Journal of Economic Geography Advance Access published August 27, 2015

Journal of Economic Geography (2015) pp. 1–20 doi:10.1093/jeg/lbv031

Highways, local economic structure and

urban development
Marco Percocoy
Department of Policy Analysis and Public Management, Università Bocconi, via Roentgen 1, 20121 Milano,
Italy
y
Corresponding author: Marco Percoco, Department of Policy Analysis and Public Management, Università
Bocconi, via Rontgen 1, 20121 Milano, Italy. email [email protected]

Abstract

Downloaded from http://joeg.oxfordjournals.org/ at Rutgers University on August 28, 2015

Transport costs are widely considered as a key driver of competitive advantage of
countries, regions and cities. Their relevance is even greater when scale economies
are at work since production is concentrated and goods must be shipped. Recent
literature has found that highways, by decreasing transport costs, are crucial in
influencing agglomeration economies and ultimately urban development. In this
article, we contribute to this literature by studying the effect of highway construction
on the structure of local economies. In particular, we consider the effect of highways
in Italian cities in terms of firm location by explicitly recognizing the pivotal role played
by the transport sector and by intersectoral linkages in promoting development. The
main research hypothesis is that the location of an highway exit in a given city
attracts firms operating in the transport service sector and consequently transport-
intensive firms. Our empirical evidence concerns Italian cities over the period
1951–2001 and exploits variation in employment, population and plants induced by
the construction of the highway network. To deal with the endogeneity of the
geography of highways exits, we propose as an instrument the geography of Roman
roads. To this end, we have coded the whole network of Roman roads in Italy. We
have found that the location of highway exits increases employment and the number
of plants and that this growth is concentrated in transport service-intensive sectors.
This result is robust to a number of checks, including eventual instrument non-validity
and selection into treatment.

Keywords: Highways, urban development, accessibility

JEL classifications: L91, N70, R11, R49
Date submitted: 18 June 2013 Date accepted: 20 July 2015

1. Introduction
Transport costs are generally considered as an important driver of economic
development and of economic geography. The recent World Development Report
(World Bank, 2009) has effectively summarized the literature and argued that the effects
of the reduction in transport costs occurred over the past two centuries has resulted in
an increase in international trade and in spatial concentration of production as a
consequence of tougher competition. In the case of scale economies in production, the
effect of a change in transport costs is nonlinear and the net effect on development

ß The Author (2015). Published by Oxford University Press. All rights reserved. For Permissions, please email: [email protected]
2 of 20 . Percoco

depends on initial conditions (Martin and Rogers, 1995; Fujita and Mori, 1996).
For developing countries and lagging regions, the impact of transport policies designed
for decreasing freight costs can be unclear or even negative if interventions are not such
that treated areas move from one equilibrium to another. Similarly, the reduction in
transport costs may generate unclear effects also at local level. Most of the recent
literature has in fact focused on the impact of expansion of infrastructure network (as a
proxy for transport cost reduction) on urban development. Baum Snow (2007) studied
the effect of highway expansion on urban sprawl in a large sample of US Metropolitan
Statistical areas. Among several sources of endogeneity of the shape of highway
network, the author identifies political bargaining as one of the most problematic
and difficult to deal with. To identify causally the effect of highways he ingeniously
proposes to use the map of the initial project of highway network in the USA as an

Downloaded from http://joeg.oxfordjournals.org/ at Rutgers University on August 28, 2015

instrument for actual road development. This choice is justified by the fact that the map
used is a representation of the planned network before political bargaining took place
and hence is a good approximation of how the network would have been. A similar
approach was adopted by Duranton and Turner (2012) who consider the effect of road
and transport service on urban growth. They find that infrastructure causes growth,
although given their construction costs, the effectiveness of their further expansion can
be questioned.1
Donaldson (2013) finds that the expansion of railroads in India has promoted
international and interregional trade as well as price convergence between districts.
According to the geographical economics literature, the effect of a change in transport
costs can be nonlinear in a world characterized by multiple equilibria. Interestingly,
Bleakley and Lin (2012) address this issue empirically by exploiting a natural
experiment related to portage system in US counties. The authors have interestingly
found that portage sites became prosperous and specialized in the commercial sector
at the beginnings of the 19th century; their prosperity was maintained also when
portage technology lost its competitiveness with respect to other modes of transport
such as railways. Despite this persistence, Bleakley and Lin (2012) could not find
evidence of multiple equilibria. By using firm-level data, Gibbons et al. (2012)
analysed the effect of road transport innovation on firm behaviour in the UK.
Interestingly enough, they have found that improvements into road viability affect
firm location in terms of entry and exit into local markets, while they could not find
any effect on employment growth in firms located in treated areas before the
treatment was introduced.
In this article, we contribute to the literature by studying the effect of highway
construction on firm location behavior and on the structure of local economies. In
particular, we consider the effect of highways on Italian cities in terms of firm location
by explicitly recognizing the crucial role played by the transport sector and by
intersectoral spillovers in promoting development. In particular, we study the case of
the construction of the highway network in Italy occurred in the period 1950–1970 in a
quasi-experimental setting. By assembling a large dataset on all Italian cities, we have
estimated the effect of the location of an highway exit on the urban economy in terms
of population, number of plants and employment growth. Contrary to previous

1
A recent strand of literature has also focused on the impact of infrastructure on interregional trade and
eventually on price convergence (Michaels, 2008; Donaldson, 2013; Duranton et al., 2014).
 Highways and urban development . 3 of 20

literature, which uses as a proxy for accessibility the extension of the highways within a
territory, we make use of a novel dataset containing information on the location of
exits and on their catchment area. This choice has been made because we think that
accessibility is better measured in terms of actual access to the infrastructure. However,
this variable poses serious problems of endogeneity since their location is likely to
be driven by factors that drive outcomes of interest, or be determined themselves by
the outcomes of interest. To deal with the endogeneity of the geography of
highway exits, we propose as an instrument the network of Roman roads built
about 2000 years before. In particular, we use a dummy variable indicating the
presence of a Roman road as an instrument for the presence of a highway exit (a point
of access to the highway network) to explain the impact of the increase in road
accessibility on economic growth. To this end, we have coded the whole network of

Downloaded from http://joeg.oxfordjournals.org/ at Rutgers University on August 28, 2015

Roman roads, both the main and the secondary ones. We have found that highway
accessibility increases employment by 4–5% and plants by 2–3%, depending on the
specification and that the majority of this effect is concentrated in transport service-
intensive sectors.
We further contribute to the literature by testing and finding support to the
hypothesis that the impact of increased accessibility works through coagglomeration
forces driven by location decision of firms in the transport sector. This evidence is
corroborated by a number of robustness checks and deviations from baseline models,
including selection into treatment and placebo regressions. The article is organized as
follows. In Section 2 we present a brief history of the highway network in Italy, whereas
Section 3 contains a description of our methodological approach. Section 4 presents the
choice of our instrument, namely the geography of ancient Roman roads and Section 5
contains results and robustness checks. Section 6 concludes.

