AGGLOMERATION ECONOMIES IN ANTIOQUIA: AN EMPIRICAL

ANALYSIS TO TEST SOME CONCEPTS OF NEW ECONOMIC

GEOGRAPHY

Carlos Carvajal

School of Economics, Management and International Business

Universidad Pontificia Bolivariana

Abstract:

This paper aims to determine the existence of an underlying spatial pattern in the regional
distribution of income in the municipalities of Antioquia and if, moreover, the pattern corresponds
to the statement by the principles of the New Economic Geography. To this end, first it is
estimated a spatial lag model relating the level of municipal incomes with the amount of
collections from property taxes to perform statistical inference on the spatial distribution of
municipal economic agglomerations; secondly, by nonlinear least squares, an equation of market
potential is estimated to determine the characteristics of spatial pattern in the light of the new
economic geography. The methodological procedure is supported on the proposed Helpman-
Hanson (1998) theoretical model, from which it is intended to estimate its central hypothesis. The
results confirm the central hypothesis for Antioquia department economy.

Keywords: Agglomeration Economies, New Economic Geography, Market Potential, Spatial

Econometrics.

Special thanks to Guillermo David Hincapié Vélez, professor at the Universidad Pontificia
Bolivariana, who contributed immensely in the direction of this work, their contributions,
suggestions and recommendations were essential in this effort to address the Antioquian space
economy, also thanks to Carlos Londoño Yepes who was responsible for challenge me to address
the regional economy in their subjects since 2013.
 I. NTRODUCTION

Up to ten thousand years ago humanity was devoted to hunting and gathering, this consisted of
livelihoods based on extraction of necessary goods for the survival of the tribes in the Neolithic.
This type of livelihood lacked sustainability through time since the passing of the years the dams
ran out and gathering decreased forcing hunter-gatherers to travel greater distances to survive. At
that point the tribe was forced to migrate to other more virgin territory where they could settle
again for some time. This nomadism met his end when agriculture and animal domestication took
place, giving rise to the production of food and goods needed for subsistence. Thus it was possible
to settle permanently in some locations giving way to the origin of the city and civilization;
production freed us from hunting and gathering and allowed unprecedented population growth
until then.

The phenomenon of the city, that is itself, agglomeration of human beings with their respective
activities has relied heavily on geography, because this determines the conditions under which
they should develop production. In the contemporary world, the development of productive
capacities of mankind has achieved dynamics increasingly complex to explain; even human
relationships start to become factors of that environment beyond the purely physical
characteristics that have traditionally been treated as the main causes of prosperity. Emerge,
many questions about the phenomena of agglomeration in the territory, as this is not uniform in
the various explanations enters an interesting amount of variables that exert complex forces.

The space is the stage where these forces take place and evolve through time in complicated
ways. Due largely to the evolution of the production means, technologies and techniques to
overcome the space for human survival. That is why it is important economic geography, which
has developed different theories and methodological approaches to give explanation about
economic activity in space, and as acceptance of the influence of space in economic phenomena is
crucial to a proper understanding of them. From Krugman approaches, economic geography has
developed different theories to explain the economic forces that mediate the formation of
agglomeration processes; has been established in the debate, for example, the hypothesis that
there are few explicit economic benefits which human beings, firms and consumers tend to
agglomerate in space, but these benefits may be limited by the possibility that the agglomerations,
due to greater competitiveness, generating higher prices. This is commonly known as the
configuration of centripetal and centrifugal forces.

Many researchers have been interested in the empirical test of the existence of these forces that
eventually mediate the configuration of a spatial pattern of economic activity. Helpman
theoretical approaches and Hanson (1998), for example, have commonly been as reference in the
international debate for purposes of empirical testing. Their model, which is based on a rethinking
of Krugman (1991), aims to study the characteristics of a spatial pattern of agglomerations,
characterized by the direct relationship between the level of economic activity and prices, both
mobile factors such as motionless; that is, and in a basic way, economic agglomerations can
generate wage levels or higher income, but the competition for land may lead to an increase in
prices, imposing the need to economic agents to adjust their expectations.
The main objective of this article is to contrast empirically the central hypothesis of the theoretical
model of Helpman-Hanson (1998), to study the economic agglomerations of the municipalities of
Antioquia in 2011. To carry out this goal, first techniques on Exploratory Analysis and Confirmatory
spatial data was used in order to determine an initial statistical sense, the existence of a spatial
pattern of income of the municipalities of Antioquia and what are their significant factors
according the established by Helpman and Hanson (1998). Second, a model by nonlinear least
squares of market potential function was estimated to justify the approach to the characteristics
of the spatial pattern of income in the department.

