Dopeso Fernandez 2011

MARSHALLIAN AGGLOMERATION ECONOMIES AND
ENTREPRENEURSHIP: THE SPANISH CASE
ROBERTO DOPESO-FERNÁNDEZ
Department of Applied Economics
Universitat Autonoma de Barcelona
roberto.dopeso@uab.cat
Tel. (+34)935811153 Fax (+34)935812292
This paper analyzes to what extent Marshallian agglomeration economies affect the creation
of new entrepreneurial ventures at the metropolitan level. The measuring of agglomeration
economies is based on the construction of indexes using the methodology of Glaeser and Kerr
(2009). The indexes attempt to capture the effects of resource sharing, labor matching and
knowledge spillovers according to the taxonomy proposed by Marshall (1920). Also an index
to measure the influence of small suppliers to attract new business ventures, following Chinitz
(1961), is constructed. Data on new firms and employment generated is accounted for the
period 2000-2008. The analysis is based on the activity of the 15 largest metropolitan areas in
Spain. Sixty industries at two-digit CNAE-93 are considered. The results show that jobs
created by entrepreneurs are highly influenced by the ability to share suppliers and customers.
Firm creation is influenced by those factors as well as the presence of small suppliers and the
proximity to innovative activity. Agglomeration indexes with sector and city fixed effects
explain more than 90% of new entry and employment generated. The potential
multicollinearity among indexes is tested using principal component analysis. This analysis
shows some complementarities among the indexes. New regressions using the factorized
terms show that traditional measures of localization economies hide specific information
about the process of agglomeration.
JEL Classification: J23, L26, L60, L70, L80, L90, O31, R39.
Key Words: Entrepreneurship, Industrial Organization, Agglomeration, Labor Markets, Input-

Output Flows, Innovation.
1. INTRODUCTION
A large percentage of the activity occurs around cities. In the Spanish case, the five largest
metropolitan areas account for 35% of population and 38% of total employment (Boix and
Veneri, 2009). The effects of agglomeration cause that new ventures will be close to the old
ones. This shows the need to measure the impact of agglomeration for policy implications.
During the past 30 years a renewed interest in entrepreneurial activity has been shown but
there is still much scope to cover. The Spanish case is particularly interesting because the
Electronic copy available at: http://ssrn.com/abstract=2099183

need for greater international competitiveness could have its engine in entrepreneurship.
Despite this, evidence shows that the majority of new Spanish firms are in low skill sectors,
mainly the building industry and real estate services.
The aim of this work is two folded. It contributes to the large literature of agglomeration
economies and also to the new but increasing literature on entrepreneurs. With respect to
agglomeration economies the research constructs indexes that capture the essence of the
advantages of agglomeration first proposed by Marshall in 1890. With respect to
entrepreneurship economics the purpose is to explain the location behavior of entrepreneurs
(measured as firm births and new employment generated by them) through the determinants
of agglomeration economies in the 15 largest metropolitan areas in Spain. These cities
account for 51% of the population and 53% of jobs in Spain according to 2001 census data.
The importance of entrepreneurs is also present in this work by testing the Chinitz (1961)
hypothesis. This hypothesis states the importance of small suppliers and entrepreneurs as
integral part for the business dynamism of a city. Another important feature is to add richness
to the industrial analysis, especially by incorporating the service sector. This sector has
received little attention in the literature of entrepreneurship location (as compared to
manufacturing) despite the fact that it concentrates more than 60% of economic activity with
an increasing trend over the past 10 years according to INE (Spanish National Institute of
Statistics) data for the Spanish economy.
The methodology follows the work by Glaeser and Kerr (2009) for the manufacturing
sector in the United States. Regressions were estimated by ordinary least squares using fixed
effects for sectors and cities. To solve endogeneity concerns an estimation based on natural
cost advantages (Ellison and Glaeser, 1999) was used with little success for the Spanish case.
Evidence suggests that Marshallian agglomeration economies (especially sharing with
customers and knowledge spillovers) and the presence of small suppliers are significant for
entrepreneurial activity. Principal component analysis was used to test co-dependencies
between indexes employed to measure Marshallian agglomeration economies. It is also useful
to improve the efficiency of the estimation using orthogonal regressors.
The rest of the paper is organized as follows. Section 2 presents the relevant literature
concerning the study of entrepreneurs and the agglomeration economies. Section 3 presents
the measuring problem, the dataset, the construction of the indexes and the model. In section
4 the main econometric results are presented and discussed. In section 5 the conclusion is
presented with a summary, complications and possible solutions.
Electronic copy available at: http://ssrn.com/abstract=2099183

2. BACKGROUND
This section presents the relevant literature on the subject. The review is organized in four
sub-sections. The first discusses some general information about the Spanish market structure
and entrepreneurship. The second shows the progress made on the literature of the
entrepreneurs. The third analyzes the developments made in the measurement of
agglomeration economies with particular emphasis on Marshallian agglomeration economies.
The fourth shows the related empirical research.
2.1. STATE OF ENTREPRENEURSHIP IN SPAIN

Entrepreneurship in Spain had been like in the majority of the countries the motor of
employment no matter that more than 90% of the firms have less than five employees. Recent
evidence also suggest an stronger agglomeration effect from small or medium business rather
than from biggest ones (Rosenthal and Strange, 2010). Altough this information is relevant
the majority of the evidence on agglomeration is indirect for the Spanish case, or it is focused
on the manufacturing sector. This work tries to adress this problem by focusing on
entrepreneurship and agglomeration in Spain including all industries in the economy, and
measuring agglomeration in terms of localization, urbanization, customer – suppliers linkages
(including the relevance of small size suppliers), labor pooling and knowledge spillovers.
Before examining the theoretical background it is relevant to have a general look at
entrepreneurship in Spain. The tool for such purpose will be the DIRCE (DIRectorio Central
de Empresas) database elaborated by INE. Entrepreneurship is measured by patterns of firm
entry. This database allows quantifying the number of births as well as its entry size and its
geographical and industrial composition. The dataset excludes the following industries:
agricultural activities, livestock, fisheries, public administration, activities of households as
employers and activities of extraterritorial organizations and bodies.
Table 1 shows the most relevant information of DIRCE database for the period 2000-
2008 and the general information presented by SABI1. In the first column the mean and
standard deviation of new companies are presented, the second column shows the data for all
enterprises and the third shows the total enterprises in SABI. As we can observe new entrants
account 12% of total companies. It is worth noticing the strong self-employment in Spain; it
represents over 70% of new initiatives and more than half of the established companies. This
is reflected in the large gap found between per company employees for new businesses (1.24)
1
SABI (Sistema de Análisis de Balances Ibéricos) is the database used for the econometric estimation. The data
is presented to contrast the representativeness of the database.
3
and established ones (6.10). The Building and Other Services sectors show a greater
participation among the new companies compared to the established firms. This phenomenon
is consistent with the Spanish property boom (18% of Other Services sector entrants
concentrates on sector 70, real estate activities) leading to a rise in firm births that have had a
short life and are now forced to close. The geographical structure of the Autonomous
Communities (NUTS-2) shows that approximately 60% of new entrants are concentrated in
the four main geographical areas2. These geographical areas represent 58% of the population
therefore a weak agglomeration effect is present. This research includes the main cities in
these areas to corroborate these findings.
Table 1
Entrepreneurship activity descriptive statistics by DIRCE (2000 - 2008) SABI
New Entry Total Enterprises Total Enterprises
Mean S.D. Mean S.D. Media
Mean annual counts 347,536.22 36,611.60 2,967,141.00 301,664.93 1301678
Mean annual employees 430,931.67 29,130.66 18,098,202.78 1,759,698.44 -
Mean annual Employees per Firm 1.24 0.20 6.10 0.15 -
By size
No Employees 71.78% 8.21% 51.91% 4.36%
1-5 Employees 23.99% 1.76% 37.34% 4.76% 92.60%
6-9 Employees 2.30% 0.12% 4.74% 0.57%
10-19 Employees 1.27% 0.07% 3.29% 0.32%
+20 Employees 0.65% 0.08% 2.72% 0.27% 7.40%
By Sector
Mining 0.05% 0.01% 0.10% 0.00% 0.04%
Manufacturing 5.07% 0.66% 8.01% 0.12% 23.97%
Energy 0.30% 0.32% 0.16% 0.08% 2.48%
Construction 17.47% 3.23% 13.29% 2.53% 33.42%
Comerce 23.40% 0.71% 27.65% 0.75% 6.21%
Accommodation and Food Service 10.78% 0.42% 9.31% 0.36% 2.66%
Transport 5.08% 0.48% 7.88% 0.26% 1.93%
Rest of Services 37.85% 6.63% 33.60% 6.34% 29.30%
By Region
Andalucía 16.45% 2.34% 14.99% 1.84% 13.48%
Aragón 2.60% 0.43% 2.89% 0.21% 3.06%
Asturias (Principado de) 2.11% 0.48% 2.27% 0.13% 1.77%
Balears (Illes) 3.05% 0.49% 2.75% 0.31% 2.90%
Canarias 4.47% 0.50% 4.16% 0.46% 3.52%
Cantabria 1.09% 0.18% 1.20% 0.10% 0.82%
Castilla y León 4.32% 0.48% 5.31% 0.33% 4.38%
Castilla - La Mancha 3.82% 0.65% 3.88% 0.48% 3.82%
Cataluña 18.20% 3.05% 18.55% 1.64% 20.33%
Comunitat Valenciana 11.56% 1.83% 10.80% 1.26% 11.90%
Extremadura 2.10% 0.46% 1.97% 0.24% 1.44%
Galicia 5.37% 0.48% 6.12% 0.49% 6.13%
Madrid (Comunidad de) 15.02% 1.93% 14.71% 1.90% 17.18%
Murcia (Región de) 3.12% 0.58% 2.78% 0.40% 2.84%
Navarra (Comunidad Foral de) 1.45% 0.34% 1.33% 0.12% 1.18%
País Vasco 4.42% 0.42% 5.33% 0.26% 4.49%
Rioja (La) 0.64% 0.09% 0.71% 0.06% 0.63%
Ceuta y Melilla 0.23% 0.01% 0.24% 0.01% 0.12%
Source: DIRCE and SABI databases
2
Based on population the four main geographical areas in Spain are Andalucia, Cataluña, Madrid and Valencia
4
The above analysis is an interesting snapshot of entrepreneurship in Spain but a closer
look to the data will help to understand the dynamics of entrepreneurial activity. Figure 1
shows the differences between growth rates for births, deaths and all enterprises. Data is
normalized with respect to the first period. Although all series show an increasing trend
differences in volatility are quite evident. The total number of firms grew by 32% during this
period. Firm births show a similar trend but are more volatile and negatively affected by
periods of uncertainty. Firm deaths have been reduced at the beginning of the period but
uncertainty between 2006 and 2008 has altered their behavior. Firm deaths show a very steep
increase as a result of economic turmoil.
Figure 1: Spanish Firms Dynamics
1.4
1.3
1.2 Total
1.1 Births
1 Deaths
0.9
0.8
2000 2001 2002 2003 2004 2005 2006 2007 2008
Figure 2 shows the evolution of firm births by industry. The manufacturing industry
has been gradually losing importance, accounting for 6% of total births at the beginning of the
period and only 4% in 2008. Despite its relative decreasing tendency, the industrial sector is
the only that presented entrepreneurial growth during the economic recession. The number of
births in the building industry has grown by 47% over the period. Building industry has
grown 24% with respect to the other sectors. It is important to notice that the crisis in the
building sector has begun to quickly erode these figures. The service sector represents over
75% of all new businesses. Service industry declines at the beginning and end of the period
but shows a general upward trend.
5
Figure 2: Births by sectors
1.65
1.55
1.45
1.35
1.25 Industry
1.15 Construction
1.05
Services
0.95
0.85
0.75
2000 2001 2002 2003 2004 2005 2006 2007 2008
Figure 3 displays the evolution of new companies by size. The case of self-
employment draws attention again. Its high volatility seems to be the main cause of the
unstable behavior observed in Figure 1. Firm births with 1-5 and 6-9 employees present a
very similar behavior. During the first half of the period they appear to take participation out
of self-employment but since 2005 this trend was reversed. Companies with 10 to 19
employees have remained around the average for the period. Entrepreneurship in companies
with more than 20 employees has been declining steadily during the period.
Figure 3: Births by size
1.25
1.15
No employees
1.05 1-5 employees
6-9 employees
0.95
10-19 employees
0.85 More than 20 employees
0.75
0.65
2000 2001 2002 2003 2004 2005 2006 2007 2008
2.2. ENTREPRENEURS IN ECONOMIC THEORY

The study of entrepreneurs is far from being new in economic theory. In 1755 Richard
Cantillon in his "Essai sur la nature du commerce en général" recognizes the entrepreneur as
the agent who takes risks for profit unlike the land-owners or workers who receive fixed-rents
6
as a result of their endowments. In economic terms Cantillon defined the entrepreneur as an
arbitrator (Iversen et. al. 2008). The entrepreneur solves frictions and matches supply and
demand. The English school took the "entrepreneur” concept from the French school but
translated its meaning to "adventurous ", "glider" or "director ". Adam Smith’s concept of
entrepreneur was not distinguishable from the company owner (Hebert and Link, 2006).
Alfred Marshall reconciles both views. The entrepreneur is both risk taker and
administrator. He defines the entrepreneur as the responsible of shifting the production
possibility frontier of the company, identifying opportunities, reducing costs and thereby
increasing production. Marshall also made a distinction between entrepreneurs dividing them
in two categories: the actives "those who opened or improved ways of doing business" and the
passives "those who follow existing roads" (Marshall, 1920, p597). Other relevant theorists3
have addressed the issue but never found a way to formulate it explicitly.
It was not until the last two decades of the twentieth century that the entrepreneur
started to be treated explicitly in economic conventional models. Baumol noted the lack of
treatment of the subject among his colleagues in a dramatic way "The theoretical firm is
entrepreneurless the Prince of Denmark has been expunged from the discussion of Hamlet"
(Baumol, 1968, p. 66) and urged to incorporate the influence of these agents in the economic
models. In the first models developed the agents differ regarding their "entrepreneurial
ability". This variable entered into the cost function (Jovanovic, 1982; Brock and Evans,
1985) or in the production function (Evans and Jovanovic, 1989; Holmes and Schmitz, 1990).
Although these first attempts were very important for the development of a more complete
theory of the entrepreneurial phenomenon they suffered some problems. Most of them used
the number of self-employed which is not the best measure of entrepreneurship (Acs and
Szerb, 2009), but laid the groundwork for the emergence of large number of papers and
journals interested in the subject.
2.3. AGGLOMERATION ECONOMIES

