models to follow we include a control

for the difference in area between opposing sides of a wedge-pair, calculated as the area of the wedge on

side 2 of a wedge-pair (in square miles) minus the area of the wedge on side 1. In our more robust

models, this wedge-area difference variable is replaced with wedge-pair fixed effects that further control

for any underlying tendency of one side of the wedge-pair to be more heavily developed than the other.

Summing all arrivals associated with a given buffer zone, notice that there are 67,276 new

establishments in zipcodes extending into the 0 to 1 mile buffer, and 96,434 new establishments in

zipcodes extending into the 0 to 10 mile buffer. In most cases estimates from the regression models to

follow are quite similar for the two buffer zone samples, and for that reason, the results based on the 1-

mile buffer zone sample are not reported.

As described earlier, Figure 3 displays the border areas that are the focus of this study. Also

shown in Figure 3 is the intensity of development along a given segment of a border, with darker regions

indicating more intensive development. The figure makes apparent that the density of development along

the border region is highly skewed, with relatively few intensively developed areas and many lightly

developed regions. Table 3 quantifies that distribution for both the 1 mile buffer and 10 mile buffer

samples. For each sample, the table presents the distribution of total employment for the sample of new

business arrivals. For both samples, panel A treats each establishment as a separate observation while

panel B treats each wedge-pair as a separate observation. The heavily skewed distribution is evident with

a very disproportionate share of establishments concentrated in the most heavily developed areas,

mirroring the United States overall. We will draw on this stylized fact in some of the model

specifications to follow.

V. Results

5.1 Size of state government

We begin with the simplest specification that considers only the impact of the size of state

government on the location of newly established enterprises. In Table 4, we present results using both the

1 mile buffer sample and the 10 mile buffer sample. Three sets of estimates are provided for each sample

based on OLS, wedge-pair fixed effect, and state-pair fixed effect specifications. In all cases, the key

control measure is the log ratio of the per capita level of state expenditures on side 2 of the border relative

to side 1, and in the year in which the establishment is created (2002 or 2005). Also included in the OLS

and state-pair fixed effect models is the difference in the square mileage of the two wedges that comprise

a given wedge-pair, and a dummy variable for arrivals in the year 2005. When the models are estimated

by ordinary least squares (OLS), the t-ratios reported are based on robust standard errors; when wedge-

pair or state-level fixed effects are included the standard errors are clustered at the level of the specified

fixed effects. In all cases, given the linear specification the estimated coefficients on the slope variables

have the usual marginal interpretation.

Identification in the OLS models is based on variation across wedge-pair locations and across

years. For both buffer samples, it is noteworthy that the OLS estimates indicate that larger state

government is associated with fewer arrivals: the coefficient on the log ratio expenditure measure is very

similar for the two samples: roughly -0.32 and -0.28 for the 1 and 10 mile buffer samples, respectively.

Both of these estimates are also highly significant, with t-ratios of 36 and 40, respectively. Taken at face

value, this suggests that smaller state government attracts business activity. As suggested earlier,

however, a concern with the OLS specification is whether unobserved attributes close to the state borders

might bias these estimates.

The second column for each buffer sample includes controls for state-pair fixed effects.

Identification in this specification is obtained only from temporal variation in cross-border activity since

time-invariant state-pair border attributes are differenced away. Notably, the coefficient on the log ratio

of per capita expenditures becomes positive for both samples, but is also small and insignificant in each

instance. This is in sharp contrast to the negative coefficients in the OLS model. It also suggests that

cross-border differences in state tax conditions are indeed correlated with cross-border differences in

unobserved attributes that affect the side of the border on which a company locates.

To explore this issue further, the third column for each sample reports estimates based on the

wedge-pair fixed effect specification. This strips away time-invariant state-pair and local attributes in the

border segment in which a newly established enterprise is located. Observe that the coefficients on the

log ratio of the expenditure variable are nearly identical to the state-pair fixed effect models. This result

is suggestive that localized unobserved attributes are not highly correlated with cross-border tax

differentials after controlling for state-pair fixed effects.

5.2 State tax conditions

Table 5 extends the model by adding controls for the state tax rates described earlier (the

maximum corporate income tax rate, the maximum personal income tax rate, and the sales tax rate). As

before, t-ratios reported for the OLS models are based on robust standard errors while for the fixed effect

models standard errors are clustered at the level of the specified fixed effects. To conserve space, both in

Table 5 and in the tables to follow, only results from the 10-mile buffer sample are reported (results from

the 1-mile buffer sample are similar). In addition, for most of the discussion to follow we will focus on

qualitative patterns among the estimated coefficients (e.g. signs and relative size). Magnitudes for select

specifications will be discussed later.

In Table 5, notice that the coefficients on the log ratio of per capita expenditures are very similar

to those reported in Table 4. In addition, dropping the per capita expenditure variable had little effect on

the tax coefficients for most of the specifications, both in Table 5 and in the tables to follow.

These

patterns suggest that the influence of the size of state government on the state of choice for entrepreneurs

operating close to the border is largely independent of cross-border differences in the manner in which

state governments finance their expenditures. Observe also that the coefficients on the tax rates are

similar in the state-pair and wedge-pair fixed effect models (the second and third specifications for each

buffer sample). That similarity echoes patterns from Table 4 but does not always persist in the more fully

specified models to follow.

Focusing on the coefficient values in Table 5, for the OLS model the tax coefficients are positive

and highly significant for all three tax instruments. However, adding location fixed effects to the models

These results are not shown to conserve space.

causes the results to change dramatically. The coefficients on the corporate income and personal income

tax rates become negative, small, and not significant. The coefficients on the sales tax rate remain

positive and of about the same magnitude, but are now only marginally significant. This latter pattern

suggests that on average, the sales tax may tend to attract companies from across the state border. We

revisit this issue shortly. For now, and more generally, the patterns just noted provide limited support for

the idea that local taxes deter arrivals of new businesses.

The findings in Table 5 are broadly consistent with early papers in this literature which also

struggled to find evidence of tax deterrent effects. As will become apparent, that characterization changes

when additional features of the tax code are taken into account. Most important, in the following section

we highlight the influence of reciprocal agreements that affect whether workers pay personal income tax

to their state of residence or their state of employment. This is followed by industry-stratified models that

further highlight the influence of the sales tax, and models in which companies are stratified by

corporate/non-corporate status that highlight the role of the corporate income tax.