The findings of this research find their justification and relevance in the following aspects: first,
serve to characterize the subregional economic agglomerations of the department in terms of
important economic variables; Second, although the work is not intended to estimate the
structural parameters of the equation of Helpman-Hanson (Equation 18), it sets a way to approach
them from spatial econometrics techniques. On the other hand, the information provided by the
article will be especially useful in the subregional economic and competitive planning of the
department of Antioquia.

The paper is organized into 8 sections of which the first is this introduction. In the second and
third, a review on the development of economic geography and a reference to the state of the
literature on the subject in the country it is made. In the fourth, the theoretical model that frames
the study is exposed. In the fifth the methodology is exposed, the sixth presents the most
remarkable econometric issues based on international literature, the seventh analyzes the results
and finally in the eighth the conclusions.

II. ECONOMIC GEOGRAPHY

The spatial economy or economic geography studies the patterns of location of economic
activities in the territory, it has developed two streams to adress this phenomenon (Velez, 2008);
Differences of "first nature" or predestination and differences of "second nature" or self-
organization.

Differences of "first nature" or predestination explain the agglomerations of certain economic

activities from physical peculiarities of the territory; as water sources, topography, geology,
minerals disponibility, geostationary location, availability of resources, access to markets, soil
fertility, etc. This approach stems from the theoretical developments made during the nineteenth
and mid-twentieth century from mainly three economic trends. For a better understanding of
these flows and their temporary location see Figura 1 of Appendix.

However, these approaches are deficient addressing certain situations where similar sets of
variables are presented but with quite different results, for example same geography and
institutions but very different levels of economic activities. It is also deficient in explaining the
economic activities of large cities. Paul Krugman (1991) identified the major weakness in the
theories and models of the three approaches; because they contain many ad hoc assumptions,
besides that in the vast majority of cases of analysis are partial equilibria that do not incorporate
market structures or pricing mechanisms.
Krugman had developed a strong argument for international trade in 1979 and noted that these
could be applied to address the problems of economic geography highlighting the differences of
second nature.

Differences of "second nature" or self-organization emphasizes concepts such as transportation

costs or "space friction" economies of scale, mobility of factors and externalities derived from
economic activities and the territory to explain the distribution of populations and economic
activities in geography, however this current lacked a solid methodological foundation. In 1977
from the work of Dixie-stiglitz, it could have the mathematical tools that would allow addressing
models with increasing returns and imperfect market structures. Under this approach comes the
"new economic geography" from work done by Paul Krugman in 1991 in addressing market
structures and mechanisms pricing, posture academic which was synthesized in later works by
Fujita, Krugman and Venables (2000).

All of this in order to achieve a more reliable methodological apparatus to make predictions in
terms of economic policy, incorporating some insights from the three initial currents in a rigorous
formal microfounded framework.

All previous models reach consensus on that the distribution of economic activities is due to the
equilibrium of two forces: agglomeration forces or "Centripetal" and scatter or "centrifugal". The
former tend to agglomerate or concentrate economic activity and the latter tend to disperse or
decentralize it.

III. AGGLOMERATION ECONOMIES IN COLOMBIA AND ANTIOQUIA: PREVIOUS WORKS

The use of methodological tools in the analysis of activity in the geographical area for Antioquia
has been rather little, this due mainly to the lack of data for the respective analysis in the
department, this being the main problem for research production. Such analysis requires
consideration of several variables through time, i.e. the time series needed to analyze the
evolution of agglomeration processes more accurately.

However at national level the picture is much richer in bibliographic production, because it has
more consolidated information, while at regionally level disaggregated data is required and often
must to be estimated or use proxies.

From the 90s it became important economic geography in Latin America, but mostly from a more
theoretical than practical reasons for the above viewed.

For the Colombian economy they have made some spatial analyzes such as Jaime Bonet (2008) of
the Regional Economic Studies Center, CEER (for its acronym in Spanish) of the Bank of La
republica of Cartagena, where elements of economic geography and spatial analysis are
incorporated to the Colombian case. Recent work also consists of an analysis from spatial statistics
on regional disparities in the country (Quintero Loaiza, Moncada Mesa, Galeano Duque, Vélez
Villegas, & Restrepo Estrada, 2015). About agglomeration economies, the work of Juan Esteban
Vélez (2008) "Agglomeration processes in Colombia in the light of the New Economic Geography"
in which these processes are analyzed using data for Colombian municipalities, were the structural
parameters of the model developed by Helpman (1995) and Hanson (1998) are estimated, being
according to the author, the best alternative to contrast empirically the New Economic Geography
to the Colombian case.

In the Antioquian picture is the work of Yhelin Cristina Street, Lina Marcela Correa and María
Beatriz López (2008) "The Agglomeration in Antioquia (1968-2005): An Analysis of the spatial
dependence" in which is demonstrated the existence and persistence of agglomeration processes
in the Antioquian economy, determining the existence and persistence of a structure of centrality
in Medellin with high dependency effects in neighboring municipalities and the region of eastern
Antioquia, specifically Rionegro and incipient agglomeration in Urabá. At the level of Medellin (the
región capital) is the work of Galeano (2013) in which is done an analysis of the urban economy
between 2005-2010 through exploratory spatial data analysis (AEDE), in this work a
characterization of economic activity was done in detail through cluster analysis.