It is impossible to find a plausible explanation for the organization of the population around
large urban centers that has occurred in the last 150 years without considering some form of
agglomeration economies or increasing returns to scale (Duranton and Puga, 2004). It is this
mystery that has led to the growth of a large literature trying to explain this phenomenon.
Based on the new economic geography developed from Krugman’s (1992) work that focuses
3
See for example the work of David Ricardo, John Stuart Mills, Jean Baptiste Say or the German school
(Thünen and Mangoldt)
7
on the importance of space, transport costs and analytical theoretical models for study (see
Fujita, Krugman and Venables, 1999) some conclusions on agglomeration economies were
obtained. Other group of studies were focused on the economic similarities and differences
between different geographical areas (see Rosenthal and Strange, 2001) based on a more ad
hoc framework. Some research has been done at urban level studying the economic
implications of cities (see Rosenthal and Strange; 2003).
Usually, economies of agglomeration refer to the phenomenon of increasing returns as
a result of the accumulation of resources in a geographic location. The concept of
agglomeration is attributed to the seminal works of Marshall and Weber. Two main forms of
agglomeration can be distinguished: (1) Urbanization economies, resulting from the
accumulation of population at one specific place, are generally related with the work of
Jacobs (1971); and (2) Localization economies, those resulting from the accumulation of an
specific industry or sector developed by Marshall (1920). It is important to recognize that
these agglomeration economies are also external (exogenous to individuals whether they are
agents or firms). External agglomeration economies are divided into three categories
depending on whether they come from interactions in labor markets, from links with suppliers
and costumers, and those from knowledge spillovers according to the taxonomy proposed by
Marshall (1920). That is the reason why throughout this work they are called Marshallian
agglomeration economies. Although first proposed by Marshall, the concept in this paper is
more related with the micro-foundations of urban agglomeration economies developed by
Duranton and Puga (2004) under the headings of sharing, matching and learning.
2.4. EMPIRICAL EVIDENCE

The importance of entrepreneurs to influence the composition and activity of urban areas was
noted by Chinitz (1961). He highlighted the importance of small suppliers to explain the
differences existing in the development of New York and Pittsburgh in the United States. A
city with smaller enterprises will have more entrepreneurs per unit of production than an
oligopolistic one. A feedback effect exists so entrepreneurs are grouped into those areas that
are most conducive to their development. If the entrepreneurs are located in intermediate
industries the flow of entrepreneurship will be spread to other industries. This causes such
cities to be more dynamic and adaptable because the detection of opportunities is more
diversified.
A good starting point to understand the effects of agglomeration in new ventures is the
work by Rosenthal and Strange (2003). They used a rich database to test the effects of urban
8
and localization economies at the Zip Code level for the United States. Their findings support
the existence of localization economies. A later work (Rosenthal and Strange; 2005) used the
same methodology focusing on New York Metropolitan Area. The number of new
establishments with three years or less and the employment generated are used as dependent
variables. The geographical scope is very specific at the census tract level. A Tobit model is
used. They control for urbanization and localization economies using the total number of
workers at different distances form the census tract centroid. Their results show significative
effects to both forms of agglomeration economies but the magnitudes diminish4 strongly as
the distance from centroids increases.
For the Spanish case there are a number of studies that allow us analyze the effect and
extent of agglomeration economies. Holl (2004) conducted a study to measure the impact of
new road infrastructure in Spain. Ten manufacturing sectors were analyzed. Population
(urbanization economies) as well as indexes of specialization (localization economies) were
used as controls. Her results show positive and significant economies of urbanization. The
evidence is more diverse on economies of localization since positive, negative and non
significant effects are present depending on the industry. Arauzo (2005) studies the
determinants of manufacturing location in Catalonia. He finds a positive significant effect for
economies of urbanization, a significant negative effect of diseconomies5 of urbanization and
a significant negative effect of economies of localization. Both studies are interesting because
they measure the impact of agglomeration economies in Spain. However, they only focus on
the manufacturing sector.
Some evidence regarding Marshallian agglomeration economies in the literature is
more indirect. Sharing can be inferred by the work of Holmes (1999). His work shows that
outsourcing increases with the concentration level of industries. Helsley and Strange (1990)
found evidence of labour matching. Their work shows the relation between the agglomeration
of employment and enterprises. Another evidence of labor matching can be found in the paper
by Costa and Kahn (2001). They demonstrate that “Power Couples”6 tend to locate with more
probability in big cities to solve problems of job searching. Glaeser and Mare (2001) found
that workers in cities earn bigger salaries. Nonetheless, it is not clear if the reason is market
coordination. Matching for the case of Spain can be inferred from the work of Arauzo-Carod
4
For the Wholesale Trade and Business Services industries the localization economies effect increases instead of
diminishing with distance.
5
The diseconomies of urbanization are measured as the square of the urbanization economies (workers per
square kilometre)
6
Marriages in which both members work and both have at least a collage degree
9
et. al. (2009). This work shows a positive relation between employment formation and the
concentration of one year old new ventures. The evidence supports the effect of
entrepreneurship in labor creation but the matching relation is not clear. The extensive
literature on entrepreneurship and innovation give more evidence for the learning effect.
Audretsch and Feldman (1996) found that innovations are concentrated spatially and
innovative industries are concentrated geographically. Using a Cobb-Douglas framework for
German Kreise Audretsch and Keilbach (2004) show that entrepreneurial capital fosters
productivity. In Spain, Segarra (2007) found evidence that Catalonian companies benefit from
R&D spillovers thus concentrating geographically in the manufacturing sector. This spillover
effect is negative for high-tech companies highlighting that for some knowledge-intensive
services the scope of learning can be extended beyond greater geographical areas, so
agglomeration is discouraged.
The first attempt to measure directly and jointly Marshallian agglomeration
economies can be found in the work of Rosenthal and Strange (2001). They use the
concentration index proposed by Ellison and Glaeser (1997) for the manufacturing industry as
dependent variable. They also calculate the spatial influence at country, state and zip code
level. To capture sharing they use the proportion of manufactured and non manufactured
inputs divided by unitary transport costs. Matching is calculated using three proxies of labor:
productivity, the ratio of management to production workers and the educative level. To
measure the knowledge spillovers they use the number of innovations. Matching is positive
related to agglomeration at the three geographical levels. Knowledge spillovers are only
significant at zip code level. Finally, input sharing is only significant at state level.
Ellison et. al. (2010) made a first attempt to use indexes to measure Marshallian
agglomeration economies. In their study the dependent variable is the rate of co-
agglomeration of industries based on Ellison and Glaeser (1997). They measured the sharing
effect using the proportions of the Input-Output7 matrix for the case of United States and
Great Britain for each manufacturing industry. Labor matching is calculated as the correlation
between proportions of each occupation between industries. Knowledge spillovers are
accounted by technology flows8 among industries, as well as flows of patent citations. The
overall results show positive evidence of the three Marshallian agglomeration economies,
especially the relationships with suppliers and customers. This work is extremely interesting
7
Because the main objective is to measure co-agglomeration. They actually used maximum and minimum
values of these proportions for each pair of individual industries
8
These flows were obtained from the technological matrix built by Frederic Scherer
10
for its explicit modeling of Marshallian agglomeration economies. Because the purpose is to
measure co-agglomeration the combined effect for all industries is left aside.
One of the more accurate investigations to measure Marshallian agglomeration is the work
of Glaeser and Kerr (2009). Their research is also very interesting because instead of using
concentration indexes as the dependent variable they use the number of new firms and jobs
generated so it incorporates entrepreneurship. The geographic scope of analysis is the city.
The sector analysis is based on SIC3 classification. To measure Marshallian agglomeration
economies indexes are built. These indexes are weighted sums comparing the characteristics
in each pair city-industry for all industries at the metropolitan area together. As an added
value, it explores the Chinitz hypothesis by measuring the influence of suppliers inversely
weighted by firm size. The results show support for Marshallian agglomeration economies as
location determinants of entrepreneurial activity in cities. Chinitz hypothesis is also
significant and positive. The industrial analysis is only for the manufacturing sector.
The work of Jofre-Monseny et. al. (2011) provide a reference for the existence of the
Marshallian agglomeration economies in the main cities of Spain. The number of new firms
(three years of existence or less) is used as the dependant variable. In order to measure the
Marshallian agglomeration economies they use indexes as in the papers presented before. The
calculation of the indexes is based on co-aglomeration patterns for each pair of industries as
in Ellison et. al. (2010) but trying to agregate among the most similar industries like in
Glaeser and Kerr (2009). They use a Poisson regression as the econometric specification.
They focus exclusevely on the manufacturing industry at three-digit level (NACE-93 Rev.1).
As a value added, the authors try to quantify the effects between and within cities. The
between cities estimation is based on the 477 cities bigger than 10,000 inhabitants in Spain.
The within estimation is based on the 755 municipalities that form the main 19 cities in Spain.
Their results show that labor matching and sharing with suppliers are the main drivers of
agglomeration, while the knowledge spillovers have a signifcant but geographically limited
scope in comparison.
3. DATA AND VARIABLES

3.1. MEASUREMENT PROBLEMS
The ambiguous definition of entrepreneurship causes an empirical problem in its
measurement. Self-employment is not the best metric because it fails to capture the
entrepreneurial phenomenon. The relationship between economic growth and self-
11
employment is negative (Congregado and Millan, 2008). Despite this, several studies have
been developed using this metric9, as it is the case with Blanchflower and Oswald (1998) or
Van Praag and Van Ophem (1995).
Alternative measures can be used: the number of firm owners or number of businesses
per capita, entry and exit rates from self-employment (Carree and Thurik, 2003), R&D
expenditure (Audrestch and Feldman, 1996) or construction of entrepreneurial indexes (Acs
and Szerb, 2009)
For purposes of this research entrepreneurship will be measured by the total number of
births as well as jobs generated by them (as in Rosenthal and Strange, 2005; or Glaeser and
Kerr, 2009). The INE database for enterprise births DIRCE only gives information at the
province level (NUTS-3) so metropolitan areas can not be correctly defined and not all the
industries are considered. This measurement problem is solved by using firm births and jobs
created from de SABI10 database. This database provided information at the municipal level
(NUTS-4) so the metropolitan areas can be constructed. Table A1 shows a comparative
between DIRCE and SABI information.
A second problem is that Spain does not have an official metropolitan area definition
(Feria-Toribio, 2004). Studies at the urban level in Spain do not consider metropolitan areas
(Arauzo and Teruel, 2005) or define them ad hoc11 (Arauzo and Viladecans, 2009). To solve
the above problem this study uses the definition of metropolitan areas by Boix and Veneri
(2009). They employ an iterative methodology based on four phases from the identification of
core cities. This methodology is attractive because its similar to the one used by the U.S.
Census Bureau. A total of 67 cities are identified for the Spanish case using this technique.
The process guarantees that the city is built on a functional principle12. City and metropolitan
area are used interchangeably for the rest of the paper.
9
The main objective of these studies was measuring moves between employment and self-employment. The
essence is to measure earnings differentials between employment and self-employment.
10
System for Analysis of Iberic Balance Sheets. It is a private database reflecting the status of more than one
million businesses for the case of Spain and Portugal. It is produced by Bureau Van Dijk
11
The Arauzo and Viladecans (2009) work states “The metropolitan area considered for each city covers the area
within a 35-km radius of the centre”
12
The functional approach is based on a commuting principle (greater or equal to 15% from the outskirts to the
core city) to be integrated into the metropolitan area
12
3.2. VARIABLES AND INDEXES CONSTRUCTION
Fifteen metropolitan areas13 and the entire set of 60 two-digit industries (CNAE-93) are
considered. This gives a total of 900 observations for the analysis. The dependent variable is
entrepreneurial activity (Entrepreneurship). The independent variables measure city
characteristics as well as city-industry sources of agglomeration based on the index
constructed by Glaeser and Kerr (2009). The following list details information about each of
the variables used; where the sub-indexes i , c , k and o represents industry, city, other
industries and occupations respectively:
Entrepreneurship ci : It is measured in two ways: (1) firm births and (2) jobs created. The
variable represent the average firm births (or jobs created) during the period 2000-2008 for
each city-industry pair. These births have at least one employee and less than 20. Firm births
with at least one employee were used to get rid of self-employment. Restricting the sample
only to firms with fewer than 20 employees can avoid (partially) re-locations or re-openings
of existing enterprises and the over-representation of this type of firms in the SABI database.
Pop c : Total population in the city. The variable will measure the degree of urbanization
economies. It is measured according to the INE 2001 Population Census. This was the last
official Census for the country.
Emp ci : Total employment for each pair city-industry. This variable measures the localization
economies. It is taken from de INE 2001 Census.