2. The highway network in Italy

An efficient highway system is probably among the prominent needs for industrialized
economies as most of freight is carried by trucks. According to recent Eurostat statistics
on modal split of freight transport, in 2012, the share of freight transport carried by
trucks was 75% in Europe and 86% in Italy (in terms of total inland freight tonne-
kilometre). In Italy, massive extension of highways took place during the 50s and 60s of
20th century and coincided with a period of sustained growth and mass diffusion of
cars, although some highways were built well before (during the 20s) in Lombardy,
between Milan and the lakes on the border with Switzerland and between Naples and
its suburban town Capua. A significant effort in the extension of the highway network
came in the aftermath of World War II and in May 1955 when the Romita law was
approved. This act planned to build more than 1200 km of new highways, with the most
important being the A1 Milan–Naples, the so-called ‘Autostrada del Sole’ (Maggi,
2009). Figure 1 reports the temporal evolution of the highway network in Italy. It shows
how it more than doubled between 1955 and 1960. In 1972, the quantity of kilometers
was more than 10 times the one in 1955. Along this period, almost 208 km were built
every year, whereas Germany built 170 and France 127. At the end of 1974, the Italian
network of highways was almost double than the one of France and UK and was
smaller only than USA and German ones. Figure 2 shows the highway system as in
2001, in terms of geography of the network.
 4 of 20 . Percoco

7000

6000

5000
Km of highways

4000

3000

Downloaded from http://joeg.oxfordjournals.org/ at Rutgers University on August 28, 2015

2000

1000

0
1955 1960 1963 1966 1969 1972 1975 1978 1981 1984 1987 1990

Year
Source: Maggi (2009)

Figure 1. The expansion of the highway network over time.

The most important highway was certainly the A1 Milan–Naples whose construction
lasted 5 years between 1959 and 1964 to build almost 700 km of roads. In San Donato
Milanese, municipality located in South of Milan, 19 May 1956, it laid the foundation
stone of the Autostrada del Sole that day in the presence of President Giovanni Gronchi
and Archbishop Giovanni Battista Montini of Milan a marble stone with the
inscription, which linked the motorway to the roads of ancient Rome (Menduni, 1999).
In July 1959 the trait Milan–Bologna was completed and the following year, in
December 1960, the highway touched Florence and, finally, in October 1964, arrived in
Naples. Within 8 years, then, was an artery in the light of 755 km, for a long time to
become the main transport axis of the peninsula, through it, it was thought, would have
met the conditions for osmosis in ‘hundred cities of Italy’, because not only was
breaking the physical border between North and South, but would also eventually loose
economic and cultural ones that still separated the two poles of the nation (Menduni,
1999). The Highway of the Sun was the carrier through (and long), which came to life
the incredible economic development that marked those years, although today there are
few historians and economists who remember the role it played, and still less the literary
evidence and film, the footprints it left on the land are important and are still evident
(Menduni, 1999; Cardinale, 2000; Maggi, 2003). Its construction, and the grafts that
followed, helped to trigger social phenomena (including mass tourism and commuting)
and economic (primarily the relocation space industries) which were accompanied by
important changes to zoning and the industrial fabric of the country.
The construction of the highway network made accessible and competitive for firm
location some areas of the country (especially in the centre), that before that policy
 Highways and urban development . 5 of 20

Downloaded from http://joeg.oxfordjournals.org/ at Rutgers University on August 28, 2015

Figure 2. The highway network in Italy (2001).

interventions were relatively underdeveloped. Accessibility to highways is granted by

the location of exits. In Figure 3, the opening of highway exits across time is shown. In
conformity with the pattern of investment in Figure 1, Figure 3 shows that the majority
of highway exits was opened during the 60s and 70s. In our analysis, exits play a crucial
role since we will assume that the opening of an exit in the surrounding of a city is a
good proxy for a change in the accessibility. If we consider this assumption as plausible,
then Figure 3 shows the timing of the treatment. For reasons related to the endogeneity
of the treatment (i.e., the decision to locate an exit) we will not exploit fully the
temporal pattern of the treatment, although in Section 5.3.1 we will also present results
of a panel difference-in-difference model.
 6 of 20 . Percoco

250

219

200
N. of exit openings

162
150

100

Downloaded from http://joeg.oxfordjournals.org/ at Rutgers University on August 28, 2015

50 51

12 11 10
4
0
Before 1950 Years '50 Years '60 Years '70 Years '80 Years '90 Years 2000

Source: CERTeT (2006).

Figure 3. Openings of highway exits.

3. Methodology
The objective of this study is to estimate the impact of a change in highway accessibility
on urban development indicators such as population, employment and the number of
plants. For this purpose, let us consider the function:
yit ¼ ai þ dt þ accessibilityit þ controlsit þ "it ; ð1Þ
where yit is either the population, the number of plants or jobs in city i at time t;
accessibilityit is an accessibility indicator, whereas controlsit is a set of control variables.
Parameters ai ; dt indicate city-specific and year-specific fixed effects. Taking first
differences we obtain:
4yit ¼ dt þ 4accessibilityit þ 4controlsit þ 4"it : ð2Þ
Note that a function like that in Equation (2) eliminates any fixed effects from Equation (1).
However, it should be noted that the elimination of city-specific fixed effects holds only if
those parameters are stable. We shall consider the period 1951–2001, which is long enough
not to consider this assumption plausible. However, we shall consider narrower time
windows to obtain more precise estimates as well as time-invariant controls to proxy
changes in city-specific fixed effects.
We assume that the opening of an highway exit makes a significant change in the
accessibility of the city, so that we estimate the following version of Equation (2):
4yit ¼ dt þ exitit þ 4controlsit þ 4"it ; ð3Þ
where it is assumed that 4accessibilityit ffi exitit : Variable exitit indicates whether an
highway exit has been located in the surrounding of city i between t and t1.
 Highways and urban development . 7 of 20