IV. RATIONALITY OF THE HELPMAN-HANSON (1998) MODEL

The seminal model of the New Economic Geography of greater importance was developed by Paul
Krugman (1991), in this model the appearance of spatial agglomerations is explained as a
dependent phenomenon of relations between increasing scale returns at the enterprise level,
transportation costs and the mobility of production factors. In this model it is considered a
composite of two regions economy with two economic sectors: agriculture and manufacturing,
with the work as only input factor for production. The model to check in this work is a modification
of the model of Krugman, because it presents problems to produce a "monocentric" equilibrium in
determining the wage structure in the long term, in addition to a trend in prices of non-tradable
goods 1 to be higher in agglomerations that represents some relevant empirical problems (Vélez
Villegas, 2008), for that reason, a based on Helpman (1998) and Hanson (1998) model is used.

In the model the agricultural sector of the seminal Krugman (1991) model was replaced by the
services sector housing H which reflects a sector of non-tradable goods, correctin thus the two
aforementioned problems generating virtually identical results to the primordial model of the New
Economic Geography as Hanson (1998) explained.

The explanation of the model is based on Mion (2003), where an economy consisting of φ regions,
with two productive sectors and assuming one factor of production as input. The manufacturing or
industrial sector M, the housing services sector or non-tradable goods H, along with the
production factor labor L. The M sector produces a continuum of varieties of differentiated
products under increasing scale returns and work as only input for production. Each variety of
differentiated goods can be traded between different regions incuring in friction 2 or iceberg
transportation costs. Suppose two generic i and k regions, where per each unit produced in i that

1
In the Krugman model non-tradable goods are produced by the agricultural sector.
2
The term friction costs covers the range of the various costs necessary to overcome the distance between
regions; transportation costs, information costs, opportunity costs and any impediment that must to be
incurred to do business in other regions markets (Polèse & Rubiera, 2009). The literature usually treats these
costs as transportation costs for simplicity.
is shipped and reaches the región k, a fraction of product gets lost on the way, as melting in the
path to the delivery region, this involves costs dependent on distance. Being Pm, i mill price of a
variety in the region i and Ti,k are the transport costs between i and k determined by Ti,k=f(di,k)
where di,k d is the distance between the two regions, we have that the price of a product of a
variety made in i and traded in k is P*m,i = Pm,i/ Ti,k. We can see that the price through the distance
is an inverse function. H is a sector of homogeneous goods that can not be traded between
regions, the price of goods in the H sector, that is to say, PH,i can be different between different
regions and is determined by the equilibrium of market forces in each region. Labor is supposed to
be freely mobile with the total amount in the economy represented by L, the equilibrium of the
spatial distribution among workers-consumers is determined by real wages, i.e. nominal wages
(wi) and prevailing prices for each region. Li is the work in the region i with∑𝛷 𝑖=1 𝐿𝑖 = 𝐿 , the labor
participation rate of the region i in the economy is λi=Li/L, preferences and technologies are not
specified.

The consumer preferences are described by a standard Cobb-Douglas utility function with
constant elasticity of substitution CES.

𝑈 = 𝐶𝑀 µ 𝐶𝐻 1−µ 0 < µ < 1 (1)

CM is the consumption rate of varieties of M sector and CH is the H sector comsumtion. Assuming a
modern sector that provides a continuum of varieties of size N, the index of consumer CM is given
by:
1
𝑁
𝐶𝑀 = [∫0 𝐶𝑚 (𝑗)𝜌 𝑑𝑗 ]𝜌 0 < ρ < 1 (2)

Cm(j) represents the consumption of variety j between N varieties, therefore every consumer is
1
attracted by the variety and parameter 𝜀 ≡ 1−𝜌 varies between 0 and ∞, ε is the constant
elasticity of substitution between two varieties; the greater is ε more varieties will be sustituted
and when it approaches the unit, the consumption tends to diversify varieties. Y represents
consumer income, optimizing utility the demand for a variety j is derived as follows:

∗
𝑐𝑚 (𝑗) = 𝑝𝑚 (𝑗)−𝜀 µ𝑌(𝑃𝑀 )𝜀−1 𝑗 ∈ [0, 𝑁] (3)

PM is a differentiated products price index defined by:

𝑁
∗ (𝑗)−(𝜀−1)
𝑃𝑀 ≡ [∫0 𝑝𝑚 𝑑𝑗]−1/(𝜀−1) (4)
Technical and technological progress is assumed to be homogeneous in the economy, allowing to
describe the relationship between work required 𝑙(𝑗) and the production of variety j with the
equation:

𝑙(𝑗) = 𝑓 + β𝑞(𝑗) (5)

Where f is the fixed labor requirement, β is the requirement of marginal labor 3 and 𝑞(𝑗) are the
quantities produced of a variety j. Companies knowledgeable of consumer demand are assumed
with seeking to optimize their profits through price:

𝜋(𝑗) = 𝑝𝑚 (𝑗)𝑞(𝑗) − 𝑤[𝑓 + 𝛽𝑞(𝑗)] (6)

Where 𝜋(𝑗) represents the profits of the company producing the variety j, and w is the wage rate
paid to the required labor force 4. Solving first order conditions and using the constant elasticity of
substitution ε can rewrite equation (6) as a common equilibrium relationship:

𝛽𝑤 𝛽𝑤
𝑝𝑚 (𝑗) = 1 = (7)
1−� � 𝜌
𝜀

As it can be appreciated, βw reflects the marginal cost of variety j. Being this a scenario of
monopolistic competition with a continuum of new businesses and substitute varieties, prices
tend to fall as competition increases until get the equilibrium, where the benefits become zero by
price competitivity slowing the entry of new firms by loss of incentives. This creates equilibrium of
amounts and varieties given by:

𝑓(𝜀−1)
𝑞(𝑗) = 𝑞 = (8)
𝛽

In equilibrium, the quantity supplied by each firm is fixed, as equally the number of companies and
varieties regardless the location of production from these companies as same that demand for

3
The requirement f to be a fixed factor is not dependent on the quantities produced, while the β factor
depends on these, which certainly implies increasing scale returns because as production increases the
required amount of f is reduced by each unit produced.
4
Note that these wages are from the industrial or manufacturing sector. At the Mion (2004) work β marginal
labor requirement is matched to the unit.
labor will not be affected by the location. The total demand for a variety of region i is given by the
domestic demand for the region itself and other regions that incur in transportation costs to
import the variety. From this price index PM,I is obtained as a weighting between the prices of
industrial goods produced locally and industrial goods from other regions:

1 1
𝛽 𝛾𝐿
𝑃𝑀,𝑖 = ( )( )1−𝜀 (∑Φ
𝑘=1 𝜆𝑘 (𝑤𝑘 𝑇𝑖,𝑘 )
1−𝜀 1−𝜀
) (9)
𝜌 𝑓𝜀

The total demand for a specific range depends not only on prices but also on wages, both in the
region of production and other importing regions and friction costs:

𝛽 −𝜀 −𝜀
𝑞𝑖= 𝜇 � � (∑Φ 𝜀−1
𝑘=1 𝑌𝑘 �𝑤𝑖 𝑇𝑖,𝑘 � 𝑃𝑀,𝑘 ) (10)
𝜌

Equating equation (8) with equation (10) to clear 𝑤𝑖 the wage equation is obtained for the region i
given by:

1 1
𝜇 1−𝜀 𝜀−1 𝜀
𝑊𝑖 = 𝜌𝛽 −𝜌 ( )𝜀 (∑Φ
𝑘=1 𝑌𝑘 𝑇𝑖,𝑘 𝑃𝑀,𝑘 ) (11)
𝑓(𝜀−1)

This equation shows the ratio of wages (w) in the region i relative to the distance through the
friction costs (𝑇), the intensity of competition between varieties ( 𝑃𝑀 ) and the size of the
markets (𝑌). Another important factor is the expeditures in housing services of H sector denoted
by GH,i we have for each region of the economy:

𝐺𝐻,𝑖 ≡ 𝑝𝐻,𝑖 𝐻𝑖 = 𝑌𝑖 (1 − 𝜇) (12)

Where 𝑝𝐻,𝑖 are the prices of housing services or non-tradable goods in i, and 𝐻𝑖 represents the
quantity of these goods and services. If we assume that people have the same amount of goods
and services H, with the assumption that industrial profits become zero in equilibrium, then the
value of manufacturing output in each region equaled the returns factor (𝜆𝑖 𝑤𝑖 𝐿) then it can be
expressed the income of region i as:

1−𝜇 Φ
𝑌𝑖 = �𝜆𝑖 ∑𝑘=1 𝜆𝑘 𝑤𝑘 𝐿� + 𝜆𝑖 𝑤𝑖 𝐿 (13)
𝜇
There will be no incentive to labor migration, if the spatial distribution of workers is balanced,
assuming perfect labor mobility will generate the equalization of real wages in the long run
through the equilibrium given by:

𝑤𝑖 𝑤𝑘
𝑃𝑀,𝑖 𝜇 𝑃𝐻,𝑖 1−𝜇
=𝑃 𝜇𝑃 1−𝜇 ∀𝑖, 𝑘 = 1,2, … , Φ (14)
𝑀,𝑘 𝐻,𝑘

From equations (12), (13) and (14) we can conclude that in regions where agglomeration is
generated, large revenues are also generated, this creates centripetal forces that attract workers
to the agglomeration, which in turn creates increased demand for goods, especially non-tradable,
increasing the price of these expenses due to the arrival of new labor from other less
agglomerated regions. This generates as a result some centrifugal forces that tend to spread
economic activity between regions because the high prices in housing services become a barrier to
migration of labor. This implies then that companies should raise wages to offset higher prices of
agglomeration 5 and thus create incentives for migration of labor. Some normalizations 6 can be
expressed without loss of generality, the equilibria of equations (9), (11) and (13) respectively as:

1−𝜀 1
𝑃𝑀,𝑖 = (∑Φ
𝑘=1 𝜆𝑘 �𝑤𝑘 𝑇𝑖,𝑘 � )1−𝜀 (15)

1
1−𝜀 𝜀−1 𝜀
𝑊𝑖 = (∑Φ
𝑘=1 𝑌𝑘 𝑇𝑖,𝑘 𝑃𝑀,𝑘 ) (16)

𝑌𝑖 = 𝜆𝑖 𝑤𝑖 𝐿 (17)

Substituting equations (15) and (17) (16), applying logarithms and simplifying the equation, the
Helpman-Hanson model is finally obtained:

1−𝜀(1−𝜇) (1−𝜇)(𝜀−1) (𝜀−𝜇)

𝐿𝑛(𝑤𝑖 ) = 𝑘 + 𝜀 −1
𝐿𝑛(∑Φ
𝑘=1 𝑌𝑘
𝜇
𝐻𝑘 𝜇
𝑤𝑘 𝜇
𝑇 𝑑𝑖,𝑘 (1−𝜀) ) (18)

5
As noted the equilibrium equation (14) relates the nominal wages and market prices, the ratio shows a
balance between real wages.
6
These normalizations are available in Chapter 4 of The Spatial Economy (Fujita, Krugman & Venables, The
Spatial Economy: Cities, Regions, and International Trade, 1999), in paragraph 4.4 pages 54 and 55.
K is a parameter that reflects the behavior of preference for industrial goods (μ), the elasticity of
substitution (ε) and the friction cost function(𝑇).

The ε expression 𝜀(1 − 𝜇) indicates which forces are influencing the agglomeration, if this is less
than unity means that wages in the region i behave as an increasing function that depends on
nearby markets, that is to say that centripetal forces are acting and tends to agglomerate. When
the ε expression 𝜀(1 − 𝜇) is greater than unity then the centrifugal forces prevail over the
centripetal generating greater spread of economic activities in the territory 7.
𝜀
The expression indicates the power to generate economies of scale, if this is greater than 1
(𝜀−1)
means that there is potential to generate profits due to the existence of scale economies.

The expression (18) is a nonlinear equation, which complicates the implementation empirical
model as expressed Mion (2003). To this can be approximated a linearized version 8 such as
proposed Mion performing the following linearization 9 of the equation (18) resulting in:

𝐿𝑛(𝑤𝑖 ) = 𝛼 + ∑Φ � �
𝑘=1[(𝛽1 𝑌𝑘,𝑡 + 𝛽2 𝐻𝑘,𝑡 + 𝛽3 𝑤
−1
�𝑘,𝑡 )𝑇𝑖,𝑘 ] + 𝐸𝑖,𝑡 (19)

In this equation each β is equivalent to:

1−𝜀(1−𝜇) (1−𝜇)(𝜀−1) (𝜀−𝜇)

𝛽1 = 𝜃 , 𝛽2 = 𝜃 , 𝛽3 = 𝜃
𝜀𝜇 𝜀𝜇 𝜀𝜇

While each variable is given by:

𝑌𝑘,𝑡 𝐻𝑘,𝑡 𝑤𝑘,𝑡

𝑌�𝑘,𝑡 = ∑Φ
�𝑘,𝑡 =
, 𝐻 ∑Φ
, 𝑤
� 𝑘,𝑡 = ∑Φ
𝑘=1 𝑌𝑘,𝑡 𝑘=1 𝐻𝑘,𝑡 𝑘=1 𝑤𝑘,𝑡

V. METHODOLOGY

7
In the model of Krugman (1991) a small ε and a large μ makes the expression 𝜀(1 − 𝜇) < 1 generates a
full agglomeration region, and in the opposite case, a perfect dispersion, which obviously does not match
𝜀−1
with the observable reality, to solve this problem the condition of "no black hole" consisting of = 𝜌 > 𝜇,
𝜀
for a more complete explanation of this condition see chapter 4, Section 4.6 of The Spatial Economy (Fujita,
Krugman, and Venables, The Spatial Economy: Cities, Regions, and International Trade, 1999) pg. 58 and 59.
8
In NEG this kind of approach are not new to applied models and have yieled promising results from its
beginnings as seen in works like Combes and Lafurcade (2001).
9
To see the formal linearization of equation (18) See Appendix 1 Mion (2003). While this is not relevant for
this work has been shown as a reference for possible future work.
Through spatial data analysis pretends to contrast the model assumptions outlined in equation
(18), for this purpose an exploratory analysis is employed, other confirmatory analysis and a
regression equation of market potential. The tools used in this paper are fully explained in much of
the literature on spatial econometrics and spatial data analysis, so that rigorous conceptualization
of these tools is not addressed in this paper.