5.3 Reciprocal agreements and the personal income tax

Recall that if a reciprocal agreement is in force (e.g. Table 1c), then individuals pay personal

income tax to their state of residence, while if a reciprocal agreement is not in force, then individuals pay

personal income tax to the state in which they are employed. To our knowledge, these provisions have

been completely overlooked by previous studies of tax deterrent effects even though most reciprocal

agreements have been in place for many years (see Table 1c, for example). Nevertheless, these

provisions provide a powerful source of identification, and especially so when considering activity along

a state border.

For companies situated within easy commuting distance of a state border, reciprocal agreements

should have a pronounced effect on the state in which an entrepreneur chooses to locate a business, and

the related deterrent effect of the personal income tax. To understand why, consider first two adjacent

states for which a reciprocal agreement is not in force. In this instance, workers employed on the high

personal income tax side of the border pay higher taxes regardless of where they live. Provided the cross-

border difference in tax rates is not fully capitalized into land rents (e.g. Figure 2 of Section 2),

individuals employed on the high tax side of the border should receive higher nominal wages.

Otherwise, they would seek employment on the low tax side of the border until wages adjust and a spatial

equilibrium is attained. If instead, reciprocal agreements are in force, then for establishments within easy

commuting distance of the state border, workers could live on the tax-advantaged side of the border

regardless of where they work. In this instance, the deterrent effect of the personal income tax should be

eliminated.

Recall also (from Section 2) that the equilibrium location of a tax sheltered company is still

sensitive to a local tax if it competes for space with sectors that pay that tax. An implication of this

principle is that if we eliminate the deterrent effect of the personal income tax – as when reciprocal

agreements are in force – then the deterrent effects of the corporate income tax and sales tax are likely to

increase since those tax measures would gain in relative importance in driving industry bid-rent functions

that affect the equilibrium location of business establishments.

These priors are tested in Table 6 where we present separate regressions for locations with and

without a reciprocal agreement in force. To conserve space, only estimates based on the wedge-pair fixed

effect specification are presented, both in Table 6 and in the tables to follow.

In Table 6, notice that reciprocal agreements have a dramatic impact on the deterrent effect of

state tax rates for companies situated close to the border. With reciprocal agreements in place, the

personal income tax attracts companies from across the border while the corporate income tax and the

sales tax act as a deterrent. These effects are all statistically significant. In contrast, when reciprocal

agreements are not in force, the personal income tax deters new business activity while the sales tax lures

companies from across the border, and both are significant. The corporate income tax has a positive

This implicitly assumes that there are multiple types of companies bidding for space, some more sensitive to the

personal income tax than others, as described in Figure 2 of Section 2. If instead all companies were alike, then

higher personal income taxes should be fully capitalized into lower land rents and would not affect the equilibrium

sorting of companies across different sides of the state border.

coefficient, but it is small and not significant. Together, these patterns support the arguments above that

the presence of reciprocal agreements mitigates the deterrent effect of the personal income tax while

increasing the deterrent effect of other sources of tax revenue.

Why might a tax attract business activity from across the border, as with the personal income tax

when reciprocal agreements are in force and the sales tax when they are not? The answer is that

companies favor locations that raise revenue in ways that impose little cost on their activities relative to

the value of government services provided. With reciprocal agreements in place, the personal income tax

imposes little direct cost on establishments operating close to a state border. In such instances, companies

should favor locations that raise more of their revenue through the personal income tax. Analogously,

companies should tend to avoid locations where revenue is raised through taxes from which they are not

sheltered, as with the corporate income tax and sales tax when reciprocal agreements are not in place.

5.4 Differences across industries and the sales tax

In this section, we focus on the sales tax for which we have sharp priors about different effects

across industries. Recall that most retail activity is subject to whatever sales tax may be in place. In

contrast, only a subset of services are subject to the sales tax, and manufacturing establishments only

rarely face the sales tax. This suggests that the retail sector ought to be most sensitive to the sales tax and

the manufacturing sector should be the least sensitive to the sales tax.

Table 7 enables us to test these priors by repeating the analysis from Table 6 with separate

regressions for Manufacturing, Retail, and Services. In Table 7, note that when reciprocal agreements are

in place, a higher sales tax rate deters business arrivals. This is true for all three industries, but the

coefficient is especially large and significant for retail (-0.149 with a t-ratio of -3.56), moderate for

In Table 6, note also that for locations with a reciprocal agreement in force (the middle column), the coefficient on

the size of state government is negative but not significant. When reciprocal agreements are not in force (the third

column), the coefficient is positive and marginally significant. On balance, these patterns and related estimates in

extensions to follow do not point to a clear and robust pattern. In that sense, based on the estimates in Table 6 and

those to follow, we cannot reject the null that larger government has a relatively neutral effect on the tendency to

lure companies from across the state border.

services (-0.054 with a t-ratio of -1.57), and much smaller and not at all significant for manufacturing.

This is consistent with the view that reciprocal agreements enhance the deterrent effects of tax measures

outside of the personal income tax. The patterns just noted also support the view that retail should be

more sensitive to the sales tax than are services, which in turn are more sensitive than manufacturing.

When reciprocal agreements are not in force, the pattern is quite different. In that instance, the sales tax

tends to lure manufacturing and service establishments from across the border (the coefficients are both

positive and roughly 0.01 with t-ratios of 1.46 and 1.91, respectively), while the coefficient for retail is

positive but close to zero. This is consistent with findings from Table 6 that in the absence of a reciprocal

agreement the deterrent effect of the sales tax is reduced.

5.5 Corporate status and the corporate income tax

Table 8 revisits the models in Table 6 once again, but this time separate models are run for

establishments that are corporations and those that are either sole proprietorships or partnerships.

discussed earlier in the paper, because only corporations pay corporate income tax, the deterrent effect of

the corporate income tax seems likely to be more pronounced for corporations.

In the absence of a reciprocal agreement results are somewhat mixed relative to patterns in the

earlier tables. The corporate income tax and the sale tax have small and insignificant effects on both

samples. The personal income tax, in contrast, has a highly significant deterrent effect on unincorporated

establishments (with a coefficient of -0.0165 and a t-ratio of -3.15) but a smaller, positive, and marginally

significant effect on corporations (with a coefficient of 0.01 and a t-ratio of 1.77).

When reciprocal agreements are present, findings are largely consistent with priors. Notice that

the personal income tax lures both corporations and unincorporated companies from across the border.

This is as before. In addition, the sales tax has a deterrent effect on unincorporated establishments but

little effect on corporations. This may reflect some difference in the mix and appeal of larger retail

outlets that are incorporated versus smaller outlets that tend to not be incorporated. Most noteworthy, and

We are able to identify the ownership structure for roughly two-thirds of our sample.

consistent with the primary focus of this section, the corporate income tax has a strongly significant

deterrent effect for corporations and unincorporated establishments (the t-ratios on the relevant

coefficients are -3.15 and -2.51, respectively). Moreover, the deterrent effect is notably larger for

corporations with a coefficient value of -0.54 versus -0.28 for unincorporated companies. Once again,

therefore, deterrent effects appear to be larger when companies are most sensitive to the tax in question.