It is worth noting that exit is a point where vehicles can enter or exit a highway and
are defined in terms of catchment area, i.e., the variable takes the value 1 is not only the
town name or the highway, but also for all municipalities served. The definition of the
basins is contained in CERTeT-Bocconi (2006) and consider as treated (i.e., with an
highway exit) cities within 15 km arc distance from an exit. Given this simple definition,
there are 469 catchment areas, consisting of 658 cities. This coding of the variable is
particularly useful because it also captures effects of spatial spillovers. The rationale for
using exit as the treatment variable is that it captures changes in accessibility in a more
compelling way. In fact, at least one highway passes within the administrative
boundaries of 1207 cities, whereas only 658 are considered treated in our sample. The
reason for this difference should be found in the fact that no exit is located (within
15 km) in 549 cities (the difference between 1207 and 658). For those cities, the presence

Downloaded from http://joeg.oxfordjournals.org/ at Rutgers University on August 28, 2015

of a highway on the territory has not meant an increase in accessibility since no direct
access to the infrastructure was provided. Given this rationale, we think at exit as a
sharper indicator of the change in road accessibility.
The geography of highways is probably endogenous with respect to development, so
that, our parameter of interest, , might be biased. The reasons for this is that the
decision concerning the location of exits (as well as the location of a road) is likely to be
driven by actual development or growth potential of cities, so that both reverse
causality and omitted variables may bias econometric estimation. To deal with this
issue, a recent strand of literature has proposed the use of historical instrument to
identify the parameter. In particular, Baum Snow (2007), Duranton et al. (2014), and
Michaels (2008) propose the use of the initial 1947 project for the US highways system
to estimate the impact of highways on a series of outcomes in US cities. In the case of
Italy, the road network has an important antecedent represented by the network of
Roman roads, originally built to provide an efficient postal and communication service
throughout the Empire.
This complex arterial system was the geographical basis on which the Italian
motorway system was designed. Therefore, for the purposes of the present study, we
consider the presence of a Roman road on the urban territory as an instrument for exit,
so that the estimated model was a system of equations:
4yi19512001 ¼ exiti þ controlsi þ "i ; ð4Þ

exiti ¼ controlsi þ $Zi þ i ; ð5Þ

where the instrument is Z ¼ 1½RomanRoad, that is, it takes the value of 1 if a stretch of
a Roman road was located in city i and 0 otherwise. As for controls, we used of the
city’s surface area, altitude, changes in population over the same time period and a full
set of province-specific fixed effects. All data are from Censuses, are at city-level and
were taken from the Atlante Statistico dei Comuni by ISTAT. It should be noted that we
excluded all cities that in 1951 already had a highway exit, hence the ones located on the
Milano–Laghi and Napoli–Capua highways. Furthermore, it must be stated that our
analysis is at city level, not at catchment-area level, that is, our treatment variable is
measured at catchment area level, whereas the outcome variable is at city level.
The estimation of model (4) and (5) implies the acceptance of the identifying
restriction that Z influences the current level of development of the Italian cities only
through the variable exit, conditional on control variables, or through a change in
transport costs. In particular, we shall propose the use of a variable indicating the
8 of 20 . Percoco

existence of a stretch of the Roman roads network under the assumption that this
variable may be correlated with levels of covariates of urban development but it is less
likely to be correlated with variables that predict changes in economic activity,
conditional on control variables. This assumption cannot be tested explicitly, however,
in Section 4 we will discuss extensively the choice of the instrument as well as possible
threats to its validity in Section 5. Finally, because the variable exit is a dichotomous
variable, the estimation of model (4) and (5) is possible only through the estimation of a
linear probability model in the first stage.
We were also interested in estimating the distribution of the effect of highway exits
across sectors within cities and towns. Our hypothesis is that the fundamental player in
the process of firm location is the transport sector. However, through a process of
coagglomeration also firms belonging to other sectors but with high demand for

Downloaded from http://joeg.oxfordjournals.org/ at Rutgers University on August 28, 2015

transport services will preferably locate in cities with a highway exit as in those locations
transport costs are lower and firms in the transport service sector have higher
productivity (Ellison et al., 2010). In such simple framework a pivotal role is played by
the technological linkages between sectors, i.e., by input–output (I–O) coefficients. To
test this simple proposition, we have estimated an equation in the form:
4yis19512001 ¼ ai þ as þ exiti  aTSP
s þ controls þ "is : ð6Þ
where the dependent variable measures the change in the number of plants or in the
number of employees in city i and sector s between 1951 and 2001. Variable aTSP s is the
technological coefficient for transport inputs in sector s as reported in the input–output
tables made available in Rampa (2001). It should be noted that this coefficient has been
computed as in 1959, i.e., at the early available year. Furthermore, we make use, in the
dependent variable, a 95 sector NACE classification, whereas I–O tables are available
with an ESA1979 classification, so that we have made a concordance between the two
classifications and imputed consequently transport technological coefficient. The
technological coefficient in specification (6) measures the dependence of a given sector
on the output of the road transport service sector. Hence, it does not capture precisely
the quantity of transport services effectively used by firms since auto-produced
transport inputs are not accounted for. However, we may reasonably think at transport
technical coefficients as a good measure for the relevance of transport usage for a given
sector.
The rationale behind specification in Equation (6) is that the location of a highway
exit affects positively firms operating in the transport service sector and then, through
them also other firms. In other words, we are testing whether or not there is a co-
location pattern between transport firms and other firms (Ellison et al., 2010).
As regards the aforementioned endogeneity issues, in the case of a city-by-sector
panel, the first-stage regression we have estimated was
aTSP
s exiti ¼ ai þ as þ $Zi  aTSP
s þ controls þ i : ð7Þ
To deal with province-specific specialization patterns, Equations (6) and (7) also include
a series of province–sector interactions as well as the initial level of the dependent
variable.
Table 1 reports descriptive statistics. The employment and plants cumulative growth
rates are both found to be larger in cities with an highway exit. Furthermore, it is
interesting to note that about 25% of cities with an highway exit have had also a
 Highways and urban development . 9 of 20

Table 1. Descriptive statistics

Mean and standard deviation Test of

significance
With highway exit Without highway exit of difference

Cumulative employment growth 0.851 (1.131) 0.715 (1.022) **

Cumulative plant growth 0.640 (0.946) 0.525 (0.719) ***
Distance from an exit (km) 4.67 (3.22) 22.02 (10.03) *
Roman roads 0.253 (0.529) 0.053 (0.252) **

Notes: Signifiance values: ***P50.001, **p50.01, *p50.05.