Xc
: This variable represents controls for the city level.
Proportion of firms births in the core municipality. (SABI)
Proportion of jobs created in the core municipality. (SABI)
Age distribution of the population. (INE 2001 Census)
Ratio of men respect to women. (INE 2001 Census)
Education level distribution of the population. (INE 2001 Census)
Human capital. It is measured with an index elaborated by IVIE (Valencian
Institute of Economic Research)
Culturec : Measures the entrepreneurial culture. This index presents the proportion of
employment that the specific industry represents in the city (INE 2001 Census). It is then
13
Madrid, Barcelona, Valencia, Sevilla, Bilbao, Oviedo, Málaga, Zaragoza, La Palma de Gran Canaria, Alicante,
Murcia, Vigo, Tenerife, Granada and Palma de Mallorca. Tabla A0 in the annex show information about the
population and employment of each city. See Boix and Veneri (2009) for more information.
13
multiplied by the proportion of firm births for the specific industry divided by the proportion
of births for all industries at the national level (Source DIRCE). Higher values of the index
represent a better entrepreneurial culture.
Empic % Entry i
Culturec = ∑
i =1,..; I Emp c % Entry
Diversityc : Measures the concentration at the city level. It is calculated with the Herfindhal-
Hirshman index. Values near one represent a very concentrated city while values near zero
represent a totally diversified city
Empic
Diversity c = ∑ ( Emp
i =1,..; I
)2
c
Chinitz ci : Accounts for the relevance of the small suppliers for each city-industry pair. It
measures the proportion of Inputs (INE 2005 symmetric Input-Output table) that the industry
receives from other industries. Then the index is multiplied by the inverse of the average size
of the supplier industry in the city. The index is near one for a high number of suppliers and
near zero if the industry is supplied by a small number of firms.
Firms kc
Chinitz ci = ∑
k =1,..; K Empc
Input i ← k
Input ci : Measures the relative strength of sharing with suppliers. The calculation is based on
the proportion of Inputs received from each industry at the national level compared with the
presence of the industry in the city. Values near zero represent a good presence of suppliers in
the specific city-industry while values near one are indicative of a poor presence. The index is
multiplied by -1 in order to obtain positive coefficients in the regression.
Empck
Input ci = − ∑
k =1,..; K
Input i ← k −
Empc
Output ci : Measures the relative strength of sharing with customers. The calculation is based
on the proportion of Output (INE 2005 symmetric Input-Output table) that the industry sends
to other industries at the national level multiplied by the size of the customer industry in the
city. In order to avoid scale effects the index is divided by the relative size of the customer
market in the whole economy and its relative presence in the city. Greater values of the index
represent good opportunities of sharing with customer.
−1
 Empck   Empck 
Output ci =  ∑ Output i → k  ⋅  ∑ Output • → k 
 k =1,..; K Emp k  k =1,..; K Empc 
14
Laborci : Accounts for the importance of labor matching. It is calculated comparing the
proportion of labor by occupation (INE 2001 Census using CNO-94 classification) in the
industry relative to the presence of the occupation across industries in the specific city. The
index is multiplied by -1 to obtain positive coefficients in the regression consistent with good
labor matching.
 E ck 
Laborci = − ∑ Lio − ∑ 
k =1,..., K  E c
Lko 
o =1,..;O 
Techci : Measures the potential learning spillovers in the city-industry pair. This index is
calculated differently than that proposed by Glaeser and Kerr (2009). The main difference is
that this index uses proportion of innovative companies14 (Source: INE 2003-2005 “R&D
Statistics”) per industry instead of patent citations. A second difference is that the index
developed here measures contact with innovator suppliers and customers (Glaeser and Kerr
only measure supplier’s relations). The index try to capture the proportion of Inputs (Outputs)
that the industry receive from (send to) other industries multiplied by the proportion of
innovative companies. Values near zero indicate good contact with innovators. The index is
multiplied by -1 to obtain positive coefficients in the regression related to the presence of
knowledge spillovers.
 1 Emp ck   1 Emp ck 
2
(
Techci = −  × ∑ Input i ← k ⋅ Innov k − )
Emp
 +  × ∑ Output
  2 k =1,..; K i
→ k
⋅ Innov k − (
Emp

 )
 k =1,..; K c   c 
Table 2 shows the main descriptive stats. Columns 1 to 4 display means, standard deviations,
maximums and minimums respectively. The maximum and minimum column also inform
about the sector15 and city16 that represent that value.
Mean number of new enterprises during the period 2000-2008 for each city-industry
pair has been close to 200 companies. The large standard deviation tells us about the
heterogeneity across industries and cities. The industry with more entry (9542 companies) is
Other Business Services (s74) located in the metropolitan area of Madrid. The average
number of employees in new enterprises has been 824. The maximum value with 38,416
14
Acs and Audretsch (2003) compare equations that measure innovation through R&D inputs, intermediate
process such as number of patents, and final production of innovations. They found that the conclusions differ
depending on the type of indicator used. They also add that direct measures of innovation are more desirable.
15
The sector is represented by the code snn where nn is the two digit code according to CNAE-93 classification.
16
The following codes are used for the 15 cities in the analysis: Madrid (MAD), Barcelona (BCN), Valencia
(VAL), Sevilla (SEV), Bilbao (BIL), Oviedo (OVI), Malaga (MAL), Zaragoza (ZAR), Palma de Gran Canaria
(PGC), Alicante (ALI), Murcia (MUR), Vigo (VIG), Santa Cruz de Tenerife (TEN), Granada (GRA) and Palma
de Mallorca (PML)
15
employees is present in the building industry (s45) in Madrid. The average number of
employees per firm is 4. The core municipality represents in average 45% of the firms and
43% of employment generated in the whole city.
Demography appears very stable across cities in comparison with the differences in
industrial structure. The measure of entrepreneurial culture shows a maximum value for
Palma de Mallorca and minimum for Zaragoza. The concentration index shows that Palma de
Mallorca is the most specialized city which suggests that the entrepreneurial culture is related
with the great participation of the construction (s45) and tourism (s55) industries in the city.
Barcelona is the most diversified city. Palma de Mallorca has the maximum value for the
Chinitz index and this occurs precisely in the building industry (s45) which supports
somehow the notion of high concentration and firm births in the building and tourism industry
in the city. The Input index reflects the best opportunities to share relations with suppliers for
the education industry in Madrid and the worst for the oil industry in Murcia. The Output
index is characterized by great heterogeneity showing the best customer relationships for the
leather industry in Alicante. The Labor index shows that the best match in terms of
occupation is found in agriculture in Barcelona and the worst is in the maritime transport
industry in Madrid. The Tech index suggests the best chance for knowledge spillovers are for
the building industry in Granada, while the worst are for the recycling industry in Malaga.
Table 2
Descriptive Stats
Mean S.D. Maximun (Sector) Minimum (Sector)
1 2 3 (City) 4 (City)
(A) Actual Data
Number of Entrepreneurial Firms 199.34 724.40 9542.00 (s74) (MAD) 0.0000 Many
Number of Employeed in Entrepreneurial Firms 824.44 2939.90 38416.00 (s45) (MAD) 0.0000 Many
Number of Employees per firm 4.0573 2.6613 20.0000 (S95) (ZAR) 0.0000 Many
Number of Entrep. Firms in the Core Municipality 90.10 346.71 6200.00 (s74) (MAD) 0.0000 Many
% Firms Entrep. Firms in the Core Municipality 0.4520 0.2978 1.0000 Many 0.0000 Many
Number of Entrep. Employees in the Core Municipality 355 1275 21037 (s74) (MAD) 0.0000 Many
% Employees in the Core Municipality 0.4306 0.3112 1.0000 Many 0.0000 Many
Population 1392644 1533836 5806548 (MAD) 528634 (PML)
% Population younger than 20 years 0.2091 0.0271 0.2438 (MUR) 0.1562 (OVD)
% Population 20-40 years 0.3365 0.0150 0.3607 (PGC) 0.3040 (OVD)
% Population older than 60 years 0.1995 0.0268 0.2549 (OVD) 0.1629 (PGC)
% Men / Women 0.9488 0.0173 0.9794 (OVD) 0.9101 (PGC)
Human Capital Index 2.7344 0.1570 3.0588 (BIL) 2.5388 (MLG)
% Analfabet Population 0.0233 0.0089 0.0371 (TEN) 0.0085 (OVD)
% Population without studies 0.1143 0.0221 0.1453 (GRA) 0.0646 (BIL)
% Population with first level education 0.2137 0.0212 0.2700 (VGO) 0.1797 (MAD)
% Population second level eduaction 0.5024 0.0273 0.5719 (PML) 0.4534 (GRA)
% Population third level eduaction 0.1462 0.0246 0.1997 (MAD) 0.1084 (VGO)
Diversity Index (HHI) 0.0597 0.0091 0.0726 (PML) 0.0462 (BCN)
Entrepreneurial Culture Index 0.9349 0.0373 1.0332 (PML) 0.8882 (ZAR)
Employees by city-industry 9674 24599 260812 (s45) (MAD) 0.0000 Many
Total Employees by city 580460 696000 2594778 (MAD) 179281 (GRA)
Total resident employees 546807 672414 2517895 (MAD) 162577 (GRA)
Chinitz index 0.0048 0.0024 0.0183 (S45) (PML) 0.0000 Many
Input index -1.4086 0.1609 -0.9457 (s80) (MAD) -1.8125 (s23) (MUR)
Output index 0.9710 1.1388 26.4809 (s19) (ALI) 0.0000 Many
Labor index -1.1987 0.1844 -0.8647 (s01) (BCN) -1.7079 (s61) (MAD)
Tech index -0.9704 0.0173 -0.8810 (s45) (GRA) -1.0019 (s37) (MAL)
Source: INE, SABI and IVIE
16
3.3 THE MODEL
The model specification is based on the proposal by Glaeser and Kerr (2009) for the U.S.
case. The econometric estimation method used is ordinary least squares. White’s robust error
matrix is used to correct heteroscedasticity problems. The variables will be standardized to
have zero mean and unit standard deviation. This will help to interpret the coefficients in
terms of their variability. The variables equal to zero are transformed in ones to avoid
problems with logarithms.
Equation (1) represents the main estimation:
η i + β P ln( Pop c ) + β E ln( Emp ci ) + γ c X c 
 
ln( Entrepreneurship ci ) =  + β H Diversity c + β C Culturec + β Ch Chinitz ci  (1)
 + β Input + β Output + β Labor + β Tech + ε 
 I ci O ci L ci T ci ci 
The coefficient ηi represents the fixed effects at the industry level. The term ε ci represents
the residuals of the regression. Endogeneity is a concern. Problems of measurement error (the
indexes do not capture perfectly the essence of the agglomeration economies) or omitted
variables (Other relevant factors causing firm births) might be present.
A second estimation is proposed by dropping the city specific controls X c and
substituting them by city fixed effects φc . This estimation presented in equation (2) is also
called conditional. It receives this name because this estimation is conditional on the city
fixed effects as good substitutes for city demographics.
η i + φ c + β E ln( Emp ci ) + β Ch Chinitz ci 

ln( Entreprene urship ci ) =   (2)
 + β I Input ci + β O Output ci + β L Laborci + β T Techci + ε ci 
4. RESULTS
4.1. GENERAL REGRESSION RESULTS
Table 3 shows the results for the unconditional estimates for the dependent variable number
of employees in new firms. Column 1 shows the base estimation that captures only the
economies of urbanization and localization. Both are highly significant but the effect of
localization economies is stronger.
Column 2 adds city demography. It can be seen that the proportion of employees in the
core municipality has a small but significant impact. Human capital and educational level
17
have a negative sign showing an inverse relationship between human capital and
entrepreneurship in Spain (this is consistent with the low-skill entrepreneurship in the country
even removing self-employment). The culture and diversity measures have no effect on
entrepreneurial employment. Chinitz index is significant and positive.
Column 3 adds Marshallian agglomeration economies without the demographic
controls. The Output index is highly significant, showing that entrepreneurial activity is
primarily affected by a good representation of customers. Sharing inputs and innovation
represented by the Tech and Input indexes are significant and positive but only at 10%
significance level. The Labor index is not significant which suggests that the presence of
specific occupations are not important for entrepreneurship in a broad sense. Spanish
entrepreneurship is based mainly in industries where skilled labor is not necessarily relevant.
Another possible reason is that the localization economies, measured as total employment in
the pair city-industry are capturing the labor pooling effect.
Table 3: Non-Conditional Estimation Number of Employees
Dependent Variable:Log Number of Employees in Entrepreneurial Firms
Base Estimation City Demographics Marshallian Agglomeration Full Estimation
1 2 3 4
Coefficient (t-stat) Coefficient (t-stat) Coefficient (t-stat) Coefficient (t-stat)
Log City Population 0.157 *** 8.282 0.281 *** 3.758 0.176 *** 9.066 0.312 *** 4.234
Log Employment city-industry 0.426 *** 8.829 0.417 *** 8.454 0.322 *** 6.191 0.332 *** 6.425
% Employees in the Core Municipality 0.044 ** 2.379 0.048 ** 2.597
% Population younger than 20 years 0.542 * 1.717 0.560 * 1.809
% Population 20-40 years 0.391 * 1.955 0.415 ** 2.116
% Population older than 60 years 0.851 ** 1.973 0.887 ** 2.097
% Men / Women -0.102 * -1.742 -0.104 * -1.796
Human Capital Index -0.160 *** -2.855 -0.161 *** -2.914
% Population without studies -0.612 ** -2.150 -0.613 ** -2.191
% Population with first level education -0.224 * -1.801 -0.212 * -1.736
% Population second level eduaction -0.453 ** -2.200 -0.448 ** -2.219
% Population third level eduaction -0.338 -1.610 -0.330 -1.603
Diversity Index (HHI) 0.065 0.902 0.075 1.043
Entrepreneurial Culture Index 0.031 1.141 0.036 1.312
Chinitz index 0.072 ** 2.178 0.046 1.449
Input index 0.065 * 1.752 0.043 0.962
Output index 0.048 *** 4.526 0.053 *** 5.023
Labor index 0.030 0.607 -0.019 -0.348
Tech index 0.040 * 1.850 0.042 ** 2.057
N Obs 900 900 900 900
2
R 0.921 0.926 0.924 0.928
2
Adjusted R 0.916 0.919 0.918 0.922
Industrial Fixed Effects (CNAE-93) X X X X
Note: ***, ** and * indicate significance levels at p<0.01, p<0.05 p<0.1 respectively
18
Column 4 shows the full estimation. The results do not change dramatically. It is
worth noting that the Chinitz and the Input indexes loss significance. This can be caused by
the high degree of correlation between the Tech and the Chinitz indexes. The Output index
along with urbanization and localization economies seems to be the more meaningful
agglomeration effects. It is also important to note the high R-squared of 0.92. Industry fixed
effects alone give a value of 0.81 which shows the importance of the existing industrial
structure. On the other hand, urbanization and localization economies alone explain 0.60 of
the total variance of new jobs.
Table 4: Non-Conditional Estimation Number of Firms