5.6 The scale of local development

Table 9 considers a final extension in which we run separate models from Table 6 for lightly

developed and heavily developed wedge-pairs in the sample. To facilitate comparisons, also reported in

Table 9 are full sample estimates that pool high and low density locations.

Review of the patterns in Table 9 indicates that there are both similarities and differences in tax

deterrent effects in lightly versus heavily developed areas. When reciprocal agreements are present,

deterrent effects from all three types of taxes are mostly similar for both subsamples and similar to the full

sample estimates. The primary exception is for the personal income tax which has little effect on

establishment location in lightly developed areas but a strong attractive effect in heavily developed

locations (with a coefficient of 0.033 and a t-ratio of 3.01). When reciprocal agreements are not present,

tax deterrent effects are small and insignificant for all three types of taxes in lightly developed locations,

but are enhanced in heavily developed locations and especially so for the personal income tax (which has

a coefficient of -0.022 and t-ratio of -4.16).

Summarizing, the patterns in Table 9 indicate that the magnitude of the coefficient on the

personal income tax is larger in heavily developed areas regardless of whether a reciprocal agreement is

in place or not. That finding is consistent with the possibility suggested in the Introduction that higher

labor productivity and wages in densely developed locations (e.g. Duranton and Puga (2004), Rosenthal

and Strange (2004), Arzaghi and Henderson (2008), Glaeser and Gottlieb (2009), Combes et al (2010))

may amplify the tendency of companies to reduce their exposure to higher personal income tax rates by

choosing the low tax side of the state border.

5.7 Magnitudes

To complete our discussion some brief comments about the magnitude of our estimates is in

order. For these purposes we focus on the coefficient estimates in Table 6 which provide the core results

upon taking reciprocal agreements into account.

Suppose that reciprocal agreements are not in force and Side 2 of a state border reduces the

maximum personal income tax rate by one percentage point relative to the state on the opposite side of the

border. Based on the estimates in column three of Table 6, that change would reduce the likelihood that

an establishment would locate on Side 2 of the border by 1 percentage point. If instead, the state on Side

2 of the border increased its sales tax rate by one percentage point, that would increase the likelihood that

a newly established company would choose Side 2 over Side 1 by roughly 1 percentage point; a change in

the corporate income tax has little effect. With a reciprocal agreement in place, an increase in the side-2

personal income tax of one percentage point would increase the likelihood of a company choosing side 2

by 2.2 percentage points. For the sales tax, a one percentage point increase would reduce the probability

of a side-2 location by 6.8 percentage points, while for the corporate income tax the analogous effect

would be a reduction in the side-2 location probability of roughly 23 percentage points.

It should be emphasized that there is a wide confidence band around most of these estimates.

Nevertheless, the weight of evidence in Table 6 and subsequent tables suggests that tax deterrent effects

among companies locating close to state borders are both statistically significant and large enough in

magnitude to warrant attention by policy makers.

VI. Conclusions

This paper has revisited an old question that has eluded efforts to provide a clear answer: to what

extent do local taxes deter business activity? Any number of measurement and econometric issues have

made this a challenging question to address for reasons that are well appreciated in the literature. For that

reason, our goals throughout the paper have been targeted. Our focus has been on the influence of state

government policy on the sorting of establishments across state lines for those companies that operate

within easy commuting distance of a state border. In that regard, we do not address the impact of state

policy on the overall level of business activity close to a state border. Instead, we consider the tendency

of entrepreneurs operating close to a state border to avoid adverse tax effects through their choice of state.

New to this paper, we provide the first ever analysis of state reciprocal agreements that require

individuals to pay personal income tax to their state of residence as opposed to their state of employment.

Controlling for that feature of the tax code proves to be a powerful source of identification. When

reciprocal agreements are in force, higher personal income tax rates lure companies from across the

border while higher corporate income and sales tax rates act as a deterrent. The reverse is true when

reciprocal agreements are not in force. The tendency of higher tax rates to sometimes lure companies

from across a state border might seem surprising at first but is actually quite intuitive: for a given set of

government services, entrepreneurs will tend to favor locations in which they are relatively sheltered from

the taxes used to finance local government activity. Our findings also suggest that tax deterrent effects

are large enough in magnitude to be economically important, even allowing for a wide confidence band

around the estimates.

Additional findings demonstrate that tax effects differ with the type of tax, the type of industry,

ownership structure of the company (e.g. corporate versus non-corporate), and the local level of

agglomeration. Given the sensitivity of our estimates to these considerations and also the influence of

reciprocal agreements, it is not surprising that many previous studies have struggled to find convincing

evidence of tax deterrent effects. Nevertheless, taken as a whole, our research design and estimates

suggest that entrepreneurs and new business activity are drawn to locations in which they are relatively

more sheltered from the cost of financing local government activities.

References

Arzaghi, Muhammad and J.V. Henderson (2008), “Networking off Madison Avenue,” Review of

Economic Studies, 75(4), 1011-1038.

Bartik, Timothy (1985), “Business Location Decisions in the United States: Estimates of the Effects of

Unionization, Taxes, and Other Characteristics of States.” Journal of Business and Economic

Statistics, 3, 14-22.

Bartik, Timothy (1994), “Taxes and Local Economic Development: What Do We Know and What Can

We Know?” Proceedings of the Eighty-Seventh Annual Conference on Taxation, National Tax

Association-Tax Institute of America, 102-106.

Bleakley, Hoyt and Jeffrey Lin (2010), “Portage: Path Dependence and Increasing Returns in U.S.

History,” Federal Reserve Bank of Philadelphia working paper 10-27.

Brulhart, Marius, Mario Jametti, and Kurt Schmidheiny (2009), “Do Agglomeration Economies Reduce

the Sensitivity of Firm Location to Tax Differentials?” working paper.

Brueckner, Jan K. (1979), “Property values, local public expenditures, and economic efficiency,” Journal

of Public Economics, 11, 223-245.

Brueckner, Jan K. (1982), “A test for allocative efficiency in the public sector,” Journal of Public

Economics, 14, 311-332.

Brueckner, Jan K. and Luz A. Saavedra (2001), “Do local governments engage in strategic property-tax

competition?” National Tax Journal, 54(2), 203-229.

Carlton, Dennis (1983), “The Location and Employment Choices of New Firms: An Econometric Model

with Discrete and Continuous Endogenous Variables.” The Review of Economics and Statistics,

65(3) 440-449.