In the exploratory analyzes the behavior of each of the variables through Kernel densities and local
Moran indicator is set.

Kernel density is nonparametric for formulating density functions approach, in this not so rigid
data assumptions are imposed as in the parametric approach. As pointed out by Silverman (1986),
"assuming that the data distribution has a probability density f, they are possible" speak "more for
themselves in determining the function f, if its adaptation were required to a particular family of
parametric functions. This allows, that from a cloud of points a scheme localized density data
representing the overall trend or pattern << trend or overall pattern >> from their distribution is
obtained, where there is more concentration, less where, how gradient occurs the spatial
variation, etc.” 10

The Moran test allows appreciate the existence of spatial autocorrelation to determine whether
the spatial units are randomly distributed or respond to certain patterns 11. This allows better
understanding of the spread of a phenomenon or trend in space and time.

For confirmatory analysis Quantile map is used for the collections of industry and commerce taxes,
map-based cluster LISA test, spatial lag model, spatial error model and finally a map of bivariate
cluster.

The Quantile map allows appreciate the geographical distribution of the data coloring the the
observations of the same quartilewith same colors, the LISA test (Local Indicators of Spatial
Association) estimates a local statistical value for each observation, allowing to determine the
relevance of spatial clusters of Similar values around the observation in question 12. The bivariate
cluster map or bivariate LISA indicates the level of association between observations at a given
time compared to the average of the other observations at a different time, allowing analyzing the
persistence of spatial clustering in time 13. The model of spatial lag assesses the existence and
strength of spatial relationships, while the spatial error model is linked to the dependence of

10
To address the technical conceptualization about Kernel density see Silverman (1986) at “Density
estimation for statistics and data analysis.” A more abbreviated source in Spanish is able in “Modelización
Cartográfica mediante estimadores Kernel” (Jimenez Moreno 1991).
11
This concept is widely developed in "The Moran Scatterplot as an ESDA Tool to Assess Local Instability in
Spatial Association" (Anselin, The Moran Scatterplot as an ESDA Tool to Assess Local Instability in Spatial
Association, 1996)
12
This statistic represents the contribution of each observation in the global spatial autocorrelation. The
sum of the LISA statistics of all observations is proportional to the global Moran.
13
A thorough analysis of the LISA test is in "Local Indicators of Spatial Associations - LISA" (Anselin, 1995)
errors, and this is useful when there are potentially influential biases in spatial autocorrelation due
to spatial data. 14

Regularly it has been established in the literature the difficulties of conducting empirical exercises
on theoretical approaches proposed by Krugman (1991), and equation (18) Helpman-Hanson has
the fundamental purpose of facilitating it although remains complex, reason why it is not
estimated in this work, but an equation of market potential is estimated.

The market potential (Harris, 1954) fuction measures the potential of a region 𝑟 as the weighted
sum of purchasing power from other regions 𝑠, the distance factor behaves inversely to
purchasing power because the territory carries costs, thus a function of market potential widely
used as obtained:
1
𝑀𝑅 = ∑𝑆 𝑃 (20)
𝐷𝑟𝑠 𝑆

Where 𝐷𝑟𝑠 is the distance between the región 𝑟 and 𝑠, while 𝑃𝑠 is the purchasing power of the
region s. To test it is carried out an estimate by nonlinear least squares of a market potential
function that takes into account the approach of the New Economic Geography. Thus, equation
(21) is estimated as a strategy used to contrast the Helpman-Hampson model like Brakman et.al
(2002) sugests.

𝑙𝑜𝑔(𝑤𝑖 ) = 𝑘1 𝑙𝑜𝑔[ ∑𝑠 𝑦𝑟 𝑒 −𝑘2 𝐷𝑟,𝑠 ] + 𝐸𝑡 (21)

Where the term 𝑤𝑟 represents the level of collection of property tax in the municipality "r", the
term 𝑦𝑟 represents the income level of the municipality, measured here as the level of tax in
respect of industry and commerce. Meanwhile, 𝐷𝑟,𝑠 represents the distance from the municipality
"r" to "s", derived from the distance matrix from which the matrix of spatial weights is calculated.
𝑘1 is a sensitive parameter to trade off between the costs of non-tradable goods and income while
𝑘2 reflects the impact of friction costs. Equation (18) is quite similar to an equation of market
potential as explained Mion (2003).