Dependent Variable:Log Number of Entrepreneurial Firms
Base Estimation City Demographics Marshallian Agglomeration Full Estimation
1 2 3 4
Log City Population 0.183 *** 9.953 0.215 *** 4.249 0.191 *** 11.372 0.233 *** 4.809
% Firms in the Core Municipality -0.032 ** -2.236 -0.029 ** -2.078
% Population younger than 20 years 0.240 1.180 0.249 1.263
% Population 20-40 years 0.201 1.468 0.213 1.600
% Population older than 60 years 0.416 1.449 0.434 1.555
% Men / Women -0.042 -1.035 -0.041 -1.060
Human Capital Index -0.170 *** -3.626 -0.169 *** -3.680
% Population without studies -0.294 -1.546 -0.289 -1.573
% Population with first level education -0.127 -1.482 -0.117 -1.372
% Population second level eduaction -0.236 * -1.701 -0.229 * -1.692
% Population third level eduaction -0.112 -0.821 -0.104 -0.785
Diversity Index (HHI) -0.058 -1.211 -0.055 -1.168
Entrepreneurial Culture Index 0.054 ** 2.490 0.058 *** 2.740
Chinitz index 0.089 *** 3.072 0.073 *** 2.594
Input index 0.075 ** 2.267 0.029 0.823
Output index 0.029 1.325 0.034 * 1.877
Labor index 0.035 0.826 -0.032 -0.739
Tech index 0.038 ** 2.154 0.035 ** 2.358
N Obs 900 900 900 900

2
R 0.947 0.953 0.949 0.954
2
Adjusted R 0.943 0.949 0.945 0.950
Table 4 shows the same regressions but using as dependent variable the number of
new firms. There are three major differences with the previous estimation. First, the
demographic variables of age and education lose their significance but the human capital is
again negative and highly significant. Second, the existence of customers or the Output metric
loses much of its importance but remains significant at 10% in the full estimation. Third, the
Chinitz and the Entrepreneurial Culture metrics that were not significant in the case of new
19
jobs are now positive and highly significant. Knowledge spillovers index also seems to be
more important for firm births than for jobs creation. All this suggests that the determinants
for new employment are more related with characteristics of the population in the city while
firm births are more sensitive to the industrial structures. That conclusion can be confirmed
by noting an increase in the R-squared in this regression to 0.95, largely because industry
fixed effects alone explain a 0.83 of variability.
Table 5 presents the conditional estimation for the number of jobs generated. In
general this estimation makes more emphasis on the variables measuring agglomeration
economies. Unfortunately, the metric for urbanization economies became part of the fixed
effects. The conclusions do not change from the ones of Table 3. Localization economies are
highly significant. The Input index is significant until the Chinitz index is introduced. A
possible multicollinearity effect may exist. The Output and the Tech indexes maintain their
significances and magnitudes. The R-square is very similar suggesting that the city-level fixed
effects are a good substitute for the demographic characteristics presented above.
Table 5: Conditional Estimation Number of Employees

Base Estimation Input, Output and Labor Plus Chinitz Full Estimation
1 2 3 4
Chinitz index 0.049 1.564 0.051 1.589
Input index 0.070 * 1.685 0.055 1.297 0.047 1.101
Output index 0.054 *** 5.137 0.054 *** 5.179 0.053 *** 5.051
Labor index -0.005 -0.091 -0.005 -0.100 -0.011 -0.197
Tech index 0.036 * 1.787
N Obs 900 900 900 900

2
R 0.925 0.928 0.928 0.928
2
Adjusted R 0.918 0.921 0.921 0.921
City Fixed Effects X X X X
Table 6 shows the conditional estimation for the number of firm births. The main
conclusions from Table 4 are maintained, except for the Input index, that becomes
significant. The possible reason for this to happen is the high correlation with some
demographic characteristics. Despite some differences between the estimates for new jobs and
firm births the influence of Marshallian agglomeration economies in entrepreneurial activity
is robust, especially for the Output and the Tech indexes.
20
Table 6: Conditional Estimation Number of Firms
Base Estimation Input, Output and Labor Plus Chinitz Full Estimation
1 2 3 4
Chinitz index 0.069 ** 2.440 0.071 ** 2.400
Input index 0.091 ** 2.524 0.070 * 1.917 0.062 * 1.697
Output index 0.035 * 1.905 0.035 * 1.892 0.034 * 1.785
Labor index -0.027 -0.627 -0.028 -0.641 -0.033 -0.759
Tech index 0.036 ** 2.399
N Obs 900 900 900 900

2
R 0.953 0.954 0.954 0.955
2
Adjusted R 0.948 0.950 0.950 0.950
4.2. REGRESSIONS BY INDUSTRY

This section analyzes the effects of Marshallian agglomeration by industry. This is interesting
because part of the value added of this work is analyzing the service sector. Unfortunately,
building industry can not be analyzed because it only has 15 observations in the sample.
Table 7: Conditional Sectorial Estimation Number of Employees

Agropecuary Manufacturing Energy Services
1 2 3 4
Log Employment city-industry 0.596 *** 2.777 0.318 *** 4.970 0.180 1.196 0.052 0.439
Chinitz index -0.572 -1.030 0.326 ** 2.209 -0.100 -0.249 0.045 0.827
Input index 0.852 1.391 -0.200 * -1.777 0.151 0.649 0.110 * 1.895
Output index 0.143 0.644 0.064 *** 5.526 0.090 ** 2.293 0.191 * 1.787
Labor index -0.084 -0.087 0.084 0.648 0.332 1.048 -0.043 -0.400
Tech index -0.555 -1.509 0.001 0.024 0.081 1.009 0.036 1.441
N Obs 45 330 120 390

2
R 0.898 0.877 0.765 0.951
2
Adjusted R 0.796 0.859 0.696 0.945
Table 7 shows the results by industry using new employment as the dependent
variable. Column 1 shows the agricultural sector. As expected, only localization economies
are significant. The manufacturing industry is presented in column 2. Economies of
localization and the Output index are positive and significant. The Chinitz index is also
positive and of big magnitude indicating that small suppliers in the manufacturing industry
are important to attract new jobs. It is worth noting that the Input index is significant but
negative. This suggests that manufacturing industries tend to be concentrated near small
suppliers and far from the bigger ones. Big manufacturing industry suppliers concentrate in a
21
different region or metropolitan area than new ventures which may explain the negative
coefficient and opens the door to analyze the network effects that exist between metropolitan
areas (unfortunately, it is out of the scope of this research). For the energy industry only the
Output index is significant, indicating new ventures in the industry located close to their
customers. The service industry is only affected by the Input and the Output indexes. Both
effects are weakly significant. This suggests that entrepreneurial activity in this industry is
affected by the presence of suppliers and customers. The localization economies are not
significant for the first time. This suggests that services follow other industries to
agglomerate.
Table 8: Conditional Sectorial Estimation Number of Firms

Agropecuary Manufacturing Energy Services
1 2 3 4
Log Employment city-industry 0.622 *** 3.774 0.301 *** 5.425 0.150 *** 2.571 0.096 1.033
Chinitz index -0.674 -1.366 0.253 ** 1.994 0.147 0.565 0.087 ** 1.923
Input index 0.832 * 1.776 -0.178 ** -1.973 0.155 1.051 0.122 *** 2.650
Output index 0.045 0.223 0.055 *** 3.133 0.007 0.307 0.176 ** 2.107
Labor index -0.796 -1.116 0.037 0.289 0.215 0.997 -0.046 -0.575
Tech index -0.234 -0.859 0.019 0.474 0.041 0.890 0.026 * 1.837
N Obs 45 330 120 390

2
R 0.939 0.919 0.838 0.975
2
Adjusted R 0.879 0.907 0.790 0.972
Table 8 shows the conditional estimates for the dependent variable number of new
firms. The influence of Marshallian agglomeration factors becomes more relevant for this
estimation. The agricultural industry is again influenced by economies of localization. The big
difference is that in this case the Input index is also significant. The main supplier of the
agricultural industry is the own industry which reinforces the relevance of localization
economies. The results for the manufacturing industry do not change from the previous
specification. It should be noted that the Input index increases its significance and maintain its
negative sign. In the energy industry, however, the results change. The localization economies
are now the only significant coefficient. This reflects that new firms in this industry tend to
concentrate around old ones, but the creation of jobs is more related to customers presence.
The estimation for the service industry presents some differences with the previous one. The
Input and the Output indexes increase their level of significance. The Chinitz and the Tech
indexes become significant. This suggests that the influence of small suppliers and innovative
companies affect the location of new firms in the service industry but not the employment
size.
22
4.3. NATURAL COST ADVANTAGE ESTIMATION
To deal with endogeneity Glaeser and Kerr (2009) propose an alternate estimation. This
estimation re-calculates the agglomeration indexes based on the city-industry employment
shares derived by a natural cost advantage estimation.
This new estimation is based on the methodology proposed by Ellison and Glaeser
(1999). The new indexes account only for the agglomeration caused by the natural conditions
so it is supposed to solve endogeneity. The main results of the natural cost advantages
estimation can be seen in Table A1 in the Annex. The results are not reliable because
information about natural cost advantages in Spain is not complete. It is difficult to obtain the
prices and specific consumptions of raw materials as proposed by the methodology at the
metropolitan level. This causes that the estimates derived from this method homogenize too
much the characteristics of the metropolitan areas (The means of the re-calculated indexes are
very similar to the original ones but the standard deviations decrease a lot as shown in Table
A2) thus the conclusions can not be used to make inference about endogeneity concerns.
This can situation can be caused by the different dependence to the natural cost
advantages between US and Spanish cities. In the annex the Table A0 shows how the
methodology homogenizes the proportions of employment between cities.
4.4. ALTERNATIVE SPECIFICATIONS

A second exercise using principal component analysis is proposed. The analysis is carried out
to test the inter-dependencies between the calculated indexes and localization economies. The
idea behind this exercise is that the metric of localization economies is too broad so it can be
taking predictive capacity from the Marshallian agglomeration indexes. Using principal
component analysis can help to understand better these inter-dependencies creating lineal
combinations of the original variables. This new factors extracted from the analysis are
perfectly orthogonal by construction so they can be used instead of the original indexes to
create a more efficient regression.
Table 9 shows the main results of the principal component analysis. The Kaiser-
Meyer-Olkin (KMO) measure gives us an idea of the capacity of the original variables to be
summarized by a lineal combination. A KMO measure of 0.590 is low for standard principal
component analysis but high enough to show some type of hidden relations between the
variables. The component matrix shows the correlation between the original variables and the
orthogonal factor or components. Localization economies, Input, Tech and Chinitz indexes
23
are connected. This connection suggests that the better predictor (localization economies in
this case) can take over explanatory power from the other indexes. Finally the rotated solution
allows us to construct new factors that mimic almost perfectly the original regressors (a
correlation higher than 0.90 in all cases) but incorporates the variability of the other indexes
that is more common to them.
Table 9: Principal Component Analysis

KMO Measure of Sampling
Adequacy .590
Chi-square 705.298
df 15
Bartlett Sphericity Test Sig. .000
Componet
1 2 3 4 5 6
Emp_ci .636 -.271 -.007 .256 -.674 .039
Input .510 -.535 .015 .470 .465 .130
Output .135 .364 .902 .190 .012 .011
Labor .009 .717 -.387 .580 .017 -.015
Tech .747 .376 -.104 -.376 .083 .375
Component Matrix Chinitz .860 .126 -.051 -.177 .149 -.433
Component
1 2 3 4 5 6
Emp_ci .096 .134 .975 -.025 .008 .147
Input .021 .981 .131 -.055 -.011 .131
Output .032 -.011 .008 .011 .999 .033
Labor .044 -.052 -.023 .997 .011 .010
Rotated Component Matrix - Tech .944 .019 .102 .052 .037 .307
Equimax Method Chinitz .327 .155 .168 .011 .040 .916
Table 10 shows the regressions run with the factors extracted from the rotated
solution. The results show a generalized increase in significance levels. A second effect is that
the Labor index is now significant. This result suggests that the localization economies
(measured as total employment in each city-industry pair) are also capturing the effect of
labor pooling. The problem of using factors instead of original variables is that we do not
know the real economic meaning of the lineal combinations. The new regressions also show
some advantages solving possible multicollinearity problems and increasing significance
levels. The analysis also suggests that broad measures of agglomeration are capturing
multiple agglomeration phenomena at the same time.
In order to evaluate the robustness of the results a third exercise is carried out. This
regression tries to use an alternative measure of entrepreneurial activity in order to check the
robustness of the coefficients calculated previously. The regression is identical to the ones
presented from Table 5 to Table 8 but instead of using the SABI database the DIRCE was
used. The DIRCE database is a census of enterprises (the SABI is a sample). The DIRCE has
two main disadvantages and thus was not used originally: (1) It only gives data at the
province level (NUTS-3) so the metropolitan areas can not be accurately tracked; and (2) Not
all the industries are considered. To construct the metropolitan data, the province firm births
(and new employment) were multiplied by the share of the population that each city
24
represents in the province or provinces. This method tends to underestimate the data from
industries that have greater incentives to agglomerate and to overestimate the ones with
incentives different than agglomeration. Also 90 observations are lost due to the missing
industries.
Table 10: Conditional Estimation with Rotated Principal Components

Total Agropecuary Manufacturing Energy Services
1 2 3 4 5
Coefficient (t-stat) Coefficient (t-stat) Coefficient (t-stat) Coefficient (t-stat) Coefficient (t-stat)
(A) Dependent Variable:Log Number of Employees in Entrepreneurial Firms
Emp_ci Factor 0.045 *** 3.509 0.286 1.236 0.060 1.280 1.895 ** 2.544 0.048 ** 2.296
Chinitz Factor 0.066 ** 2.224 -0.494 -1.074 0.229 * 1.774 0.295 0.813 0.064 1.247
Input Factor 0.090 *** 2.087 0.978 * 1.911 -0.056 -0.530 0.363 1.540 0.107 * 1.892
Output Factor 0.088 *** 7.005 0.099 0.365 0.105 *** 5.771 0.124 *** 3.276 0.240 *** 2.685
Labor Factor 0.125 ** 2.431 1.029 0.954 0.304 ** 2.342 0.103 0.357 -0.014 -0.133
Tech Factor 0.063 ** 2.544 -0.711 * -1.960 0.121 1.641 0.271 1.611 0.055 * 1.761
N Obs 900 45 330 120 390