Carlton, Dennis (1979), “Births of Single Establishment Firms and Regional Variation in Economic

Costs,” Paper 2, Report 7729, Center for Mathematical Studies in Business and Economics,

University of Chicago.

Chirinko, Robert, Daniel Wilson (2008)“Tax Competition among U.S. States: Racing to the

Bottom or Riding on a Seesaw?” Federal Reserve Bank of San Francisco Working Paper

No. 2008-03

Combes, Pierre- Philippe, Gilles Duranton, Laurent Gobillon, and Sébastien Roux (2010), “Estimating

Agglomeration Economies with History, Geology, and Worker Effects,” in Agglomeration

Economics, Edward Glaeser (ed.), The University of Chicago Press, Chicago, 15-66.

Cunningham, Christopher (2007), “Growth Controls, Real Options and Land Development,” Review of

Economics and Statistics, 89 (2) 343–358.

Devereux, Michael P., Rachel Griffith, and Helen Simpson (2007), “Firm Location Decisions, Regional

Grants and Agglomeration Economies,” Journal of Public Economics, 91, 413-435.

Dubay, Curtis S. and Chris Atkins (2006), “2007 State Business Tax Climate Index,” Tax Foundation

Background Paper, October 2006, Number 52.

Duranton, Gilles and Diego Puga (2004), "Micro-foundations of urban agglomeration economies," in J.

Vernon Henderson and Jacques-Francois Thisse (eds.), Handbook of Urban and Regional

Economics, Volume 4, Elsevier B.V. North Holland, Amsterdam, 2063-2118.

Duranton, Gilles, Laurent Gobillon, and Henry Overman (2011), “Assessing the Effects of Local

Taxation Using Microgeographic Data,” Economic Journal 121 (555), 1017-1046.

Glaeser, Edward L. and David C. Mare (2001), “Cities and Skills,” Journal of Labor Economics 19(2),

316-342.

Glaeser, Edward L and Joshua D. Gottlieb (2009), “The Wealth of Cities: Agglomeration Economies and

Spatial Equilibrium in the United States,” Journal of Economic Literature, 47:4, 983-1028.

Greenstone, Michael, Rick Hornbeck and Enrico Moretti (2010), “Identifying Agglomeration Spillovers:

Evidence from Winners and Losers of Large Plant Openings. Journal of Political Economy, 118

(3) 536-598.

Greenstone, Michael and Enrico Moretti (2004), “Bidding for Industrial Plants: Does Winning a ‘Million

Dollar Plants’ Increase Welfare?” working paper.

Goolsbee, Austan and Edward L. Maydew (2000), “Coveting thy Neighbor’s Manufacturing: The

Dilemma of State Income Apportionment,” Journal of Public Economics, 75, 125-143.

Gordon, Rodger and John D. Wilson (1986), “An Examination of Multijurisdictional Corporate Income

Taxation under Formula Apportionment,” Econometrica, 54(6), 1357-1373.

Grahm, John R. (1996), “Proxies for the Corporate Marginal Income Tax Rate,” Journal of Financial

Economics, 42(2), 187-221.

Hamilton, Bruce (1976), “Capitalization of Interjurisdictional Differences in Local Tax Prices,” American

Economic Review, 66(5), 743-753.

Hines, James R. (1996), “Altered States: Taxes and the Location of Foreign Direct Investment in

American,” American Economic Review, 86, 1076-1094.

Hines, James R. (2006), “Will Social Welfare Expenditures Survive Tax Competition?" Oxford Review of

Economic Policy 22, no. 3: 330-48.

Holmes, Thomas J. (1998), "The Effect of State Policies on the Location of Manufacturing: Evidence

from State Borders," Journal of Political Economy, 106 (4), 667-705.

Institute on Taxation and Economic Policy (2008), “Corporate income Tax Apportionment and the

‘Single Sales Factor’” http://www.itepnet.org/pdf/pb11ssf.pdf

, Policy Brief #11, Washington DC.

Jofre-Monseny, Jordi and Albert Sole-Olle (2007), “Tax Differentials and Agglomeration Economies in

Intraregional Firm Location,” working paper.

Klassen, K. and Shackelford, D. (1998), “State and Provincial Corporate Tax Planning: Income, Sales,

Assets, and Compensation Management,” Journal of Accounting and Economics, 25(3), 385-406.

Oates, William. and Schwab, R. (1988), “Economic competition among jurisdictions: efficiency-

enhancing or distortion-inducing?” Journal of Public Economics, 35, 333–354.

Papke, Leslie E (1991), “Interstate Business Tax Differentials and New Firm Location: Evidence from

Panel Data,” Journal of Public Economics, 45, 47-68.

Rosenthal, Stuart and Amanda Ross (2010), “Violent Crime, Entrepreneurship, and Cities,” Journal of

Urban Economics, 67(1), 135-149.

Rosenthal, Stuart S. and William C. Strange (2001), “The Determinants of Agglomeration,” Journal of

Urban Economics, 50(2), 191-229.

Rosenthal, Stuart and William Strange (2003), “Geography, Industrial Organization, and Agglomeration,”

The Review of Economics and Statistics, 85 (2): 377-393.

Rosenthal, Stuart S., and William C. Strange (2005), “The Geography of Entrepreneurship in the New

York Metropolitan Area,” Economic Policy Review, (Special Issue on “Urban Dynamics in New

York City”). New York Federal Reserve Bank, December, 11(2), 29-54.

Rosenthal, Stuart S., and William Strange (2008), “The Attenuation of Human Capital Spillovers,”

Journal of Urban Economics, 64(2), 373-389.

Rathelot, Roland and Patrick Sillard (2008), “The Importance of Local Corporate Taxes in Business

Location Decisions: Evidence from French Micro Data,” Economic Journal, 118, 499-514.

Schmenner, Roger (1975), “City Taxes and Industry Location,” Unpublished paper, Harvard Business

School.

Schmenner, Roger (1982), Making Business Location Decisions (Englewood Cliffs, NJ: Prentice- Hall).

Slemrod, Joel (2004), “Are corporate tax rates, or countries, converging?” Journal of Public Economics,

88(6), 1169-1186.

Tiebout, Charles (1956), “A Pure Theory of Local Expenditures,” Journal of Political Economy, 64(5),

416-424.

Uhlig, Harald and Mathias Trabandt (April, 2011), “How Far are We from the Slippery Slope? The Laffer

Curve Revisited,” MFI Working paper 2009-005.