Downloaded from http://joeg.oxfordjournals.org/ at Rutgers University on August 28, 2015

Roman road, whereas only 5% of cities with Roman road do not actually have highway
access. The table also reports the significance of the test of difference in mean, as well as
the average distance of the city centroid from the nearest highway exit. To be noted is
that the distance is not zero for treated cities as the treatment is defined at the level of
catchment area.

4. The instrument
To deal with the problem of endogeneity in our regression, we make use of an
instrument that measures the presence of an intercity Roman road in the town. The
rationale for such choice is driven by the fact that the geography of local development
in Italy has been shaped by the existence of such a dense network of roads (Maggi,
2009). Furthermore, our exclusionary restriction is that the presence of a stretch of
Roman roads in a city may be correlated with the (initial) level of economic activity, but
it is less likely to be correlated with a determinant of the change in economic outcomes
in a limited time period.
Besides major routes (reported in Appendix A), a vast network of secondary roads
was developed between 4th century BC and 4th century AC. Our instrument is then a
sort of long lag of our endogenous variable and as such it could be considered as
exogenous, conditional on control variables. Roman Empire was extremely centralized
on the city of Rome so that to maintain the control on peripheral areas, a wide network
of roads was built. Before the Romans, several civilization used to build roads to
connect the territory. However, Romans re-organized the space through a well
established network of roads on which they applied new engineering solutions in order
to meet three main features (Basso, 2009):
1. firmitas (solidity of the construction);
2. utilitas (social efficiency of the investment);
3. venustas (monumental appearance).
Roman roads were built to last as for the first time in history, they have been paved
(viae stratae) with stones meant to last forever, in the intentions of the builders
(Berechman, 2003; Staccioli, 2010). In the first period of network construction, i.e., in
the 4th century BC, Roman roads were crucial for the ‘romanization’ of outermost
provinces as they facilitated contacts with Rome. They also contributed to economic
development through trade growth and to urban development through the
10 of 20 . Percoco

Downloaded from http://joeg.oxfordjournals.org/ at Rutgers University on August 28, 2015

Figure 4. Highways and Roman roads in the province of Trento.

concentration of population near principal roads. In a second period (4th–3rd centuries

BC), roads were essential for military expansion in the Italian peninsula. The third
period starts with the investment planned by Augustus between I BC and the beginning
of I AC during which the road network was extended out of Italy and probably reached
120,000 km in length. After Augustus, his successors engaged in a large programme of
maintenance up to 4th century AC in which roads were used only as defense
infrastructure to move armies to the borders. There were three different types of roads
(Basso, 2009): (a) viae publicae, i.e., roads connecting different parts of the urban space;
(b) viae vicinales, i.e., roads connecting different parts of the urban space; and (c) viae
privatae, i.e., roads used to have access to rural areas. We think at the presence of
Roman roads in cities as a measure of their long run accessibility upon which the
economic geography of Italy was built. Once the new technology of highways was made
available, we assume that all cities with a Roman road were eligible to receive the
treatment which in our case is measured through the location of an highway exit. Given
this assumption, the coefficient associated to the exit in our second-stage regression
measures locally an Average Treatment on the Treated and should be thought as the
impact of a reduction of transport costs within the set of compliers, i.e., the cities
already connected with a Roman road. The rationale behind the choice of such
instrument is hence that the network of Roman roads has been important for the spatial
organization of production for a wide range of countries (Von Hagen, 1966). The
increase in the accessibility of certain cities led the early distribution of population and
this, in the long run through forms of path dependence, drove, along with production
factors, the location of firms and industries. Hence, in our empirical analysis, we aim to
verify whether a further change in transport cost has resulted in a change in the
equilibrium and hence in firm relocation.
Figure 4 reports, only as an example, a map of the Province of Trento with the two
highways ‘Brennero–Modena’ and ‘Vicenza–Schio’, as well as traits of the Roman roads.
 Highways and urban development . 11 of 20

For the sake of clarity, it must be stated that Roman road traits within city boundaries
are purely indicative, so that it is not possible to quantify precisely the kilometers of roads
in each city. If we had to use this information as an instrument, we would have incurred
in a serious problem of measurement error. On the other hand, whether a city was located
on a Roman road or not is known without uncertainty. A possible problem with the
proposed instrument is that if Romans located roads to serve fast-growing cities and if
those cities have continued to grow until present days, then our instrument may be not
valid. To deal with this issue, we have excluded from our analysis all province chief-towns
as Roman roads connected the most important cities of the Empire and those cities seem
to be modern chief towns. In doing so, we think that the proposed instrument, although
potentially endogenous for major cities, is exogenous to minor ones. The exclusion of
those cities does not solve the problem of endogeneity, however, as our exclusionary

Downloaded from http://joeg.oxfordjournals.org/ at Rutgers University on August 28, 2015