2
R 0.922 0.875 0.864 0.786 0.951
2
Adjusted R 0.915 0.750 0.845 0.724 0.945
Industrial Fixed Effects X X X X X
City Fixed Effects X X X X X
(B) Dependent Variable:Log Number of Entrepreneurial Firms
Emp_ci Factor 0.070 *** 5.169 0.454 ** 2.228 0.122 *** 2.660 2.013 *** 6.367 0.061 *** 3.148
Chinitz Factor 0.081 *** 3.111 -0.459 -1.123 0.182 * 1.692 0.502 *** 2.846 0.099 ** 2.374
Input Factor 0.102 *** 2.729 1.014 *** 2.566 -0.064 -0.765 0.422 *** 3.790 0.123 *** 2.767
Output Factor 0.067 *** 5.474 0.006 0.024 0.090 *** 6.675 0.046 *** 2.879 0.253 *** 3.464
Labor Factor 0.101 ** 2.297 0.285 0.374 0.254 ** 2.023 -0.029 -0.214 -0.001 -0.010
Tech Factor 0.067 *** 3.525 -0.490 * -1.748 0.115 ** 2.022 0.314 *** 3.849 0.059 *** 2.962
N Obs 900 45 330 120 390

2
R 0.950 0.921 0.908 0.888 0.976
2
Adjusted R 0.945 0.842 0.895 0.855 0.972
Industrial Fixed Effects X X X X X
City Fixed Effects X X X X X
Table 11 presents the results of the estimations using DIRCE. In Column 1 panel A the
results of the complete estimation for the number of employees are presented. The main
difference is that the Tech index loses its significance and the Labor Pooling index became
significant. The labor pooling was not significant in the regressions using SABI database, a
possible explanation for this difference can be found in the firms bigger than 20 employees
that were out of the SABI sample but can not be removed from DIRCE. In the column 1 panel
B the results of the estimation for the firm births are presented. Tech and Input indexes lose
their significance and labor pooling became significant, again. DIRCE estimates show that the
localization economies, the relations with customers and labor pooling are the main drivers of
the new employment while for the firm births the presence of small suppliers is also
important.
Columns 2-4 present the regressions by industry. It is worth noting that the
conclusions remain basically the same except for the service industry. For SABI’s case the
localization economies are not relevant for the new employment and firm births in the service
25
industry while in DIRCE’s case this factor is significant and of great magnitude. Also the
presence of small suppliers and innovative companies (Chinitz and Tech indexes) were
important for the SABI estimation of firm births but were not relevant for the DIRCE case.
The only consistent effect between both estimations is for the Input index but the magnitudes,
specially for the firm births, are quite different.
Table 11: Conditional Estimation with DIRCE database

Total Manufacturing Energy Services
1 2 3 4
(A) Dependent Variable:Log Number of Employees in Entrepreneurial Firms (DIRCE)
Chinitz index 0.034 0.972 0.367 ** 2.155 0.161 0.376 -0.034 -0.758
Input index 0.033 0.650 -0.286 ** -2.328 -0.172 -0.595 0.121 ** 2.238
Output index 0.050 ** 2.152 0.059 ** 2.370 0.097 *** 2.757 -0.131 -0.961
Labor index 0.202 *** 3.349 0.105 0.712 -0.341 -0.819 0.204 1.488
Tech index 0.004 0.177 0.051 1.500 -0.005 -0.042 -0.045 -1.059
N Obs 810 330 120 345

2
R 0.919 0.890 0.813 0.920
2
Adjusted R 0.911 0.874 0.758 0.909
(B) Dependent Variable:Log Number of Entrepreneurial Firms (DIRCE)
Log Employment city-industry 0.410 *** 11.860 0.469 *** 9.193 0.167 ** 2.480 0.505 *** 5.416
Chinitz index 0.074 ** 2.570 0.298 ** 2.303 0.117 0.405 0.020 0.650
Input index 0.014 0.400 -0.165 * -1.765 0.029 0.240 0.058 * 1.762
Output index 0.042 *** 6.359 0.049 *** 6.089 0.052 *** 2.931 -0.048 -0.596
Labor index 0.087 ** 2.412 -0.016 -0.155 -0.042 -0.196 0.092 1.646
Tech index 0.016 1.121 0.043 1.441 -0.036 -0.402 -0.014 -0.543
N Obs 810 330 120 345

2
R 0.965 0.931 0.864 0.982
2
Adjusted R 0.962 0.921 0.824 0.980
Table A3 in the Annex is an analogous of Table 10 but using DIRCE’s database to

construct the dependent variables. Again the coefficient for the localization economies
remains significant but its effect is of less magnitude. The labor pooling variable appears
specially important for this regressions reinforcing the notion of the relevance of the bigger
(more specialized) companies in DIRCE data. Once again, the service industry conclusions
are radically different from SABI and DIRCE. In the first case the Output index is the most
relevant force. There are, however, other significant factors. While in the second the Labor
index is the only consistent effect for number of new employees although Output index is also
of high magnitude for the firm bith regression.
4.5 REGRESSIONS BY CITY

An individual estimation by city was also run. The results are shown in Table A4 in the
Annex. The only consistent significant effect across cities is the localization effect. The
magnitude of the localization economies is increasing with respect to city size. This indirectly
26
suggests a sharing effect making more attractive the presence of the own industry as the
market grows and specializes.
Another four regressions were run to complement the analysis by city. In table A5 a
regression using the SURE method was run. SURE method is useful to account for the
unobserved heterogeneity and obtain more efficient estimators. In Table A6 a regression by
city using the Rotated Principal Components as independent variables was used to check for
possible interdependencies between the indexes at the city level. Table A7 shows a regression
by city using DIRCE database to check robustness. Finally Table A8 presents the results of a
regression using DIRCE database and the Rotated Principal Components as independent
variable.
The general summary of this five regressions by city is presented in Table 12. The
results show the significance level for each variable in each city (Column 1 represent new
employment as dependent variable and Column 2 present firm births as dependent variable).
A strong evidence of the influence of the localization economies, the Chinitz and Tech
indexes is present across cities. The Input index becomes more relevant in the regressions
using Principal Components and DIRCE data. Altough the data in table A4 suggests
increasing returns to scale respect with city size, the evidence is not so clear across
specifications.
Table 12: City estimations Summary

MAD BAR VAL SEV BIL OVI MAL ZAR PGC ALI MUR VIG TEN GRA PML
1 2 1 2 1 2 1 2 1 2 1 2 1 21 2 1 2 1 2 1 21 2 1 2 1 2 1 2
A4 ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+
***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+
***+ ***+ ***+ ***+ ***+ ***+***+ ***+
A5 ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+
***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+
***+ ***+ ***+ ***+ ***+ ***+***+ ***+
A6 ***+ ***+ ***+ ***+ ***+ ***+ **+ **+ ***+ ***+ **+ **+ ***+ ***+
***+ **+ *+ *+ **+ **+ ***+ ***+
***+ ***+ ***+ **+ *+ ***+ ***+
A7 ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+
***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+
***+ ***+ ***+ ***+ ***+ ***+***+ ***+
Emp ci A8 ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+
***+ ***+ ***+ ***+ ***+ ***+ **+ **+
***+ ***+ ***+ ***+ ***+ **+ ***+ ***+
A4 **+ ***+ **+ ***+ ***+ **+ *+ *+ **+ *+ ***+ ***+ **+ **+ *+
A5 ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ **+ ***+ **+ ***+ **+ ***+ ***+ ***+ **+ ***+ **+ **+ ***+ ***+ ***+ **+ ***+ **+
A6 **+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ *+ ***+ **+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ **+ ***+ ***+ ***+ ***+ **+ ***+ ***+ ***+
A7 **+ *+ **+ ***+ ***+ **+ **+ *+ *+ ***+ **+
Chnitz A8 ***+ ***+ ***+ ***+ *+ *+ ***+ **+ ***+ ***+ ***+ **+ ***+ **+ ***+ **+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+
A4 *-
A5 *- **-
A6 *+ *+ *+ *+ *+ ***+ ***+ **+ **+ *+ **+ **+ ***+ ***+ ***+ ***+ **+ ***+ **+ ***+
A7 ***+ ***+ ***+ ***+ ***+ ***+ **+ **+ ***+ **+ ***+ **+ ***+ **+ ***+ **+ ***+ *+ ***+ ***+ ***+ **+ ***+ ***+ ***+ **+ **+ *+ ***+ **+
Input A8 ***+ ***+ **+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+
A4 **+ *+ **+ **+
A5 *+ **+ ***+ ***+ *+ **+ ***+ ***+ ***+ *- *+
A6 ***+ ***+ ***+ **+ ***+ **+ **+ *+ *+ **+ **+
A7 **+ **- **- *-
Output A8 ***+ ***+ ***+ ***+ *- **- **+ **+ **+ *+ *+ ***+ ***+
A4
A5 **+
A6
A7 **- *- **-
Labor A8 **+ *+
A4 ***+ **+ *+ **+ **+ **+ *+ *+ ***+ *+ **+ **+ *+ *+ *+ **+ ***+ *+
A5 ***+ ***+ **+ ***+ **+ ***+ ***+ ***+ ***+ **+ ***+ **+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+
A6 ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ **+ **+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ **+ ***+ ***+ ***+ ***+ ***+ ***+
A7 **+ **+ ***+ ***+ **+ **+ ***+ **+ *+ ***+ **+
Tech A8 ***+ ***+ ***+ ***+ **+ **+ ***+ ***+ ***+ ***+ ***+ ***+ **+ **+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ ***+ **+ *+ ***+ ***+ ***+ ***+ ***+ ***+
27
5. CONCLUSION
Marshallian agglomeration economies are present in the Spanish economy and significantly
affect entrepreneurial activity. A good costumer base is especially important to create sharing
opportunities as measured by the Output index. The relevance of suppliers for sharing was not
so strong but has some effect. Proximity to innovative companies to take advantage of
knowledge spillovers is also relevant for business formation. The Chinitz index (that
measures the relevance of small suppliers inside the city) was also relevant to explain the
location of new firms but no so much to explain the employment created by these firms. The
labor pooling was not significant for new firms. The results regarding the service industry are
particularly interesting. Localization economies have no effect in services but they are highly
influenced by some of the Marshallian agglomeration factors.
The research also presents some challenges that should be addressed. First the
potential of endogeneity was not solved. The method proposed to solve it was not robust. A
second problem is given by the own indexes, although capturing the essence of what
Marshallian agglomeration economies are, they do not cover the entire theoretical concept.
Finally, the database used was a sample and not a census of new which may have caused
representativeness problems for some industries. It serves as motivation for future work
extending the study to other countries to check for stylized facts and to test the indexes.
Future research should include a deeper analysis in the industrial and geographical
dimension. This work focused on two-digit (CNAE-93) industries and the largest 15
metropolitan areas in Spain. The next step should include three and four-digit industries.
Adding more metropolitan areas can explain if Marshallian agglomeration economies
disappear with city size. Finally explore further the Chinitz hypothesis seems plausible given
the favorable evidence presented and its relevance for new firm formation, specially for cities
like Vigo and Valencia.
ACKNOWLEDGMENTS
The author is grateful to Jose Luis Roig for his valuable comments and advice as thesis
supervisor. This research is supported by the Secretary of Universities and Research (SUR) of
the Department of Economics and Scientific Knowledge of the Generalitat de Catalunya and
the European Social Fund. Support given by the grant ECO2009-10003 of the Spanish
Ministry of Science and Innovation and the grant SGR2009-600 of the Catalonian
Government is also acknowledged.
28
REFERENCES
Acs, Z. J. and Szerb L. (2009). “The global entrepreneurship index (GEINDEX)”, Jena Economics Research
Papers N° 2009-28.
Acs, Z. J. and Audretsch, D. B (2003). “Innovation and Technological Change”. In Z.J. Acs and D.B. Audretsch
(eds.), Handbook of Entrepreneurship Research, 55-79. Kluwer Academic Publishers, Gran Bretaña.
Arauzo, J. M. (2005) “Determinants of Industrial Location. An Application for Catalan Municipalities,” Papers
in Regional Science, 84, 105–120.
Arauzo-Carod, J.M.; Teruel-Carrizosa, M. (2005) “An Urban Approach to Firm Entry: The Effect of City Size”,
Growth and Change, 36 (4), 508-528.
Arauzo-Carod, J.M. and Viladecans-Marsal E. (2009). “Industrial Location at the Intra-Metropolitan Level: The
Role of Agglomeration Economies” Regional Studies, Vol. 43.4, 545–558.
Audretsch, D.B. and Feldman, M.P.(1996). “R&D spillovers and the geography of innovation and production”,
American Economic Review 86(3), 630-640.
Audretsch, D.B. and Keilbach, M. (2004): “Does entrepreneurship capital matter?”, Entrepreneurship:Theory
and Practice, 28(5): 419-429.
Baumol, W. J. (1968). “Entrepreneurship in Economic Theory”, The American Economic Review, Vol. 58, No.
2, Papers and Proceedings of the Eightieth Annual Meeting of the American Economic Association, 64-71
Blanchflower, D. and Oswald, A., (1998). "What Makes an Entrepreneur?," Journal of Labor Economics,
University of Chicago Press, vol. 16(1), 26-60
Boix, R.;and Veneri, P. (2009). “Metropolitan Areas in Spain and Italy” IERMB Working Paper in Economics,
nº 09.01, March 2009
Brock, W. and Evans, D. (1985). "The Economics of Regulatory Tiering," RAND Journal of Economics, vol.
16(3), pages 398-409
Carree, M. A. and Thurik, A. R. (2003). “The Impact of Entrepreneurship on Economic Growth”. In Z.J. Acs
and D.B. Audretsch (eds.), Handbook of Entrepreneurship Research, 437-471. Kluwer Academic Publishers,
Great Britain.
Chinitz, B (1961) “Contrasts in Agglomeration: New York and Pittsburgh”, The American Economic Review,
Vol. 51, No. 2, Papers and Proceedings of the Seventy -Third Annual Meeting of the American Economic
Association, 279-289.
Congregado, E.; and Millán J.M. (2008). “Capital humano y transiciones al autoempleo”, en Congregado et al.
El capital humano y los emprendedores en España, pp. 101-130. IVIE- Bancaja, Instituto Valenciano de
Investigaciones Económicas, Valencia, 226 p.
Costa, D.L. and M.E. Kahn (2001), "Power Couples," Quarterly Journal of Economics 116, 1287-1315.
Duranton, G. and Puga, D. (2004). “Micro-foundations of urban agglomeration economies”. In J. V. Henderson,