Wasylenko, Michael (1997), “Taxation and Economic Development: The State of the Economic

Literature,” New England Economic Review, March/April, 37-52

Side I

Side II

Figure 1: One Sector

distance

Bid-Rent (No Tax)

Bid-Rent (With Tax)

Type-II Firm Bid-Rent: Tax

Protected Sector

Side I

Side II

Figure 2: Two Sectors

distance

Type-I Firm Bid-

Rent

(

No Tax

)

Type-I Firm Bid-

Rent (With Tax)

Figure 3: Border Region Total Employment

Figure 4: 20-by-20 mile Grid Squares Overlaid on 10-mile State Border Buffers

Table 1a: State Tax and Expenditures for the 48 Contiguous States

Panel A: State Revenue Shares and Expenditures for the 48 Contiguous States

Year

Sales Tax

Revenue Relative to

Expenditures

Household Income Tax

Revenue Relative to

Expenditures

Corporate Income Tax

Revenue Relative to

Expenditures

per Capita

2002 0.135 0.129 0.018 $4,650

2005 0.139 0.136 0.025 $5,207

Panel B: State Tax Rates for the 48 Contiguous States

Sales Tax

Rate Maximum Personal Income Tax Rate

Maximum Corporate Income Tax Rate

Year Median Std. Dev.

States with

No Tax Median Std. Dev.

States with

No Tax Median Std. Dev.

States with

No Tax

2000 5% 1.737 4 6% 2.709 6 7% 2.986 5

2003 5% 1.705 4 6% 2.750 6 7% 2.939 5

Using 2005 as an example, other major sources of state government revenue include: Intergovernmental grants, primarily from the federal

government and to a much lesser extent local government, roughly 24 percent; All other sources of tax, licensing, and general revenue,

roughly 20 percent; State-owned utilities and liquor stores, roughly 1 percent; Insurance trust revenue for government retirement and social

insurance programs including contributions by government workers and net earnings on fund investments, roughly 25 percent. See: U.S.

Census Bureau, State & Local Government Finance, Historical Data: 2005, http://www.census.gov/govs/estimate/historical_data_2005.html

Tax rate values were obtained from the Tax Foundation website for 2000 and 2003 at the following URL: www.taxfoundation.org.

Table 1b: Corporate Share of New Business Arrivals in the United States

2002:Q3 2005:Q3

Variable Total

Percent

Corporations Total

Percent

Corporations

Manufacturing 19,399 0.439 12,790 0.482

Wholesale Trade 15,489 0.463 11,481 0.486

Retail 63,601 0.266 36,831 0.373

Finance & Insurance 21,014 0.437 19,283 0.543

Services 122,968 0.325 100,429 0.473

Total 242,471 0.337 180,814 0.462

Sample includes all establishments created in the previous 12 months throughout the United States.

Data are from the Dun and Bradstreet MarketPlace file.

Table 1c: State Income Tax Reciprocal Agreements

State

Has Reciprocal

Agreement with

Year of

Inception State

Has Reciprocal

Agreement with

Year of

Inception

Illinois

Iowa 1973 Maryland Pennsylvania 1990

Kentucky 1971 Virginia 1992

Michigan 1971 West Virginia 1988

Wisconsin 1973 Michigan Minnesota 1984

Indiana Kentucky 1977 Ohio 1972

Michigan 1968 Wisconsin 1967

Ohio 1977 Minnesota North Dakota 1958

Pennsylvania 1977 Wisconsin 1968

Wisconsin 1977 Montana North Dakota 1982

Kentucky Michigan 1968 New Jersey Pennsylvania 1978

Ohio 1972 Ohio Pennsylvania 1978

West Virginia 1965 West Virginia 1972

Wisconsin 1968 Pennsylvania Virginia 1982

Virginia 1964 West Virginia 1972

Virginia West Virginia 1988

The reciprocal agreement data comes from www.gaebler.com. Note that Washington DC is not included in this list despite

having reciprocal agreements because Washington DC is not included in this study.

Table 1d: Changes in State Tax Rates Between 2000 and 2003

States that Changed

Their Tax Rate

Adjacent States for Which the

Cross-Border Difference in

Tax Rate Changed

Maximum Corporate Income Tax Rate 6 24

Maximum Personal Income Tax Rate 11 35

Sales Tax Rate 5 23

Table 2: Total Number of New Business Arrivals

Variable

2002

Side 1

2002

Side 2

2005

Side 1

2005

Side 2 Total

1 Mile Buffer Sample

14,645 21,353 13,947 17,331 67,276

10 Mile Buffer Sample

21,860 29,712 20,566 24,296 96,434

As described in the text, the 1-mile buffer sample includes new business establishments throughout the entirety of any

zipcode that is at least partially within one mile of the state border. The 10-mile buffer sample is analogous and is based on

a ten mile buffer.

Table 3: Wedge-Pair Employment Counts in 2002:Q3

Panel A: Each Establishment Treated as a Separate Observation

Percentile 1 Mile Buffer Sample

10 Mile Buffer Sample

10,594 13,779

32,596 48,854

82,441 139,607

288,511 465,073

3,522,033 3,936,475

Mean 547,247 620,103

Observations 67,276 96,434

Panel B: Each Wedge-Pair Treated as a Separate Observation

Percentile 1 Mile Buffer Sample

10 Mile Buffer Sample

2,620 2,647

7,375 7,723

14,033 16,013

30,203 35,424

91,815 135,454

Mean 33,428.48 44,868.15

Observations 832 869

As described in the text, the 1-mile buffer sample includes new business establishments

throughout the entirety of any zipcode that is at least partially within one mile of the state

border. The 10-mile buffer sample is analogous and is based on a ten mile buffer.

Table 4: Size of State Government

Dependent Variable is 1 if arrival on side 2 and 0 if arrival on side 1

(t-stats are reported in parentheses

)

1 Mile Buffer Sample

10 Mile Buffer Sample

OLS

State Pair

Wedge-Pair

FE OLS

State Pair

Wedge-Pair

Log(PerCapExp

/PerCapExp

)

-0.3267 0.0982 0.0821 -0.2795 0.0322 0.0363

(-36.47) (1.15) (1.47) (-40.23) (0.45) (0.68)

Area

– Area

(sq miles)

0.0071 0.0091 - 0.0021 0.0024 -

(166.30) (6.41) - (179.28) (10.13) -

Year 2005

-0.0363 -0.0234 -0.0179 -0.0286 -0.0191 -0.0172

(-10.27) (-2.73) (-3.73) (-9.89) (-2.65) (-3.72)

Observations 67,276 67,276 67,276 96,434 96,434 96,434

State Fixed Effects - 105 - - 104 -

Wedge-Pair Fixed Effects - - 832 - - 869

Adj. R-Squared (total) 0.15 0.24 0.42 0.19 0.30 0.46

As described in the text, the 1-mile buffer sample includes new business establishments throughout the entirety of any zipcode

that is at least partially within one mile of the state border. The 10-mile buffer sample is analogous and is based on a ten mile

buffer.