restriction is that the indicator for Roman road is not correlated with changes in
economic activity in a limited time period, restriction of the sample to relatively minor
cities, increases the likelihood of this assumption.
One of the main concerns on instrument validity is whether it meets the imposed
exclusionary restriction. We have already dropped from the dataset major cities as those
were connected by Roman roads probably because of their geographical advantage
which may have had a substantial role also in subsequent development. However, to
address further concern, in Table 2 we report a number of placebo regressions. In
particular, Models (1)–(4) present linear probability models in which the dependent
variable is Z ¼ 1½RomanRoad and explanatory variables include the growth rate of
employment rate or of plants per capita in two different time windows, i.e., 1951–1971
and 1951–1981. Other control variables are altitude, surface, city population in 1861
and province-specific fixed effects. Standard errors are clustered by province using the
Bester et al. (2011) procedure. Interestingly coefficients are never significant and equal
to 0.01–0.02 in both the whole sample and the sample including only Centre-Northern
cities. Finally, model (5) tests whether the instrument is correlated with the quantity of
non-highway roads in the city. In this case, the dependent variable is the quantity of
kilometers of roads per capita in 2001 and Z ¼ 1½RomanRoad is used as a regressor.
Also in this case we could not find any significant correlation. However, a concern
should be expressed on this last regression as the dependent variable is measured as in
2001 and not in 1951 (to be noted is the fact that it was not used as a control in the main
IV regressions).2
Models (6)–(8) report some further placebo tests for the validity of the instrument.
In particular, our restriction imposes that the existence of a Roman road is
uncorrelated with changes in economic activity affecting the outcomes of interest. An
indirect test of this assumption is given by the correlation between the Roman road
indicator and growth before the extension of the highway network. Model (6)
considers population growth over a very long time period, i.e., 1861–1951, whereas
models (7) and (8) use employment and plants growth, respectively. From models (7)
and (8), all cities treated with an exit or located in a catchment area before 1951 have
been excluded. According to estimates, the low significance of the parameter of

2
To address this last concern, in Section 5.3.2 we consider the possibility of selection into treatment, hence
we avoid the use of instrument to estimate the impact of exit location decision.
12 of 20 . Percoco

Table 2. Placebo regressions

(1) (2) (3) (4) (5) (6) (7) (8)

Roman Roman Roman Roman Roads Population Employment Firm
roads roads roads roads in 2001 growth growth growth
1951– 1951– 1951– 1951– 1861–1951 1951–1961 1951–
1971 1981 1971 1981 1961

Panel A: Whole sample

Employment growth 0.01 0.01
(1.02) (1.12)
Plant growth 0.02 0.02
(1.29) (1.42)

Downloaded from http://joeg.oxfordjournals.org/ at Rutgers University on August 28, 2015

Roman roads 0.01 0.134 0.001 0.002
(0.99) (0.221) (0.023) (0.034)
Observations 7480 7480 7480 7480 7480 7480 7404 7404
R2 0.53 0.51 0.42 0.38 0.29 0.43 0.34 0.41
Panel B: Only Centre-Northern cities
Employment growth 0.01 0.01
(0.99) (0.96)
Plant growth 0.02 0.02
(1.22) (1.23)
Roman roads 0.02 0.194 0.002 0.001
(0.66) (0.211) (0.031) (0.044)
Observations 4154 4154 4154 4154 4154 4154 4128 4128
R2 0.41 0.42 0.59 0.29 0.29 0.44 0.33 0.41

Notes: All models present OLS estimates. All specifications include a constant, surface, altitude, city
population in 1861 and a full set of province-specific fixed effects, although their coefficients are not
reported. Robust standard errors clustered by province according to the Bester et al. (2011) procedure are
in parentheses. Significance values: ***p50.01, **p50.05, *p50.1.

interest for both the whole and the Centre-North samples corroborates the
exclusionary restriction of the empirical model.

5. Results
5.1. Baseline results
In Table 3, estimates of Equation (2) are reported for Italy and for only Centre-
Northern regions, respectively. The opening of an highway exit has not produced any
significant effect on population growth in any of the considered time periods, whereas
the increase in accessibility has generated an increase in the employment growth rate by
3% over the period 1951–2001. As expected, the effect of exit location is stronger for the
period 1951–1981, i.e., during the effective development of the road network, with an
estimated impact of 4%. Results for the growth in the number of plants qualitatively
confirm those obtained in the case of employment, although point estimates are smaller
and amount, on average, to 1%, with a slightly larger effect for the period 1951–1981.
Similar results are obtained when the sample is restricted to Northern cities, as results
might be driven by the low level of development of Southern cities as only a relatively
small portion of which received the treatment, and hence may not be the appropriate
 Highways and urban development . 13 of 20

Table 3. Expansion of the highway network and urban growth (OLS estimates)

Whole sample Only Centre-North

(1) (2) (3) (4)

1951–2001 1951–1981 1951–2001 1951–1981

Panel A: Population growth

Exit 0.00 0.00 0.00 0.01
(0.007) (0.008) (0.009) (0.010)
Observations 7480 7480 4154 4154
R2 0.24 0.50 0.26 0.50
Panel B: Employment growth
Exit 0.03*** 0.04*** 0.04** 0.03***

Downloaded from http://joeg.oxfordjournals.org/ at Rutgers University on August 28, 2015

(0.010) (0.002) (0.024) (0.001)
Observations 7480 7480 4154 4154
R2 0.33 0.52 0.38 0.44
Panel C: Plants growth
Exit 0.01*** 0.02*** 0.01*** 0.02**
(0.004) (0.004) (0.001) (0.010)
Observations 7480 7480 4154 4154
R2 0.55 0.54 0.22 0.58

Notes: In Panel A, dependent variable is average decadal population growth; in Panels B and C, dependent
variables are employment and plants growth, respectively. All specifications present OLS estimates and
include a constant, surface, altitude, city population in 1861 and a full set of province-specific fixed effects,
although their coefficients are not reported. Robust standard errors clustered by province according to the
Bester et al. (2011) procedure are in parentheses. Significance values: ***p50.01, **p50.05, *p50.1.

counterfactual. In general, results are confirmed, although with relatively lower point
estimates with respect to the whole sample.
As argued in previous sections, OLS estimates of Equation (2) may be biased by
endogeneity of the treatment variable. To deal with this issue we have estimated the
system of equations in (4) and (5) by using the location of a Roman road as an
instrument for exit. Table 4 reports estimates of reduced form equations in which the
instrument is used in place of the endogenous variable. As from this table, it seems
verified our assumption of relevance of Roman roads to explain the pattern of growth
of employment and plants.
In Table 5 we report estimates of first stage regressions in which the dependent
variable is exit and is regressed on the same variables as Equation (2) and on Roman
road. In this case, we document a strong correlation between Roman road and exit with
coefficients of magnitude 0.10–0.16. Table 6 reports estimates of second stage
regressions. Interestingly enough, estimates follow the same pattern across sample
periods as in Table 3, as road accessibility improvement has not caused population
growth, whereas the impacts on employment and plants growth are estimated in 5%
and 3%, respectively, in the period 1951–1981.