P. Nijkamp, E. S. Mills, P. C. Cheshire and J. F. Thisse (ed.) Handbook of Regional and Urban Economics
Capítulo 48, Blackwell, 2063-2117
Ellison, G. and Glaeser, E. (1997). "Geographic Concentration in U.S. Manufacturing Industries: A Dartboard
Approach", Journal of Political Economy 105, 889-927.
29
Ellison, G. and Glaeser E. (1999). "The Geographic Concentration of Industry: Does Natural Advantage Explain
Agglomeration?", American Economic Review Papers and Pro-ceedings 89 (1999), 311-316.
Ellison, G.; Glaeser E. L.; Kerr W. R., (2010). "What Causes Industry Agglomeration? Evidence from
Coagglomeration Patterns," American Economic Review, vol. 100(3), 1195-1213.
Evans, D. and Jovanovic, B. (1989). “An Estimated model of entrepreneurial choice under liquidity constraints”.
Journal of Political Economy. 97, 808-827.
Feria Toribio, J.M. (2004). “Problemas de definición de las áreas metropolitanas en España”. Boletín de la
A.G.E., N° 38, 85-99
Fujita, M., Krugman, P. and Venables, A. (1999). “The Spatial Economy: Cities, Regions, and International
Trade”. Cambridge: Cambridge University Press.
Glaeser, E.L., and D. C. Mare (2001), "Cities and Skills," Journal of Labor Economics 19(2), 316-342.
Glaeser, Edward L., and William R. Kerr (2009). "Local Industrial Conditions and Entrepreneurship: How Much
of the Spatial Distribution Can We Explain?" Journal of Economics and Management Strategy 18, no. 3, 623–
663.
Hebert, R. F. and Link, A. N. (2006). "Historical perspectives of Entrepreneurship", Foundations and Trends in
Entrepreneurship: Vol. 2: No 4, pp 261-408.
Holl, Adelheid (2004c). “Manufacturing Location and Impacts of Road Transport Infrastructure: Empirical
Evidence from Spain,” Regional Science and Urban Economics, 34, 341–363.
Helsley, R. W. and Strange W.C. (1990). “Matching and agglomeration economies in a system of cities”.
Regional Science and Urban Economics 20(2):189–212.
Holmes, T. J. & Schmitz, J. A. Jr., (1990). "A Theory of Entrepreneurship and Its Application to the Study of
Business Transfers," Journal of Political Economy, vol. 98(2), pages 265-94
Holmes, T. J. (1999), “Scale of Local Production and City Size”. The American Economic Review, 89 (2),
Papers and Proceedings of the One Hundred Eleventh Annual Meeting of the American Economic Association,
317 -320
Iversen, J.; Jørgensen, R.; Malchow-Møller, N. (2008) "Defining and Measuring Entrepreneurship", Foundations
and Trends in Entrepreneurship: Vol. 4: No 1, pp 1-63.
Jacobs, J. (1971). “La Economía de las ciudades”. Barcelona : Península.
Jofre-Monseny, J.; Marín-López, R.; Viladecans-Marsal, E. (2011) "The mechanisms of agglomeration:

Evidence from the effect of inter-industry relations on the location of new firms," Journal of Urban Economics,
Elsevier, vol. 70(2-3), pp 61-74.
Jovanovic, B. (1982) "Selection and the Evolution of Industry," Econometrica, Econometric Society, vol. 50(3),
pp 649-70
Krugman, P. (1992). "Geography and Trade," MIT Press Books.
Marshalll, A. (1920b), Principles of Economics. London: Macmillan, 8th edition.
Rosenthal, S. S. and Strange, W. C. (2001), "The Determinants of Agglomeration," Journal of Urban Economics
50, 191-229.
Rosenthal, S. S. and Strange, W. C. (2003), "Geography, Industrial Organization, and Agglomeration," Review
of Economics and Statistics 85 (2), 377-393.
30
Rosenthal, S. S. and Strange, W. C. (2005), "The geography of entrepreneurship in the New York metropolitan
area," Economic Policy Review, Federal Reserve Bank of New York, issue Dec, 29-53.
Rosenthal, S. S. and Strange, W. C., (2010), "Small Establishments/Big Effects: Agglomeration, Industrial
Organization and Entrepreneurship," NBER Chapters, in: Agglomeration Economics, pages 277-302 National
Bureau of Economic Research, Inc.
Segarra-Blasco, A. (2007). "Innovation, R&D Spillovers and Productivity: The Role of Knowledge-Intensive
Services," Working Papers XREAP2007-12, Xarxa de Referència en Economia Aplicada (XREAP).
Van Praag, C.M. and H. van Ophem (1995) “Determinants of willingness and opportunity to start as an
entrepreneur”, Kyklos, 48, 513-540.
ANNEX
Tabla A0: Metropolitan Areas' Population and Employment
Employment
through Natural
Cost Advantages
Metropolitan Area Population Employment Calculation
Madrid 5,806,548 2,594,778 1,065,813
Barcelona 4,512,283 2,006,257 984,446
Valencia 1,696,810 702,494 677,409
Sevilla 1,396,538 482,528 601,116
Bilbao 1,094,644 433,310 619,220
Oviedo 897,681 324,038 514,434
Málaga 790,322 282,780 504,094
Zaragoza 786,055 329,605 549,485
La Palma de Gran Canaria 628,343 235,445 477,744
Alicante 572,617 243,040 464,221
Murcia 555,872 224,480 451,424
Vigo 550,531 220,087 456,918
Santa Cruz de Tenerife 538,450 210,682 443,160
Granada 534,326 179,281 431,143
Palma de Mallorca 528,634 238,101 466,277
TOTAL 20,889,654 8,706,906 8,706,906
31
Table A1: Estimation derived from Natural Cost Advantages
Base Estimation Input, Output and Labor Plus Chinitz Plus Tech
1 2 3 4
Log Employment city-industry 0.408 ** 2.376 0.228 1.028 0.228 0.961 0.249 1.038
Input index 0.142 0.656 0.141 0.619 0.133 0.581
Output index 0.069 1.023 0.069 1.023 0.059 0.858
Labor index -0.001 -0.005 -0.005 -0.024
Tech index 0.043 1.261
N Obs 900 900 900 900

2
R 0.911 0.911 0.911 0.911
2
Adjusted R 0.902 0.903 0.902 0.903
Log Employment city-industry 0.400 ** 2.556 0.442 ** 2.454 0.598 *** 2.995 0.617 *** 3.069
Input index 0.064 0.354 0.214 1.075 0.206 1.035
Output index -0.033 -0.760 -0.033 -0.752 -0.042 -0.954
Labor index 0.441 ** 2.439 0.437 ** 2.419
Tech index 0.039 * 1.650
N Obs 900 900 900 900

2
R 0.941 0.941 0.941 0.942
2
Adjusted R 0.936 0.936 0.936 0.936
Table A2
Descriptive Stats from Natural Cost Advantages
Mean S.D. Maximun (Sector) Minimum (Sector)
1 2 3 (City) 4 (City)
Employees by city-industry 9674 15861 126176 (s45) (MAD) 159 (s2) (GRA)
Input index -1.3031 0.1588 -0.9396 (s73) (MAD) -1.7895 (s23) (PGC)
Output index 0.9480 0.2245 1.4507 (s01) (BIL) 0.0000 Many
Labor index -1.0967 0.2443 -0.7104 (s51) (MAD) -1.7244 (s95) (ZAR)
Tech index -0.9725 0.0122 -0.9418 (s65) (VGO) -1.0091 (s45) (BIL)
32
Table A3:Conditional Estimation with Rotated Principal Components (DIRCE)
Total Manufacturing Energy Services
1 2 3 4
Emp_ci Factor 0.039 ** 2.215 0.119 * 1.741 2.847 *** 4.554 0.005 0.240
Chinitz Factor 0.051 1.438 0.236 1.464 0.555 1.358 -0.041 -0.788
Input Factor 0.101 * 1.754 -0.030 -0.222 0.258 1.203 0.098 1.620
Output Factor 0.119 *** 3.360 0.134 *** 3.187 0.161 *** 3.882 0.285 1.612
Labor Factor 0.452 *** 7.226 0.579 *** 3.589 -0.623 * -1.668 0.406 *** 3.035
Tech Factor 0.047 1.471 0.202 *** 2.699 0.300 1.394 -0.039 -0.767
N Obs 810 330 120 345

2
R 0.897 0.840 0.828 0.907
2
Adjusted R 0.887 0.818 0.778 0.894
Industrial Fixed Effects X X X X
Emp_ci Factor 0.057 *** 4.617 0.101 ** 2.398 1.620 *** 6.151 0.018 * 1.722
Chinitz Factor 0.078 *** 2.935 0.210 * 1.791 0.357 1.422 0.011 0.327
Input Factor 0.066 * 1.751 0.018 0.185 0.261 *** 2.867 0.051 1.386
Output Factor 0.088 *** 7.460 0.101 *** 5.651 0.084 *** 4.648 0.235 *** 3.113
Labor Factor 0.260 *** 6.444 0.304 ** 2.469 -0.224 -1.326 0.235 *** 4.125
Tech Factor 0.058 *** 2.798 0.158 *** 2.719 0.167 1.173 0.001 0.059
N Obs 810 330 120 345