Robust standard errors are used for the OLS model. Standard errors are clustered at the level of the fixed effects for the state-

pair and wedge-pair fixed effect models.

PerCapExp

and PerCapExp

are the state per capita levels of expenditures on sides 1 and 2 of the border.

Area

and Area

are the square mileage of the wedges from sides 2 and 1 that belong to a given wedge pair.

Table 5: Tax Measures

Dependent Variable is 1 if arrival on side 2 and 0 if arrival on side 1

(10 mile buffer sample

; t-stats are reported in parentheses

)

OLS

State Pair

Wedge-Pair

Log(PerCapExp

/PerCapExp

)

-0.3275 0.05351 0.0579

(-43.40) (0.73) (1.07)

Max Corp Tax Rate (Side 2 – Side 1) 0.0042 -0.0020 -0.0016

(5.52) (-0.21) (-0.23)

Max Personal Tax Rate (Side 2 – Side1) 0.0068 -0.0020 -0.0013

(8.96) (-0.32) (-0.25)

Sales Tax Rate (Side 2 – Side 1) 0.0100 0.0073 0.0080

(13.17) (1.53) (1.71)

Area

– Area

(sq miles)

0.0021 0.0023 -

(178.06) (10.09) -

Year 2005

-0.0249 -0.0195 -0.0172

(-8.63) (-2.94) (-3.97)

Observations 96,434 25,012 71,422

State-Pair Fixed Effects - 104 -

Wedge-Pair Fixed Effects - - 869

Adj. R-Squared (total) 0.36 0.42 0.34

As described in the text, the 10-mile buffer sample includes new business establishments

throughout the entirety of any zipcode that is at least partially within ten miles of the state

border.

Robust standard errors are used for the OLS model. Standard errors are clustered at the

level of the fixed effects for the state-pair and wedge-pair fixed effect models.

PerCapExp

and PerCapExp

are the state per capita levels of expenditures on sides 1

and 2 of the border.

Area

and Area

are the square mileage of the wedges from sides 2 and 1 that belong to a

given wedge pair.

Table 6: Reciprocal Agreements

Dependent Variable is 1 if arrival on side 2 and 0 if arrival on side 1

(10 mile buffer sample

; t-stats are in parentheses

)

Full Sample

Reciprocal

Agreements

Reciprocal

Agreements

Log(PerCapExp

/PerCapExp

)

0.0579 -0.2043 0.0852

(1.07) (-0.89) (1.53)

Max Corp Tax Rate (Side 2 – Side 1)

-0.0016 -0.2298 0.0021

(-0.23) (-2.27) (0.36)

Max Personal Tax Rate (Side 2 – Side1)

-0.0013 0.0226 -0.0092

(-0.25) (2.54) (-1.97)

Sales Tax Rate (Side 2 – Side 1)

0.0080 -0.0678 0.0082

(1.71) (-2.27) (1.80)

Year 2005

-0.0172 0.0018 -0.0205

(-3.97) (0.22) (-4.42)

Observations 96,434 25,012 71,422

Wedge-Pair Fixed Effects 869 195 674

Adj. R-Squared (total) 0.36 0.42 0.34

As described in the text, the 10-mile buffer sample includes new business establishments

throughout the entirety of any zipcode that is at least partially within ten miles of the state

border.

Robust standard errors are used for the OLS model. Standard errors are clustered at the

level of the fixed effects for the state-pair and wedge-pair fixed effect models.

PerCapExp

and PerCapExp

are the state per capita levels of expenditures on sides 1 and

2 of the border.

Exp

and Exp

are the state levels of expenditures on sides 1 and 2 of the border.

Table 7: Reciprocal Agreements and Stratification by Industry

Dependent Variable is 1 if arrival on side 2 and 0 if arrival on side 1

(10 mile buffer sample

; t-stats are in parentheses

)

Reciprocal Agreements No Reciprocal Agreements

Manufacturing Retail Services Manufacturing Retail Services

Log(PerCapExp

/PerCapExp

)

-0.1899 -0.3517 -0.2483 -0.1228 -0.0154 0.0590

(-0.22) (-0.93) (-0.82) (-0.48) (-0.16) (0.91)

Max Corp Tax Rate (Side 2 – Side 1)

-0.3637 -0.1600 -0.2947 0.1778 0.0007 -0.0007

(-1.11) (-1.32) (-2.11) (1.25) (0.09) (-0.10)

Max Personal Tax Rate (Side 2 – Side1)

-0.0283 0.0293 0.0313 -0.0076 -0.0091 -0.0064

(-1.42) (3.19) (2.51) (-0.59) (-1.57) (-1.20)

Sales Tax Rate (Side 2 – Side 1)

-0.0160 -0.1495 -0.0536 0.0108 0.0028 0.0090

(-0.18) (-3.56) (-1.57) (1.46) (0.67) (1.91)

Year 2005

-0.0654 0.0296 0.0013 -0.0009 -0.0190 -0.0230

(-1.76) (2.07) (0.12) (-0.06) (-2.56) (-4.13)

Observations 937 4,705 12,690 2,744 16,105 32,072

Wedge-Pair Fixed Effects 73 161 180 187 547 600

Adj. R-Squared (total) 0.41 0.4 0.42 0.32 0.34 0.34

As described in the text, the 10-mile buffer sample includes new business establishments throughout the entirety of any zipcode that is at least partially within ten

miles of the state border.

Robust standard errors are used for the OLS model. Standard errors are clustered at the level of the fixed effects for the state-pair and wedge-pair fixed effect

models.

PerCapExp

and PerCapExp

are the state per capita levels of expenditures on sides 1 and 2 of the border.