VI. DATA AND PROCEDURES

One of the biggest problems in trying to empirically test economic theories in Colombia is the poor
quality of the information available, since there is not a good culture of data, much of this is due to
the high costs of compiling the information plus various economic dynamics that sometimes are
immeasurable, such as illegal activities, informality and lack of institutional consolidation afflicting
the country. In addition for empirical microfunded models purpose it is quite common to have
trouble choosing data that serve as good proxies. This paper analyzes the smallest unit available in
Colombia, i.e. the data were taken at municipal level for each of the 125 municipalities of
Antioquia. Theoretically, a small unit of analysis enables a better reflecting of externalities that the
model seeks to explain, and avoid that errors be spatially correlated as explained Velez (2008).

14
The concepts of lag and error spatial model can be found widely described in "Chapter 14: Spatial
Econometrics" Luc Anselin (2001) of "A Companion to Theoretical Econometrics"
For the variable Y the average tax collections took per capita for each municipality, taking the tax
collection and dividing by the number of inhabitants in each municipality. This variable represents
the demand for goods, most ideal would be to take the GDP of each municipality instead tax
collection, but unfortunately in Colombia that information is only able at the departmental and
large cities level, this strategy is based on the rationality of that tax revenue is proportional to
GDP. For the H variable, which represents the non-tradable goods consumption tax collection was
taken by average per capita property tax concept as this reflects to some extent the consumption
of housing services. For the w variable, reflecting the consumption of tradable goods average per
capita collection it was made by way of trade and industry under similar to those of the previous
two variables strategy. As Velez (2008) clarifies, this proxy is not free of problems as it also
includes the commercial activity which may itself include any consumption of non-tradable,
although its rationale underlying that a high proportion of collection of a municipality for industry
and trade shows signs of industrial activity, or at least tradable goods commerce. All tax
information for previous proxies were obtained from the report of budget execution of the
Departamento Nacional de Planeación for 2011, while data on the number of inhabitants were
taken from census projections 2005 to 2011 of the Departamento Administrativo Nacional de
Estadística.

As is usual in the application of statistical techniques and spatial econometrics, it is important the
specification of the spatial weights matrix. For purposes of this exercise, a matrix of road distances
between municipalities is based on information provided by the secretaría de infraestructura de la
Gobernación de Antioquia. It is also important to note that the various tests, such as global
MORAN local and the spatial lag model, were made under different spatial weights matrices, and
corroborating the models that best fit reported according to the different matrices (Queen matrix
type or Rook spatial weights type), however, the distance matrix in travel times was the most
consistent.

Another important aspect to the estimation is undertaken here, and that is commonly reported in
the literature, it has to do with the potential endogeneity problem by simultaneity, or
measurement problems in the variables. In this aspect, it is started from the arguments of Le Gallo
(2004) regarding the possibility of spatial lag models, can help to solve these problems. In this
regard, these authors point out that:

“Furthermore, data on some of these explanatory variables may not be easily

accessible and/or reliable. Spatial autocorrelation may therefore act as a proxy to all these
omitted variables and catch their effects. This is particularly useful in the case of European
data, where explanatory variables are scarce... ”
 VII. RESULTS ANALYSIS
As was stated in the previous section, the study was based primarily on the development
of three major phases, namely on analysis using statistical techniques and spatial
econometrics of the spatial pattern of regional distribution of income in the municipalities
of Antioquia Finally, the results with the estimate by nonlinear least squares equation of
market potential for Antioquia are compared.
A. Exploratory Spatial Data Analysis on Regional Income Distribution in the
municipalities of Antioquia: 2011.
The aim of this section is establish, by statistical techniques the spatial pattern of income
distribution for the municipalities of Antioquia department. While the goal of the equation
Helpman-Hanson (1998) is precisely to model a spatial pattern of income in the light of
the new economic geography, the objective of this course is justified in certain mesure, as
stated Mion (2004), exploratory spatial data analysis serves to support further - generally
by variables Dummies referred to the existence of spatial clusters or specific
differentiation between rural and urban areas, to empirical tests that may be done about
the equation of Hanson (1998). This paper is primarily part of the exploratory and
confirmatory analysis of spatial data.
As a first step, it is necessary to clarify the distributional characteristics of the variables
under analysis. The graph (1) illustrates Kernel densities of variables with special emphasis
on their average behavior.
Figure (1). Kernel density of Industry and Commerce Collection per capita

Source: Own calculations based on DANE and DNP budget executions

As can be seen, the distribution of Kernel reflects a significant feature of the distribution
of economic activity in the department of Antioquia, showing the characteristics of
regional inequality.
The graph (2), meanwhile, the behavior of the collection by way of property tax, which to
some extent we approach the analysis of the behavior of housing prices, information
necessary to contrast the basic assumptions of the model proposed by Helpman-Hanson
(1998). Again, the behavior of inequality of municipalities in the department for that year
is highlighted. Light fashions at right of distribution, (with a selected bandwidth through
multimodality approach suggested by Silverman (1981)).
Figure (2). Kernel density of collection of property tax per capita