5.2. The effect on economic structure

Results in the previous section are to be interpreted as average treatment effect across
sectors. However, the decision to locate an exit in a given territory may
14 of 20 . Percoco

Table 4. Road network and urban growth (OLS estimates; reduced forms)

Whole sample Only Centre-North

(1) (2) (3) (4)

1951–2001 1951–1981 1951–2001 1951–1981

Panel A: Population growth

Exit 0.008* 0.001*** 0.002 0.030***
(0.004) (0.0004) (0.006) (0.007)
Observations 7480 7480 4154 4154
R2 0.24 0.50 0.26 0.50
Panel B: Employment growth
Exit 0.0021*** 0.0032*** 0.013*** 0.041***

Downloaded from http://joeg.oxfordjournals.org/ at Rutgers University on August 28, 2015

(0.0002) (0.0001) (0.003) (0.012)
Observations 7480 7480 4154 4154
R2 0.35 0.48 0.37 0.48
Panel C: Plants growth
Exit 0.0038*** 0.0061*** 0.039*** 0.080***
(0.0013) (0.0018) (0.002) (0.022)
Observations 7480 7480 4154 4154
R2 0.34 0.54 0.59 0.61

Notes: In Panel A, dependent variable is average decadal population growth; in Panels B and C, dependent
variables are employment and plants growth, respectively. All specifications present OLS estimates and
include a constant, surface, altitude, city population in 1861 and a full set of province-specific fixed effects,
although their coefficients are not reported. Significance values: ***p50.01, **p50.05, and *p50.1.
Robust standard errors clustered by province according to the Bester et al. (2011) procedure are in
parentheses.

Table 5. Road network and roman roads (first stage regressions)

Whole sample Only Centre-North

(1) (2) (3) (4) (5) (6)

Population Employment Plants Population Employment Plants

Roman road 0.10*** 0.13*** 0.13*** 0.14*** 0.15*** 0.16***

(0.025) (0.025) (0.024) (0.029) (0.034) (0.035)
F-statistics 27.12 29.22 31.49 22.89 27.39 32.55
Observations 7478 7214 7214 4153 3984 3984
R2 0.50 0.47 0.66 0.35 0.51 0.67

Notes: All specifications present first stage estimates and include a constant, surface, altitude, city
population in 1861 and a full set of province-specific fixed effects, although their coefficients are not
reported. Robust standard errors clustered by province according to the Bester et al. (2011) procedure are
in parentheses. Significance values: ***p50.01, **p50.05, *p50.1.

have heterogeneous impacts depending on the initial economic structure and may also
influence the economic structure itself. Table 7 reports estimates of Equations (6) and
(7) in which we assume that the impact of highways on a given sector depends on the
technology of the sector. In Panel A, results for the whole sample indicate that the point
 Highways and urban development . 15 of 20

Table 6. Road network and urban growth (second stage regressions)

Whole sample Only Centre-North

(1) (2) (3) (4)

1951–2001 1951–1981 1951–2001 1951–1981

Panel A: Population growth

Exit 0.082 0.014 0.01 0.03
(0.073) (0.213) (0.058) (0.172)
Observations 7480 7480 4154 4154
R2 0.23 0.50 0.22 0.29
Panel B: Employment growth
Exit 0.04*** 0.05*** 0.05*** 0.05***

Downloaded from http://joeg.oxfordjournals.org/ at Rutgers University on August 28, 2015

(0.011) (0.002) (0.002) (0.002)
Observations 7480 7480 4154 4154
R2 0.44 0.55 0.59 0.59
Panel C: Plants growth
Exit 0.04** 0.03** 0.04*** 0.04***
(0.016) (0.012) (0.012) (0.014)
Observations 7480 7480 4154 4154
R2 0.35 0.41 0.44 0.42

Notes: In Panel A, dependent variable is average decadal population growth; in Panels B and C, dependent
variables are employment and plants growth, respectively. All specifications present IV estimates and
include a constant, surface, altitude, city population in 1861 and a full set of province-specific fixed effects
although their coefficients are not reported. Robust standard errors clustered by province according to the
Bester et al. (2011) procedure are in parentheses. Significance values: ***p50.01, **p50.05, *p50.1.

Table 7. Road network and urban growth (second-stage regressions)

Whole sample Only Centre-North

(1) (2) (3) (4)

1951–2001 1951–1981 1951–2001 1951–1981

Panel A: Employment growth

Exit 0.10** 0.14** 0.14*** 0.12**
(0.04) (0.05) (0.04) (0.05)
Observations 225,766 225,766 175,299 175,299
R2 0.54 0.59 0.58 0.59
Panel B: Plants growth
Exit 0.12* 0.16*** 0.11* 0.12***
(0.05) (0.05) (0.05) (0.04)
Observations 225,766 225,766 175,299 175,299
R2 0.54 0.62 0.29 0.63

Notes: In Panels A and B, dependent variables are employment and plants growth, respectively. All
specifications present IV estimates and include a full set of city-specific fixed effects and province–sector
interaction dummies, although their coefficients are not reported. Robust standard errors clustered by
province according to the Bester et al. (2011) procedure are in parentheses. Significance values: ***p50.01,
**p50.05, *p50.1.
16 of 20 . Percoco

estimate is 0.10 for the period 1951–2001 and 0.14 for the period 1951–1981. The
technical coefficient for transport services ranges from 0.0015 for the Real Estate sector
to 0.1282 for Postal services, implying that the location of an highway exit increased the
growth rate of employment in the Real Estate sector by about 0.02%, whereas the
Postal services sector benefits by 1.68% increase in employment. Similar results were
found also in the case of growth in the number of plants and in models for which we use
only Centre-Northern cities.
If we assume that tradable sectors have larger technical coefficients for transport
services, hence, our results indicate a larger impact of a reduction in transport costs on
the production of tradables. Furthermore, given the substantial similarity between
estimates using the whole sample of data and those for only Centre-Northern cities, we
could not find differential effects of the highways across the country. These results seem

Downloaded from http://joeg.oxfordjournals.org/ at Rutgers University on August 28, 2015

hence to confirm the positive role played by highways for urban development and that
this effect acts through the co-location of firms with firms operating in the transport
service sector. In the following sub-section we provide a number of robustness checks
for our general models.