Table 8: Reciprocal Agreements and Corporate Versus Non-Corporate Status

Dependent Variable is 1 if arrival on side 2 and 0 if arrival on side 1

(10 mile buffer sample

; t-stats are in parentheses

)

Reciprocal Agreements No Reciprocal Agreements

Corporations

Sole Proprietorships

and Partnerships Corporations

Sole Proprietorships

and Partnerships

Log(PerCapExp

/PerCapExp

)

-0.2699 0.2251 0.0611 0.0720

(-0.78) (0.70) (0.58) (0.90)

Max Corp Tax Rate (Side 2 – Side 1)

-0.5406 -0.2849 -0.0025 0.0027

(-3.15) (-2.51) (-0.32) (0.45)

Max Personal Tax Rate (Side 2 – Side1)

0.0157 0.0421 0.0108 -0.0165

(1.47) (2.46) (1.77) (-3.15)

Sales Tax Rate (Side 2 – Side 1)

0.0088 -0.0685 0.0033 0.0087

(0.23) (-1.75) (0.86) (1.15)

Year 2005

-0.0361 0.0273 0.0016 -0.0234

(-4.16) (2.41) (0.20) (-3.62)

Observations 10,116 9,807 27,956 28,177

Wedge-Pair Fixed Effects 185 188 611 654

Adj. R-Squared (total) 0.41 0.42 0.33 0.35

As described in the text, the 10-mile buffer sample includes new business establishments throughout the entirety of any

zipcode that is at least partially within ten miles of the state border.

Robust standard errors are used for the OLS model. Standard errors are clustered at the level of the fixed effects for the

state-pair and wedge-pair fixed effect models.

PerCapExp

and PerCapExp

are the state per capita levels of expenditures on sides 1 and 2 of the border.

Table 9: Reciprocal Agreements and Stratification by Agglomeration

Dependent Variable is 1 if arrival on side 2 and 0 if arrival on side 1

(10 mile buffer sample

; t-stats are in parentheses

)

Reciprocal Agreements No Reciprocal Agreements

Full Sample

Lightly Developed

(< 95

Percentile)

Heavily Developed

(> 95

Percentile)

Full Sample

Lightly Developed

(< 95

Percentile)

Heavily Developed

(> 95

Percentile)

Log(PerCapExp

/PerCapExp

)

-0.2043 -0.1091 -0.3173 0.0852 0.0618 0.1761

(-0.89) (-0.40) (-0.52) (1.53) (0.97) (1.53)

Max Corp Tax Rate (Side 2 – Side 1)

-0.2250 -0.1180 -0.3594 0.0021 -0.0028 0.1476

(-2.27) (-1.42) (-1.75) (0.36) (-0.45) (2.02)

Max Personal Tax Rate (Side 2 – Side1)

0.0226 -0.0077 0.0330 -0.0092 -0.0007 -0.0217

(2.54) (-0.62) (3.01) (-1.97) (-0.13) (-4.16)

Sales Tax Rate (Side 2 – Side 1)

-0.0678 -0.0583 -0.0922 0.0082 0.0015 0.0192

(-2.27) (-2.29) (-1.25) (1.80) (0.59) (2.94)

Year 2005

0.0018 0.0065 0.0065 -0.0206 -0.0196 -0.0177

(0.22) (0.50) (0.62) (-4.42) (-3.36) (-1.53)

Observations 25,012 10,682 14,330 71,422 37,403 34,019

Wedge-Pair Fixed Effects 195 180 15 674 646 28

Adj. R-Squared (total) 0.42 0.39 0.44 0.34 0.36 0.33

As described in the text, the 10-mile buffer sample includes new business establishments throughout the entirety of any zipcode that is at least partially within ten miles

of the state border.

Robust standard errors are used for the OLS model. Standard errors are clustered at the level of the fixed effects for the state-pair and wedge-pair fixed effect models.

PerCapExp

and PerCapExp

are the state per capita levels of expenditures on sides 1 and 2 of the border.

Lightly and heavily developed locations are defined as those less than or greater than the 95

percentile as defined in Panel B of Table 3.

Appendix: Supplemental Tables Based on Tax Revenue Measures

Table A-1: Tax Measures

Dependent Variable is 1 if arrival on side 2 and 0 if arrival on side 1

(10 mile buffer sample

; t-stats are in parentheses

)

OLS State FE

Wedge-Pair

Log(PerCapExp

/PerCapExp

)

-0.2611 0.0434 0.0523

(-32.08) (0.58) (0.98)

CorpTaxRev/Exp

– CorpTaxRev/Exp

-0.4212 0.7819 1.013

(-3.65) (0.93) (1.86)

IncTaxRev/Exp

– IncTaxRev/Exp

-0.0052 -0.2148 -0.2233

(-0.24) (-1.36) (-1.72)

SalesTaxRev/Exp

– SalesTaxRev/Exp

0.1631 0.5000 0.4751

(7.22) (3.46) (3.30)

Area

– Area

(sq miles)

0.0021 0.0024 -

(170.19) (10.09) -

Year 2005

-0.0299 -0.0171 -0.0140

(-10.27) (-2.55) (-3.20)

Observations 96,434 25,012 71,422

State-Pair Fixed Effects - 105 -

Wedge-Pair Fixed Effects - - 869

Adj. R-Squared (total) 0.19 0.30 0.46

As described in the text, the 10-mile buffer sample includes new business establishments

throughout the entirety of any zipcode that is at least partially within ten miles of the state

border.

Robust standard errors are used for the OLS model. Standard errors are clustered at the

level of the fixed effects for the state-pair and wedge-pair fixed effect models.

PerCapExp

and PerCapExp

are the state per capita levels of expenditures on sides 1

and 2 of the border.

Exp

and Exp

are the state levels of expenditures on sides 1 and 2 of the border.

Area

and Area

are the square mileage of the wedges from sides 2 and 1 that belong to a

given wedge pair.

Table A-2: Reciprocal Agreements

Dependent Variable is 1 if arrival on side 2 and 0 if arrival on side 1

(10 mile buffer sample

; t-stats are in parentheses

)

Full Sample

Reciprocal

Agreements

Reciprocal

Agreements

Log(PerCapExp

/PerCapExp

)

0.0523 -0.1987 0.1181

(0.98) (-0.78) (2.24)

CorpTaxRev/Exp

– CorpTaxRev/Exp

1.013 -2.060 1.976

(1.86) (-2.02) (3.47)

IncTaxRev/Exp

– IncTaxRev/Exp

-0.2233 -0.6233 -0.1129

(-1.72) (-1.22) (-0.90)

SalesTaxRev/Exp

– SalesTaxRev/Exp

0.4751 0.0879 0.6539

(3.30) (0.09) (4.87)

Year 2005

-0.0140 -0.0281 -0.0146

(-3.20) (-3.00) (-2.97)

Observations 96,434 25,012 71,422

Wedge-Pair Fixed Effects 869 195 674

Adj. R-Squared (total) 0.4556 0.2836 0.5166

As described in the text, the 10-mile buffer sample includes new business establishments

throughout the entirety of any zipcode that is at least partially within ten miles of the state

border.

Robust standard errors are used for the OLS model. Standard errors are clustered at the

level of the fixed effects for the state-pair and wedge-pair fixed effect models.