2
R 0.957 0.903 0.885 0.978
2
Adjusted R 0.952 0.890 0.851 0.974
Industrial Fixed Effects X X X X
33
Table A4: Conditional Estimation by Cities
Madrid Barcelona Valencia Sevilla Bilbao
1 2 3 4 5
Log Employment city-industry 0.642 *** 5.513 0.676 *** 5.546 0.615 *** 6.867 0.615 *** 4.304 0.579 *** 5.052
Chinitz index -0.014 -0.155 0.057 0.647 0.003 0.041 -0.133 -1.295 0.061 0.607
Input index 0.357 0.891 0.575 ** 2.239 0.551 ** 2.076 0.102 0.173 0.136 ** 1.980
Output index -0.116 -1.588 -0.017 -0.284 0.090 1.586 0.000 0.006 0.005 0.062
Labor index 0.326 ** 2.549 0.227 1.375 0.144 1.465 0.280 ** 2.202 0.179 1.000
Tech index 0.108 0.909 0.153 1.603 0.272 *** 2.852 0.196 1.104 0.225 * 1.648
N Obs 60 60 60 60 60
2
R 0.704 0.792 0.808 0.690 0.646
2
Adjusted R 0.670 0.768 0.787 0.655 0.605
Industrial Fixed Effects (CNAE-93) No No No No No
City Fixed Effects No No No No No
Chinitz index 0.004 0.045 0.046 0.576 -0.040 -0.491 -0.086 -0.840 0.013 0.147
Input index 0.481 1.197 0.572 * 1.933 0.330 1.143 0.130 0.213 0.070 1.436
Output index -0.114 -1.457 -0.030 -0.475 0.057 0.894 -0.056 -0.625 -0.046 -0.559
Labor index 0.297 ** 2.176 0.259 1.549 0.186 1.527 0.287 ** 2.285 0.160 0.833
Tech index 0.184 1.534 0.223 ** 2.135 0.287 ** 2.378 0.220 1.195 0.270 * 1.903
N Obs 60 60 60 60 60
2
R 0.735 0.799 0.779 0.690 0.680
2
Adjusted R 0.705 0.777 0.753 0.655 0.644
Oviedo Málaga Zaragoza La Palma de Gran Canaria Alicante
6 7 8 9 10
Chinitz index -0.031 -0.262 -0.136 -1.202 0.034 0.431 -0.030 -0.319 -0.169 * -1.925
Input index 0.025 0.884 0.148 0.274 0.268 0.970 -0.185 -1.119 0.038 ** 2.137
Output index -0.022 -0.209 -0.021 -0.283 -0.008 -0.119 0.034 0.446 -0.008 -0.106
Labor index 0.274 1.564 0.156 * 1.826 0.244 1.276 0.246 ** 2.166 0.252 1.360
Tech index 0.053 0.342 0.184 1.012 0.123 1.065 0.216 1.625 0.142 0.839
N Obs 60 60 60 60 60
2
R 0.578 0.686 0.729 0.691 0.678
Adjusted R2 0.530 0.650 0.699 0.657 0.641
Chinitz index -0.047 -0.571 -0.093 -0.878 -0.075 -0.961 -0.012 -0.159 -0.099 -1.313
Input index -0.037 -1.375 0.081 0.139 0.267 0.841 -0.086 -0.509 0.034 ** 2.108
Output index -0.029 -0.326 -0.027 -0.335 -0.040 -0.552 0.018 0.268 -0.033 -0.458
Labor index 0.233 1.471 0.142 * 1.680 0.270 1.472 0.211 * 1.910 0.206 1.253
Tech index 0.189 1.414 0.262 1.408 0.215 * 1.804 0.234 * 1.839 0.242 1.550
N Obs 60 60 60 60 60
2
R 0.666 0.675 0.718 0.695 0.704
2
Adjusted R 0.628 0.638 0.687 0.661 0.670
Murcia Vigo Santa Cruz de Tenerife Granada Palma de Mallorca
11 12 13 14 15
Chinitz index -0.078 -0.846 -0.043 -0.517 -0.049 -0.524 -0.044 -0.546 -0.106 -1.154
Input index 0.046 0.280 0.167 0.882 -0.018 -0.779 -0.158 -0.253 0.021 0.052
Output index 0.083 1.067 -0.003 -0.038 -0.070 -0.951 0.000 0.003 0.015 0.164
Labor index 0.213 ** 2.133 -0.065 -0.910 0.156 * 1.813 0.197 ** 2.278 0.217 * 1.802
Tech index 0.153 1.242 0.387 *** 4.407 0.273 ** 2.079 0.120 0.670 0.057 0.626
N Obs 60 60 60 60 60
2
R 0.712 0.668 0.722 0.657 0.624
2
Adjusted R 0.679 0.631 0.690 0.619 0.582
Chinitz index -0.087 -0.977 -0.039 -0.486 -0.044 -0.570 -0.066 -0.698 -0.081 -0.885
Input index -0.028 -0.156 0.078 0.414 -0.060 -1.352 -0.113 -0.230 0.102 0.267
Output index 0.041 0.510 -0.007 -0.109 -0.026 -0.335 -0.003 -0.043 0.005 0.055
Labor index 0.200 * 1.905 -0.043 -0.727 0.169 * 1.668 0.220 *** 2.967 0.131 1.228
Tech index 0.202 1.575 0.434 *** 5.515 0.263 ** 2.014 0.166 0.998 0.169 * 1.820
N Obs 60 60 60 60 60
2
R 0.699 0.657 0.694 0.675 0.626
2
Adjusted R 0.665 0.618 0.660 0.638 0.584
34
Tabla A5: Conditional Estimation by Cities using SURE
1 2 3 4 5
Chinitz index 0.186 *** 2.658 0.194 *** 3.009 0.311 *** 5.493 0.251 *** 3.255 0.276 *** 3.969
Input index -0.017 -0.296 0.040 0.825 -0.007 -0.173 -0.111 * -1.948 0.053 0.968
Output index -0.064 -0.472 0.248 * 1.831 0.316 *** 3.612 -0.048 -0.390 0.082 * 1.756
Labor index -0.071 -1.329 -0.026 -0.565 0.067 1.603 -0.001 -0.020 0.004 0.079
Tech index 0.271 *** 3.524 0.194 ** 2.451 0.135 ** 2.488 0.244 *** 4.065 0.128 1.612
N Obs 60 60 60 60 60
2
R 0.671 0.758 0.780 0.662 0.614
2
Adjusted R 0.634 0.730 0.755 0.624 0.570
Chinitz index 0.222 *** 3.408 0.230 *** 4.294 0.281 *** 5.631 0.206 *** 3.268 0.263 *** 5.058
Input index 0.000 0.004 0.050 1.250 -0.010 -0.292 -0.026 -0.556 0.047 1.195
Output index 0.038 0.283 0.267 ** 2.449 0.260 *** 3.815 0.128 1.355 0.006 0.206
Labor index -0.050 -0.998 -0.005 -0.124 0.076 ** 2.089 -0.025 -0.534 -0.014 -0.336
Tech index 0.265 *** 3.634 0.207 *** 3.156 0.148 *** 3.131 0.243 *** 4.931 0.125 *** 2.236
N Obs 60 60 60 60 60
2
R 0.696 0.762 0.746 0.648 0.637
2
Adjusted R 0.662 0.735 0.717 0.608 0.596
6 7 8 9 10
Chinitz index 0.151 ** 2.094 0.178 ** 2.108 0.152 ** 2.454 0.224 *** 3.242 0.152 ** 2.110
Input index -0.034 -0.578 -0.080 -1.333 0.042 0.885 0.010 0.180 -0.141 ** -2.562
Output index 0.021 0.869 0.160 1.099 0.094 0.920 -0.145 *** -2.593 0.053 *** 5.256
Labor index -0.021 -0.365 -0.027 -0.471 -0.025 -0.558 0.054 0.954 -0.039 -0.796
Tech index 0.152 ** 2.370 0.160 ** 2.824 0.202 *** 2.941 0.203 *** 3.624 0.235 *** 3.533
N Obs 60 60 60 60 60
2
R 0.545 0.665 0.696 0.679 0.655
Adjusted R2 0.493 0.627 0.662 0.643 0.616
Chinitz index 0.202 *** 3.828 0.212 *** 2.959 0.241 *** 4.820 0.206 *** 4.232 0.217 *** 3.941
Input index -0.013 -0.309 -0.036 -0.716 -0.046 -1.242 0.043 1.086 -0.043 -1.064
Output index -0.042 ** -2.119 0.142 1.215 0.108 1.425 -0.007 -0.182 0.050 *** 6.146
Labor index -0.002 -0.053 -0.012 -0.236 -0.008 -0.214 0.050 1.237 -0.012 -0.315
Tech index 0.163 *** 3.536 0.144 *** 2.847 0.157 *** 2.877 0.172 *** 4.971 0.190 *** 3.829
N Obs 60 60 60 60 60
2
R 0.628 0.640 0.671 0.659 0.664
2
Adjusted R 0.585 0.599 0.633 0.621 0.626
11 12 13 14 15
Chinitz index 0.184 ** 2.230 0.383 *** 5.787 0.263 *** 3.914 0.125 1.590 0.050 0.857
Input index -0.057 -1.021 -0.031 -0.542 0.005 0.100 0.004 0.073 -0.027 -0.430
Output index -0.012 -0.119 0.095 1.233 0.023 1.097 -0.056 -0.385 0.045 0.398
Labor index 0.060 1.072 -0.040 -0.733 -0.054 -0.968 -0.019 -0.307 -0.002 -0.033
Tech index 0.187 *** 2.665 0.027 0.710 0.143 *** 2.792 0.202 *** 3.504 0.248 *** 3.647
N Obs 60 60 60 60 60
2
R 0.690 0.636 0.704 0.633 0.599
2
Adjusted R 0.655 0.595 0.671 0.591 0.553
Chinitz index 0.170 ** 2.456 0.390 *** 7.000 0.230 ** 4.402 0.146 *** 2.601 0.112 ** 2.479
Input index -0.038 -0.821 -0.002 -0.051 0.006 0.139 -0.009 -0.213 0.008 0.170
Output index -0.115 -1.412 0.048 0.751 -0.026 * -1.848 -0.074 -0.795 0.162 * 1.855
Labor index 0.055 1.166 0.008 0.192 0.004 0.101 0.011 0.256 0.011 0.232
Tech index 0.199 *** 3.373 0.018 0.575 0.140 *** 3.599 0.213 *** 5.274 0.163 *** 3.156
N Obs 60 60 60 60 60
2
R 0.664 0.624 0.661 0.633 0.584
2
Adjusted R 0.626 0.581 0.623 0.592 0.537
35
Tabla A6: Conditional Estimation by City using Rotated Principal Components
1 2 3 4 5
Emp_ci Factor 0.188 *** 4.566 0.302 *** 6.002 0.569 *** 4.138 0.528 ** 2.263 0.770 *** 2.838
Chinitz Factor 0.199 ** 2.090 0.240 *** 2.955 0.362 *** 4.474 0.318 *** 2.684 0.297 *** 2.983
Input Factor 0.080 0.859 0.160 * 1.920 0.148 * 1.764 0.067 0.963 0.118 1.159
Output Factor 0.571 1.261 1.268 *** 4.704 1.077 *** 2.870 0.630 1.325 -0.038 -0.521
Labor Factor -0.045 -0.395 0.034 0.379 0.128 1.643 0.035 0.326 0.002 0.017
Tech Factor 0.500 *** 3.659 0.435 *** 2.891 0.373 *** 3.144 0.487 *** 4.229 0.470 *** 3.447
N Obs 60 60 60 60 60
2
R 0.532 0.656 0.653 0.519 0.507
2
Adjusted R 0.479 0.617 0.614 0.464 0.451
Industrial Fixed Effects No No No No No
Emp_ci Factor 0.215 *** 4.738 0.311 *** 5.899 0.635 *** 3.888 0.540 ** 2.086 0.835 *** 2.846
Chinitz Factor 0.262 *** 2.910 0.307 *** 3.776 0.372 *** 3.906 0.342 *** 2.839 0.325 *** 3.227
Input Factor 0.112 1.229 0.156 * 1.942 0.118 1.501 0.109 * 1.698 0.086 0.980
Output Factor 0.720 1.609 1.231 *** 4.928 0.897 ** 2.414 0.632 1.351 -0.096 -1.535
Labor Factor -0.038 -0.317 0.024 0.288 0.103 1.308 -0.021 -0.203 -0.045 -0.496
Tech Factor 0.511 *** 3.493 0.472 *** 3.187 0.403 *** 2.906 0.488 *** 4.360 0.451 *** 3.301
N Obs 60 60 60 60 60
2
R 0.569 0.693 0.637 0.545 0.566
2
Adjusted R 0.520 0.658 0.596 0.493 0.517
6 7 8 9 10
Emp_ci Factor 0.833 ** 2.040 1.115 *** 3.410 1.023 *** 2.824 0.810 * 1.783 1.591 ** 2.450
Chinitz Factor 0.219 * 1.954 0.299 ** 2.219 0.338 *** 3.844 0.377 *** 3.697 0.384 *** 2.719
Input Factor 0.081 0.810 0.157 * 1.850 0.241 *** 2.935 0.166 ** 2.029 0.122 1.217
Output Factor 0.016 0.484 1.082 *** 2.633 0.635 ** 2.148 -0.084 -0.464 0.042 1.376
Labor Factor -0.071 -0.654 0.025 0.289 -0.025 -0.292 0.078 0.805 0.050 0.475
Tech Factor 0.465 *** 3.857 0.280 ** 2.421 0.573 *** 4.159 0.461 *** 4.254 0.565 *** 4.310
N Obs 60 60 60 60 60
2
R 0.425 0.533 0.572 0.544 0.525
Adjusted R2 0.360 0.480 0.523 0.492 0.472
Emp_ci Factor 0.847 ** 2.166 1.130 *** 3.326 0.949 ** 2.351 0.918 * 1.886 1.482 ** 2.408
Chinitz Factor 0.324 *** 3.072 0.361 *** 2.626 0.400 *** 4.465 0.387 *** 3.724 0.447 *** 3.379
Input Factor 0.084 1.055 0.197 *** 2.795 0.130 1.576 0.188 ** 2.811 0.176 * 1.902
Output Factor -0.042 -1.306 0.961 ** 2.372 0.588 * 1.882 0.018 0.097 0.038 1.314
Labor Factor -0.076 -0.815 0.018 0.202 -0.042 -0.538 0.061 0.737 0.017 0.187
Tech Factor 0.461 *** 3.938 0.282 ** 2.468 0.588 *** 4.770 0.430 *** 4.222 0.519 *** 4.375
N Obs 60 60 60 60 60
2
R 0.527 0.553 0.606 0.566 0.587
2
Adjusted R 0.474 0.503 0.561 0.517 0.541
11 12 13 14 15
Emp_ci Factor 1.526 *** 3.005 1.628 *** 3.115 1.257 *** 2.526 0.892 1.413 1.156 *** 3.183
Chinitz Factor 0.361 *** 3.144 0.610 *** 5.813 0.487 *** 4.926 0.356 ** 2.223 0.264 *** 3.342
Input Factor 0.190 ** 2.335 0.252 *** 2.608 0.199 *** 2.765 0.214 ** 2.530 0.189 ** 2.275
Output Factor 0.263 1.323 0.378 * 1.712 0.000 0.004 0.319 0.649 0.708 ** 2.122
Labor Factor 0.090 0.981 -0.038 -0.435 -0.035 -0.392 -0.027 -0.280 0.052 0.433
Tech Factor 0.525 *** 4.577 0.229 ** 2.308 0.425 *** 3.802 0.401 *** 4.128 0.363 *** 3.091
N Obs 60 60 60 60 60
2
R 0.547 0.494 0.552 0.430 0.481
2
Adjusted R 0.496 0.437 0.501 0.366 0.422
Emp_ci Factor 1.504 *** 2.791 1.592 *** 2.893 1.328 ** 2.378 1.043 * 1.758 1.156 *** 3.275
Chinitz Factor 0.387 ** 3.281 0.635 *** 6.075 0.470 *** 4.318 0.390 *** 2.851 0.333 *** 4.243
Input Factor 0.178 ** 2.220 0.250 *** 2.617 0.196 *** 2.649 0.187 *** 2.747 0.201 *** 2.833
Output Factor 0.168 0.808 0.257 1.168 -0.039 -1.392 0.339 0.856 0.699 ** 2.368
Labor Factor 0.048 0.538 -0.037 -0.465 0.008 0.086 -0.019 -0.217 0.036 0.330
Tech Factor 0.506 *** 4.484 0.250 *** 3.011 0.419 *** 3.826 0.418 *** 4.534 0.306 *** 2.791
N Obs 60 60 60 60 60
2
R 0.557 0.520 0.575 0.505 0.533
2
Adjusted R 0.507 0.466 0.527 0.449 0.480
36
Tabla A7: Conditional Estimation by Cities (DIRCE)
1 2 3 4 5
Chinitz index -0.096 -1.084 0.063 0.748 0.198 * 1.890 -0.040 -0.315 0.113 1.087
Input index 0.256 *** 3.246 0.206 *** 3.273 0.287 *** 3.929 0.274 ** 2.647 0.272 *** 3.100
Output index -0.098 -0.410 0.533 ** 2.009 0.320 0.945 -0.707 ** -2.109 0.008 0.105
Labor index -0.041 -0.685 -0.099 * -1.736 -0.001 -0.018 -0.024 -0.305 0.055 0.723
Tech index 0.176 ** 2.008 0.109 1.035 0.048 0.481 0.268 *** 3.036 0.090 0.701
N Obs 54 54 54 54 54
2
R 0.851 0.884 0.831 0.818 0.752
2
Adjusted R 0.832 0.869 0.810 0.794 0.720
Chinitz index 0.054 0.548 0.196 ** 2.013 0.259 ** 2.106 0.159 1.246 0.309 *** 3.279
Input index 0.301 *** 3.457 0.207 *** 2.860 0.237 *** 2.759 0.244 ** 2.344 0.192 ** 2.413
Output index 0.061 0.230 0.239 0.781 0.296 0.742 -0.700 ** -2.081 0.044 0.608
Labor index -0.131 ** -1.982 -0.108 -1.636 -0.036 -0.480 -0.074 -0.937 -0.067 -0.965
Tech index 0.239 ** 2.456 0.155 1.281 0.126 1.070 0.252 *** 2.846 0.004 0.037
N Obs 54 54 54 54 54
2
R 0.847 0.853 0.805 0.813 0.793
2
Adjusted R 0.828 0.834 0.780 0.789 0.766
6 7 8 9 10
Chinitz index 0.038 0.405 -0.079 -0.529 0.099 1.021 0.099 0.699 -0.084 -0.700
Input index 0.326 *** 3.780 0.400 *** 3.603 0.285 *** 3.895 0.388 *** 2.996 0.380 *** 3.807
Output index 0.020 0.437 -0.263 -0.747 0.188 0.748 0.051 0.370 0.001 0.033
Labor index -0.012 -0.170 -0.110 -1.395 -0.127 ** -2.039 0.112 1.242 -0.058 -0.769
Tech index 0.133 1.468 0.230 ** 2.809 0.078 0.668 0.093 1.020 0.332 *** 3.202
N Obs 54 54 54 54 54
2
R 0.780 0.822 0.826 0.761 0.812
Adjusted R2 0.752 0.800 0.804 0.730 0.788
Chinitz index 0.298 *** 2.747 0.152 0.961 0.194 * 1.842 0.283 ** 2.054 0.096 0.764
Input index 0.239 ** 2.374 0.273 ** 2.309 0.190 ** 2.389 0.240 * 1.917 0.317 *** 3.035
Output index -0.035 -0.659 -0.192 -0.513 -0.137 -0.503 -0.145 -1.082 -0.006 -0.270
Labor index -0.094 -1.119 -0.119 -1.424 -0.109 -1.601 -0.028 -0.324 -0.085 -1.084
Tech index 0.072 0.684 0.175 ** 2.009 0.175 1.377 0.087 0.982 0.231 ** 2.127
N Obs 54 54 54 54 54
2
R 0.753 0.805 0.804 0.773 0.794
2
Adjusted R 0.721 0.781 0.779 0.744 0.767
11 12 13 14 15
Chinitz index 0.062 0.551 0.195 * 1.920 0.127 1.005 0.130 1.117 0.049 0.586
Input index 0.264 *** 3.213 0.352 *** 3.365 0.362 *** 3.180 0.269 ** 2.612 0.300 *** 2.921
Output index -0.076 -0.267 0.189 1.110 0.058 1.501 -0.480 * -1.740 0.349 1.320
Labor index -0.031 -0.438 0.039 0.484 0.050 0.631 -0.079 -1.090 0.092 1.119
Tech index 0.142 1.580 -0.069 -0.958 0.140 * 1.731 0.195 *** 2.813 0.060 0.662
N Obs 54 54 54 54 54
2
R 0.821 0.768 0.804 0.836 0.801
2
Adjusted R 0.798 0.739 0.779 0.815 0.775
Chinitz index 0.205 * 1.809 0.296 *** 2.877 0.281 ** 2.190 0.190 1.396 0.133 1.529
Input index 0.193 ** 2.337 0.294 *** 2.769 0.243 ** 2.104 0.224 * 1.864 0.276 ** 2.551
Output index -0.196 -0.686 -0.036 -0.211 -0.039 -0.993 -0.434 -1.347 0.110 0.397
Labor index -0.076 -1.056 -0.028 -0.348 -0.046 -0.575 -0.065 -0.774 -0.026 -0.301
Tech index 0.128 1.415 -0.111 -1.519 0.098 1.194 0.204 ** 2.526 0.087 0.906
N Obs 54 54 54 54 54
2
R 0.816 0.744 0.797 0.779 0.777
2
Adjusted R 0.793 0.712 0.771 0.751 0.748
37
Tabla A8: Conditional Estimation by City using Rotated Principal Components (DIRCE)
1 2 3 4 5
Emp_ci Factor 0.220 *** 6.910 0.270 *** 7.612 0.540 *** 4.271 0.690 *** 3.876 0.942 *** 4.855
Chinitz Factor -0.035 -0.315 0.232 *** 2.903 0.363 *** 3.837 0.109 0.662 0.179 * 1.882
Input Factor 0.318 *** 2.854 0.223 ** 2.228 0.378 *** 4.636 0.443 *** 3.788 0.325 *** 2.993
Output Factor 0.229 0.397 1.736 *** 3.852 1.396 *** 3.171 0.220 0.402 -0.207 * -1.737
Labor Factor 0.077 0.601 -0.007 -0.072 0.040 0.396 0.003 0.020 0.100 0.857
Tech Factor 0.364 *** 2.973 0.403 *** 3.481 0.188 ** 2.187 0.430 *** 3.841 0.422 *** 3.248
N Obs 54 54 54 54 54
2
R 0.580 0.717 0.732 0.587 0.574
2
Adjusted R 0.527 0.681 0.698 0.535 0.519
Emp_ci Factor 0.240 *** 6.792 0.294 *** 6.501 0.563 *** 3.835 0.691 *** 3.341 0.958 *** 4.091
Chinitz Factor 0.131 1.179 0.338 *** 3.605 0.434 *** 3.986 0.271 * 1.833 0.323 *** 3.144
Input Factor 0.380 *** 3.295 0.251 *** 2.959 0.343 *** 3.198 0.427 *** 3.386 0.284 *** 3.260
Output Factor 0.348 0.762 1.324 *** 3.953 1.374 *** 3.216 0.093 0.201 -0.155 -1.452
Labor Factor -0.024 -0.184 -0.015 -0.136 0.013 0.123 -0.043 -0.345 -0.019 -0.165
Tech Factor 0.438 *** 3.711 0.446 *** 3.734 0.276 ** 2.556 0.442 *** 4.015 0.382 *** 2.826
N Obs 54 54 54 54 54
2
R 0.684 0.734 0.725 0.651 0.640
2
Adjusted R 0.644 0.700 0.690 0.606 0.594
6 7 8 9 10
Emp_ci Factor 1.095 *** 4.770 0.899 *** 3.509 0.962 *** 4.341 1.408 *** 4.350 1.342 *** 3.481
Chinitz Factor 0.203 ** 2.195 0.106 0.588 0.314 *** 4.036 0.352 ** 2.634 0.254 ** 2.312
Input Factor 0.486 *** 4.543 0.598 *** 4.630 0.441 *** 5.125 0.592 *** 4.451 0.595 *** 4.451
Output Factor -0.009 -0.356 0.853 ** 1.996 0.750 ** 2.345 0.134 0.806 0.002 0.074
Labor Factor -0.027 -0.254 -0.050 -0.439 -0.159 -1.566 0.230 ** 2.063 0.048 0.421
Tech Factor 0.350 *** 3.587 0.266 ** 2.424 0.440 *** 4.109 0.335 *** 4.217 0.555 *** 5.678
N Obs 54 54 54 54 54
2
R 0.609 0.690 0.683 0.646 0.696
Adjusted R2 0.560 0.651 0.643 0.601 0.657
Emp_ci Factor 1.029 *** 3.544 0.993 *** 3.327 1.038 *** 3.728 1.354 *** 3.450 1.426 *** 3.314
Chinitz Factor 0.400 *** 3.972 0.286 1.602 0.404 *** 4.179 0.485 *** 3.726 0.378 *** 3.213
Input Factor 0.428 *** 4.187 0.511 *** 3.492 0.373 *** 4.081 0.479 *** 3.183 0.562 *** 3.716
Output Factor -0.053 ** -2.020 0.903 ** 2.276 0.379 1.385 -0.057 -0.392 -0.006 -0.190
Labor Factor -0.105 -0.887 -0.049 -0.432 -0.127 -1.161 0.095 0.872 0.021 0.185
Tech Factor 0.350 *** 3.132 0.283 ** 2.424 0.542 *** 5.262 0.371 *** 3.653 0.507 *** 5.107
N Obs 54 54 54 54 54
2
R 0.639 0.695 0.688 0.654 0.699
2
Adjusted R 0.593 0.656 0.648 0.609 0.660
11 12 13 14 15
Emp_ci Factor 0.956 ** 2.303 1.806 *** 4.061 1.376 *** 3.507 1.173 *** 2.826 1.225 *** 6.454
Chinitz Factor 0.247 ** 2.171 0.543 *** 4.825 0.406 *** 3.726 0.368 *** 2.765 0.264 *** 3.826
Input Factor 0.443 *** 3.820 0.628 *** 4.706 0.570 *** 4.241 0.523 *** 4.669 0.536 *** 5.214
Output Factor 0.732 1.497 0.422 * 1.845 0.068 * 1.798 0.258 0.508 1.130 *** 5.146
Labor Factor -0.084 -0.672 0.046 0.369 0.151 1.329 -0.102 -0.872 0.179 * 1.727
Tech Factor 0.397 *** 4.453 0.204 ** 2.059 0.371 *** 4.097 0.395 *** 3.316 0.245 *** 3.194
N Obs 54 54 54 54 54
2
R 0.633 0.581 0.687 0.654 0.729
2
Adjusted R 0.587 0.527 0.647 0.610 0.694
Emp_ci Factor 0.994 ** 2.139 1.721 *** 3.460 1.393 *** 3.227 1.151 ** 2.184 1.177 *** 4.460
Chinitz Factor 0.356 *** 2.926 0.571 *** 4.563 0.513 *** 4.471 0.413 *** 2.879 0.327 *** 4.175
Input Factor 0.398 *** 3.208 0.572 *** 4.273 0.482 *** 3.478 0.469 *** 3.092 0.517 *** 4.756
Output Factor 0.556 1.137 0.148 0.631 -0.025 -0.856 0.237 0.513 0.822 *** 3.164
Labor Factor -0.122 -1.021 -0.022 -0.180 0.058 0.544 -0.081 -0.705 0.055 0.546
Tech Factor 0.412 *** 4.208 0.167 * 1.739 0.367 *** 3.566 0.409 *** 3.553 0.272 *** 2.852
N Obs 54 54 54 54 54
2
R 0.648 0.602 0.691 0.637 0.723
2
Adjusted R 0.603 0.551 0.651 0.590 0.687
38