5.3. Robustness checks and extensions

5.3.1. Difference-in-difference estimates and restrictions of the sample
As a first extension and check of the robustness of our results, we exploit the full
temporal variation in our sample. To this end, we have estimated a slightly different
model in the form:
yit ¼ ap þ dt þ exitit þ controlit þ "i ; ð8Þ
and
yist ¼ ap þ as þ dt þ exitit  aTSP
s þ "is ; ð9Þ
where the treatment variable is 1 after the treatment and zero before or otherwise. Both
models have been estimated via IV regressions over different time periods: 1951–2001,
1951–1981, 1951–1971, and 1971–2001. Furthermore, it should be noted that, by using a
difference-in-difference approach as in Equations (8) and (9), we control for unobserved
heterogeneity through province fixed effects, p, capturing all provincial time-invariant
potential confounders.
Panel A in Table 8 reports estimates of both model (8) and (9) for the whole sample.
Interestingly enough, we have found that the average treatment on the treated generally
decreases over time in most of the cases, both for the employment rate and for the
number of plants. This result points at a confirmation of the positive impact of
highways on local economic growth in an order of magnitude of 4–5% for employment
growth and 2–3% for plants. In panel B results of models (8) and (9) with city-specific
time trends are reported. In this case, a reduction in the coefficients is detected, along
with a decrease in statistical significance. However, these estimates should be considered
with caution since the statistical power of the regressions is relatively low, given the
large number of cities in the sample.
Table 9 reports some further robustness checks for the whole sample. In particular, in
the first part of the table, models have been estimated on a sub-sample of cities
all crossed by a highway stretch but not necessarily treated with an exit. Instrumental
 Highways and urban development . 17 of 20

Table 8. Robustness check: Difference-in-difference

Aggregate outcomes Sector-based outcomes

Employment Plant Employment Plant

Panel A: Baseline regressions

1951–2001 0.05*** 0.03*** 0.11*** 0.12***
(0.012) (0.011) (0.011) (0.001)
1951–1981 0.03*** 0.03*** 0.11*** 0.11***
(0.011) (0.011) (0.000) (0.001)
Panel B: Models with city-specific trends
1951–2001 0.05** 0.02** 0.12*** 0.11**
(0.021) (0.010) (0.015) (0.045)

Downloaded from http://joeg.oxfordjournals.org/ at Rutgers University on August 28, 2015

1951–1981 0.04** 0.02* 0.12** 0.11**
(0.018) (0.015) (0.013) (0.045)

Notes: All models are estimated via IV regressions. Specifications for the aggregated outcomes include a
constant, surface, altitude, city population in 1861 and province-specific dummies. Specifications for sector-
based outcomes include a full set of city-specific fixed effects and province–sector interaction dummies. In
panel B, city-specific trends are included. Robust standard errors clustered by province according to the
Bester et al. (2011) procedure are in parentheses. Significance values: ***p50.01, **p50.05, *p50.1.

Table 9. Robustness checks (growth in 1951–2001)

Only cities with highways (second stage) Only cities with Roman roads (OLS)

(1) (2) (3) (4) (5) (6)

Population Employment Plants Population Employment Plants

Exit 0.01 0.06*** 0.05*** 0.01* 0.02** 0.03***

(0.031) (0.013) (0.014) (0.004) (0.01) (0.01)
Observations 1047 1047 1047 792 792 792
R2 0.20 0.57 0.69 0.35 0.31 0.47

Notes: All specifications include a constant, surface, altitude, city population in 1861 and a full set of
province-specific fixed effects, although their coefficients are not reported. Robust standard errors clustered
by province according to the Bester et al. (2011) procedure are in parentheses. Significance values:
***p50.01, **p50.05, *p50.1.

variable estimates of the coefficient of interest are slightly larger than the ones in
Table 6 with very similar significance. These results are important since they
corroborate the choice of the instrument by excluding eventual correlation with the
presence of a highway over the whole sample.3
The second part of the table presents OLS models of the sub-set of compliers, i.e.,
only cities located on a Roman road. Also in this case, results are confirmed from the
point of view of significance, although with smaller coefficients.

3
Further evidence on this point will be given with models in Table 10.
18 of 20 . Percoco

Table 10. Oaxaca–Blinder regression estimates

(1) (2) (3)

1951–2001 1951–1981 1951–1971

Panel A: Employment growth–whole sample

Exit 0.02 0.03 0.03
(0.001) (0.001) (0.001)
Panel B: Plant growth–whole sample
Exit 0.02 0.02 0.03
(0.003) (0.006) (0.001)

Note: Oaxaca–Blinder regressions include a constant, surface, altitude, city population in 1861 and a full set

Downloaded from http://joeg.oxfordjournals.org/ at Rutgers University on August 28, 2015

of province-specific fixed effects.

5.3.2. Selection into treatment as a treat to identification

Although in our regressions we control for a number of predetermined variables, some
pre-treatment differences may still remain unobservable (although this concern is minor
in the case of difference-in-differences estimates). Furthermore, one may argue that our
instrument could be not completely exogenous because of morphological characteristics
of territory. To deal with those issues, we have estimated counterfactual changes for
treated cities via Oaxaca–Blinder regressions (Kline, 2011).
The procedure runs as follows. We first fit regression model to control cities of the
form 4yi19512001 ¼ a þ controlsi þ "i . We then use the vector estimated coefficients
for pre-program characteristics to predict the counterfactual mean for the treated cities.
The Oaxaca–Blinder regression has the advantage over standard regression methods
of identifying the average treatment on the treated in the presence of treatment effect
heterogeneity, hence it may account for another potential source of endogeneity, such
as unobserved heterogeneity. In addition, Kline (2011) has proven that the Oaxaca–
Blinder regression method can be interpreted as a propensity score reweighting
estimator, so that it accounts also for selection into treatment.
Table 10 reports estimate for the average effect across sector for the whole sample.
Interestingly enough, also in this case, results are largely confirmed with estimates very
close to the ones obtained by standard IV regressions in Section 5.1. However, although
in the first stage we have used exactly the same regressors as in model (4), treatment
effects obtained via Oaxaca–Blinder regressions are generally very sensitive to model
specification.