PerCapExp

and PerCapExp

are the state per capita levels of expenditures on sides 1 and

2 of the border.

Exp

and Exp

are the state levels of expenditures on sides 1 and 2 of the border.

Table A-3: Reciprocal Agreements and Stratification by Industry

Dependent Variable is 1 if arrival on side 2 and 0 if arrival on side 1

(10 mile buffer sample

; t-stats are in parentheses

)

Reciprocal Agreements No Reciprocal Agreements

Manufacturing Retail Services Manufacturing Retail Services

Log(PerCapExp

/PerCapExp

)

-2.1504 -0.0879 -0.2313 -0.1480 -0.0045 0.1053

(-1.74) (-0.19) (-0.76) (-0.66) (-0.05) (1.74)

CorpTaxRev/Exp

– CorpTaxRev/Exp

-2.9976 -2.0789 -2.1746 -1.7941 0.8342 2.6259

(-0.81) (-1.34) (-1.75) (-0.77) (0.85) (4.06)

IncTaxRev/Exp

– IncTaxRev/Exp

-2.8034 -0.6110 -0.6709 0.1467 -0.1027 -0.1824

(-1.00) (-0.61) (-1.10) (0.26) (-0.38) (-1.38)

SalesTaxRev/Exp

– SalesTaxRev/Exp

6.4390 1.2754 -0.8489 0.4401 0.4592 0.7558

(1.46) (0.77) (-0.79) (1.63) (3.10) (5.36)

Year 2005

-0.0966 -0.0023 -0.0363 -0.0280 -0.0160 -0.0139

(-2.34) (-0.15) (-3.11) (-1.55) (-2.12) (-2.42)

Observations 937 4,705 12,690 2,744 16,105 32,072

Wedge-Pair Fixed Effects 73 161 180 187 547 600

Adj. R-Squared (total) 0.41 0.4 0.42 0.32 0.34 0.34

As described in the text, the 10-mile buffer sample includes new business establishments throughout the entirety of any zipcode that is at least partially within ten

miles of the state border.

Robust standard errors are used for the OLS model. Standard errors are clustered at the level of the fixed effects for the state-pair and wedge-pair fixed effect

models.

PerCapExp

and PerCapExp

are the state per capita levels of expenditures on sides 1 and 2 of the border.

Exp

and Exp

are the state levels of expenditures on sides 1 and 2 of the border.

Table A-4: Reciprocal Agreements and Corporate Versus Non-Corporate Status

Dependent Variable is 1 if arrival on side 2 and 0 if arrival on side 1

(10 mile buffer sample

; t-stats are in parentheses

)

Reciprocal Agreements No Reciprocal Agreements

Corporations

Sole Proprietorships

and Partnerships Corporations

Sole Proprietorships

and Partnerships

Log(PerCapExp

/PerCapExp

)

-0.7365 0.2518 0.1575 0.0923

(-1.94) (0.77) (1.39) (1.19)

CorpTaxRev/Exp

– CorpTaxRev/Exp

-2.2333 -1.3207 1.8628 1.9642

(-1.44) (-0.83) (1.61) (2.36)

IncTaxRev/Exp

– IncTaxRev/Exp

0.0031 -0.9314 0.4322 -0.1839

(0.01) (-1.30) (1.67) (-0.80)

SalesTaxRev/Exp

– SalesTaxRev/Exp

-1.1302 -0.3560 0.3762 0.7486

(-0.98) (-0.28) (2.38) (3.09)

Year 2005

-0.0630 -0.0106 -0.0009 -0.0150

(-4.64) (-0.75) (-0.13) (-2.25)

Observations 10,116 9,807 27,956 28,177

Wedge-Pair Fixed Effects 185 188 611 654

Adj. R-Squared (total) 0.41 0.42 0.33 0.35

As described in the text, the 10-mile buffer sample includes new business establishments throughout the entirety of any

zipcode that is at least partially within ten miles of the state border.

Robust standard errors are used for the OLS model. Standard errors are clustered at the level of the fixed effects for the

state-pair and wedge-pair fixed effect models.

PerCapExp

and PerCapExp

are the state per capita levels of expenditures on sides 1 and 2 of the border.

Exp

and Exp

are the state levels of expenditures on sides 1 and 2 of the border.

Table A-5: Reciprocal Agreements and Stratification by Agglomeration

Dependent Variable is 1 if arrival on side 2 and 0 if arrival on side 1

(10 mile buffer sample

; t-stats are in parentheses

)

Reciprocal Agreements No Reciprocal Agreements

Full Sample

Lightly Developed

(< 95

Percentile)

Heavily Developed

(> 95

Percentile)

Full Sample

Lightly Developed

(< 95

Percentile)

Heavily Developed

(> 95

Percentile)

Log(PerCapExp

/PerCapExp

)

-0.1986 -0.2359 -0.3361 0.1180 0.05158 0.2572

(-0.78) (-0.93) (-0.50) (2.24) (0.77) (2.65)

CorpTaxRev/Exp

– CorpTaxRev/Exp

-2.0601 -3.7418 -3.2733 1.9757 1.5944 2.4236

(-2.02) (-3.74) (-1.19) (3.47) (2.00) (4.22)

IncTaxRev/Exp

– IncTaxRev/Exp

-0.6233 -0.1744 -1.5136 -0.1129 0.1234 -0.2115

(-1.22) (-0.30) (-1.53) (-0.90) (0.70) (-1.27)

SalesTaxRev/Exp

– SalesTaxRev/Exp

0.0878 0.7495 0.4717 0.6538 0.3388 0.9301

(0.09) (0.81) (0.24) (4.87) (1.55) (5.77)

Year 2005

-0.0281 -0.0056 -0.0602 -0.0145 -0.0157 -0.0220

(-3.00) (-0.49) (-3.91) (-2.97) (-2.40) (-2.96)

Observations 25,012 10,682 14,330 71,422 37,403 34,019

Wedge-Pair Fixed Effects 195 180 15 674 646 28

Adj. R-Squared (total) 0.42 0.39 0.44 0.34 0.36 0.33

As described in the text, the 10-mile buffer sample includes new business establishments throughout the entirety of any zipcode that is at least partially within ten miles

of the state border.

Robust standard errors are used for the OLS model. Standard errors are clustered at the level of the fixed effects for the state-pair and wedge-pair fixed effect models.

PerCapExp

and PerCapExp

are the state per capita levels of expenditures on sides 1 and 2 of the border.

Exp

and Exp

are the state levels of expenditures on sides 1 and 2 of the border.

Lightly and heavily developed locations are defined as those less than or greater than the 95

percentile as defined in Panel B of Table 3.