Dopeso Fernandez 2011

Transféré par

Informations du document

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Dopeso Fernandez 2011

Transféré par

Droits d'auteur :

Formats disponibles

MARSHALLIAN AGGLOMERATION ECONOMIES AND

ENTREPRENEURSHIP: THE SPANISH CASE

Key Words: Entrepreneurship, Industrial Organization, Agglomeration, Labor Markets, Input-

Electronic copy available at: http://ssrn.com/abstract=2099183

Electronic copy available at: http://ssrn.com/abstract=2099183

2.1. STATE OF ENTREPRENEURSHIP IN SPAIN

Figure 1: Spanish Firms Dynamics

Figure 3: Births by size

2.2. ENTREPRENEURS IN ECONOMIC THEORY

2.3. AGGLOMERATION ECONOMIES

2.4. EMPIRICAL EVIDENCE

3. DATA AND VARIABLES

economies. It is taken from de INE 2001 Census.

(A) Actual Data

η i + φ c + β E ln( Emp ci ) + β Ch Chinitz ci 

Base Estimation City Demographics Marshallian Agglomeration Full Estimation

Coefficient (t-stat) Coefficient (t-stat) Coefficient (t-stat) Coefficient (t-stat)

% Employees in the Core Municipality 0.044 ** 2.379 0.048 ** 2.597

% Population younger than 20 years 0.542 * 1.717 0.560 * 1.809

% Population 20-40 years 0.391 * 1.955 0.415 ** 2.116

% Population older than 60 years 0.851 ** 1.973 0.887 ** 2.097

% Men / Women -0.102 * -1.742 -0.104 * -1.796

Human Capital Index -0.160 *** -2.855 -0.161 *** -2.914

% Population without studies -0.612 ** -2.150 -0.613 ** -2.191

% Population with first level education -0.224 * -1.801 -0.212 * -1.736

% Population second level eduaction -0.453 ** -2.200 -0.448 ** -2.219

% Population third level eduaction -0.338 -1.610 -0.330 -1.603

Diversity Index (HHI) 0.065 0.902 0.075 1.043

Entrepreneurial Culture Index 0.031 1.141 0.036 1.312

Chinitz index 0.072 ** 2.178 0.046 1.449

Input index 0.065 * 1.752 0.043 0.962

Output index 0.048 *** 4.526 0.053 *** 5.023

Labor index 0.030 0.607 -0.019 -0.348

Tech index 0.040 * 1.850 0.042 ** 2.057

N Obs 900 900 900 900

Industrial Fixed Effects (CNAE-93) X X X X

Table 4: Non-Conditional Estimation Number of Firms

Base Estimation City Demographics Marshallian Agglomeration Full Estimation

% Firms in the Core Municipality -0.032 ** -2.236 -0.029 ** -2.078

% Population younger than 20 years 0.240 1.180 0.249 1.263

% Population 20-40 years 0.201 1.468 0.213 1.600

% Population older than 60 years 0.416 1.449 0.434 1.555

% Men / Women -0.042 -1.035 -0.041 -1.060

Human Capital Index -0.170 *** -3.626 -0.169 *** -3.680

% Population without studies -0.294 -1.546 -0.289 -1.573

% Population with first level education -0.127 -1.482 -0.117 -1.372

% Population second level eduaction -0.236 * -1.701 -0.229 * -1.692

% Population third level eduaction -0.112 -0.821 -0.104 -0.785

Diversity Index (HHI) -0.058 -1.211 -0.055 -1.168

Entrepreneurial Culture Index 0.054 ** 2.490 0.058 *** 2.740

Chinitz index 0.089 *** 3.072 0.073 *** 2.594

Input index 0.075 ** 2.267 0.029 0.823

Output index 0.029 1.325 0.034 * 1.877

Labor index 0.035 0.826 -0.032 -0.739

Tech index 0.038 ** 2.154 0.035 ** 2.358

N Obs 900 900 900 900

Industrial Fixed Effects (CNAE-93) X X X X

Table 5: Conditional Estimation Number of Employees

N Obs 900 900 900 900

N Obs 900 900 900 900

4.2. REGRESSIONS BY INDUSTRY

Table 7: Conditional Sectorial Estimation Number of Employees

N Obs 45 330 120 390

Table 8: Conditional Sectorial Estimation Number of Firms

% Employees in the Core Municipality 0.044 2.379 0.048 2.597

% Population older than 60 years 0.851 1.973 0.887 2.097

Human Capital Index -0.160 * -2.855 -0.161 * -2.914

% Population without studies -0.612 -2.150 -0.613 -2.191

% Population second level eduaction -0.453 -2.200 -0.448 -2.219

Output index 0.048 * 4.526 0.053 * 5.023

% Firms in the Core Municipality -0.032 -2.236 -0.029 -2.078

Human Capital Index -0.170 * -3.626 -0.169 * -3.680

Entrepreneurial Culture Index 0.054 2.490 0.058 * 2.740

Chinitz index 0.089 * 3.072 0.073 * 2.594

Tech index 0.038 2.154 0.035 2.358