6. Conclusion
The reduction of transport costs is at the heart of major policy interventions, although
their effect is not always clear. In this article, we have analysed an important policy
experiment in Italy, namely the construction of the whole highway network.
In particular, we have focused on the location of highway exit in Italian cities as a
policy treatment and conducted an econometric analysis in which the growth rates
of population, employment rate and plants per capita have been considered to be
the main outputs. To take into account the endogeneity of the location decision of
exits, we have coded the geography of ancient Roman roads and used it as an
 Highways and urban development . 19 of 20

instrument. Our results indicate that access to an highway for a city has a positive
impact on urban development, in terms of employment growth (þ4–5%) and firm entry
(þ2–3%).
However, it should be noted that we cannot disentangle redistribution from global
growth, that is our results might be driven by business relocation. Furthermore, we have
found that one of the transmission channels through which a decrease in transport costs
works is an improvement in the competitive advantage of the transport service sector
which, through co-location patterns, propagates to other sectors.

Acknowledgements

Downloaded from http://joeg.oxfordjournals.org/ at Rutgers University on August 28, 2015

I am grateful to Gianmarco Ottaviano, Michel Serafinelli, Dirk Stelder and audiences at the
Università di Modena e Reggio Emilia, LSE, 2012 NARSC Conference in Ottawa, the First
Meeting on Transport Economics and Infrastructure in Barcelona for useful comments.
Francesca Cattaneo and Francesca Scaturro provided superb research assistance.

References
Basso, P. (2009) Strade romane: Storia e archeologia. Roma: Carocci.
Baum Snow, N. (2007) Did highways cause suburbanization? Quarterly Journal of Economics,
122: 775–805.
Berechman, Y. (2003) Transportation—economic aspects of Roman highway development: The
case of via appia. Transportation Research A, 37: 453–478.
Bester, C. A., Conley, T., Hansen, C. (2011) Inference with dependent data using cluster
covariance estimators. Journal of Econometrics, 165: 137–151.
Bleakley, H., Lin, J. (2012) Portage and path dependence. Quarterly Journal of Economics, 127:
587–644.
Cardinale, A. (2000) Aspetti sociologici dell’organizzazione del lavoro. Da Adam Smith alla
globalizzazione dell’economia, CUEM, Milano.
CERTeT-Bocconi (2006) Rapporto AISCAT—L’impatto territoriale dei caselli autostradali:
Un’indagine sui comuni, Università Bocconi.
Donaldson, D. (2013) Railroads of the Raj: Estimating the impact of transportation
infrastructure. American Economic Review, forthcoming.
Duranton, G., Turner, M. (2012) Urban growth and transportation. Review of Economic Studies,
79: 1407–1440.
Duranton, G., Morrow, P., Turner, M. (2014) Roads and trade: Evidence from the US. Review of
Economic Studies, 81: 681–724.
Ellison, G., Glaeser, E., Kerr, W. (2010) What causes industry agglomeration? Evidence from
coagglomeration patterns. American Economic Review, 100: 1195–1213.
Fujita, A., Mori, T. (1996) The role of ports in the making of major cities: Self-agglomeration and
hub-effect. Journal of Development Economics, 49: 93–120.
Gibbons, S., Lyytikäinen, T., Overman, H. G., Sanchis-Guarner, R. (2012) New road
infrastructure: The effects on firms. LSE Spatial Economics Research Centre SERCDP0117.
London School of Economics Spatial Economics Research Centre.
Kline, P. (2011) Oaxaca–Blinder as a reweighting estimator. American Economic Review: Papers
and Proceedings, 101: 532–537.
Maggi, S. (2003) Storia dei trasporti in Italia. Bologna: il Mulino.
Maggi, S. (2009) Storia dei trasporti in Italia. Bologna: il Mulino.
Martin, P., Rogers, C. A. (1995) Industrial location and public infrastructure. Journal of
International Economics, 39: 335–361.
Menduni, E. (1999) L’Autostrada del Sole. Bologna: Il Mulino.
Michaels, G. (2008) The effect of trade on the demand for skill: Evidence from the Interstate
Highway System. Review of Economics and Statistics, 90: 683–701.
20 of 20 . Percoco

Rampa, G. (2001) Yearly series of input–output tables (ESA1979) for the Italian economy, 1959–
1987. Rivista Internazionale di Scienze Sociali, 59: 449–478.
Staccioli, R. A. (2010) Strade romane. Rome: L’Erma di Bretschneider.
Von Hagen, V. W. (1966) Roman Roads. New York: World Publishing.
World Bank, (2009) World Development Report 2009. Reshaping Economic Geography,
Washington: World Bank.

Appendix
A. List of Roman roads
The complex of Roman roads in Italy consisted of 13 principal routes (Basso, 2009): (1)
The Appian Way was built in 312 BC by the Consul Appius Claudius was initially

Downloaded from http://joeg.oxfordjournals.org/ at Rutgers University on August 28, 2015

drawn to Capua to be then extended to Beneventum, Venosa, Tarantum and
Brundisium. (2) The Via Aemilia, which was the continuation of the Via Flaminia
north-west, joining with Ariminum Placentia, touching Caesena, Forum Livi, Bononia
Mutina, Regium Lepidum and Parma. (3) The Way of Capua–Rhegium, which, starting
from the Via Appia to Capua, went up to Rhegium, through Consentia and Vibo
Valentia. (4) The Via Aurelia, which connected Rome to Go Sabatia (Vado Ligure),
through Pisae, Moon and Genua. (5) The Way Domitiana that is separated from the
Via Appia in Sinuessa (Mondragone) and finally to Neapolis. (6) The Popilia-Via Annia
was another continuation of the Via Flaminia and started from Ariminum through
Ravenna, Atria (Adria), Patavium (Padua), Altinum, Aquileia, Tergeste (Trieste). (7)
The Latin Way connected Rome to Capua through Anagnia, Frusino and Casinum. (8)
The Via Flaminia linking Rome with Ariminum (Rimini), touching Fanum Fortunae
(Fano) and Pisuarum. (9) The Via Salaria was the main track on which ran the supply
and sale of salt, one of the most valuable materials for the time. The route started from
Rome and came to Castrum Truentinum (Porto d’Ascoli), and through Reate Asculum.
(10) The Way Postojna joined Genua and Aquileia, through the Po Valley. (11) The Via
Valeria connected the Roman coast to Aterni (Pescara), past Tibur (Tivoli) and Teate
Marrucinorum (Chieti). (12) The Via Cassia, linking Rome to northern Italy, touching
Arretium, Florentia, Pistoia and Luke. (13) Finally, the Via Clodia connected Rome to
Saturnia.