Source: Own calculations based on DANE and DNP budget executions

The graph (3), presents the case of variable tax revenue per inhabitant. In line with the previous
graphs, this behavior still has significant regional inequality. The intuition underlying approaches
Hanson (1998), it lies in the fact that economic agglomerations are configured under the trade-
off 15 between higher income probability by scale economies that can ease and centrifugal forces
reflected by high land prices, housing or nontradable as is considered. While the above graphs are
basic statistical information, their analysis allows approximating indications of this phenomenon,
and finding that inequalities manifest a regional correlation evidenced. By applying econometric
techniques, this paper advances in the realization of these hypotheses.
As mentioned, the theoretical approaches considered here as a reference, start from the
consideration of the acceptance of space as an important determinant of economic phenomena

15
The term trade-off relates to offsetting effects between variables for balance. A more detailed analysis of
this concept can be found in Chapter 4 of Fujita and Thisse (2013).
and precisely the equation Hanson aims to show the existence of a spatial pattern. Thus once they
are i considered the findings supported by analysis of the distribution of the variables to be
analyzed is passed to show whether the variables in question have spatial dependence by
techniques framed in the exploratory analysis of spatial data.

Figure (3). Kernel density of Recaudo Tax Per Capita

Source: Own calculations based on DANE and DNP budget executions

In order to verify the hypothesis of spatial dependence, it carried out the implementation
of local Moran test on analysis variables16. Table (1), the results of Moran scatterplot of
these variables are presented, see Annexes to find the p-value associated with hypothesis
testing. As can be seen clearly, the behavior of the scatter diagrams suggest definitively
the existence of positive spatial autocorrelation, indicating that can statistically be
concluded that, for each of the variables, the behavior of the municipality is conditioned
by the average behavior of their neighborhood, i.e. the level of tax collection industry and
commerce of a municipality depends, in general, on the average of the revenues of
industry and commerce of the neighboring municipalities. Statistical then allow to
conclude that there is an underlying spatial pattern in the variables under analysis, the
next step being the econometric modeling of the pattern, such as compliance with the
proposed objectives.

16
In the annexes, can be found the results of the application of local MORAN test for the different variables
under analysis and cluster maps.
 Table (1). Moran scatterplots

Recaudo de Industria y Comercio por habitante Recaudo de Impuesto Predial por habitante

Recaudo Tributario por Habitante

Source: Own calculations based on DANE and DNP budget executions

 B. Confirmatory Analysis of Spatial Data
Once it evidenced the existence of an underlying spatial pattern in all the variables under
analysis, it can continue, in order to statistically model the pattern, especially in the
dependent variable equation suggested by Hanson (1998).
On the map (1), the spatial distribution of the dependent variable of the model is
presented. As can be seen, the level of collection of industry and trade per capita tends to
be higher in the center of the department, in the municipalities of the Aburrá Valley and
the nearby municipalities, such as those located in the subregion of eastern Antioquia;
also in the north subregion as some municipalities, Entrerrios, Santa Rosa de Osos, among
others. Meanwhile, some municipalities in the subregion of Urabá like the municipality of
Necocli, San Pedro de Uraba stand out, among others, which have a significant level of
collection of industry and commerce average per capita.
Map (1). Quantile map for Industry and Commerce Collection Per capita

Source: Own calculations based on DANE and DNP budget executions

On the map (2), meanwhile, the LISA cluster map for collection of industry and commerce
per capita is presented. As can be seen, in terms of this variable, sub-regional disparities
are significant. Indeed, in blue are the municipalities that belong to the low-low cluster,
which are characterized by a low level of collection and whose neighbors also have a low
level of collection by way of trade and industry; This cluster includes many of the
municipalities in southwestern Antioquia and Urabá subregion. Consequently, it can be
concluded that the spatial distribution of these clusters is characterized by a similar
distribution center-periphery. This type of spatial pattern, then, must be approximated
with the confirmatory analysis of spatial data, which comprises spatial error models and
spatial lag model. To make explicit, in addition, the influence of high-high in modeling,
cluster dummy variable that contains the municipalities that make up this cluster was
formed.
Table (2), the results of applying the confirmatory analysis of spatial data, specifically
considering a spatial lag model is presented. According to the information criteria for
choosing between models, especially the AIC-Akaike information criterion criterion-, the
model that minimizes this criterion proved to be the spatial lag.
Map (2). LISA map for Industry and Commerce Recaudo Per Capita

Source: Own calculations based on DANE and DNP budget executions

Table (2). Results of spatial lag model to explain the spatial distribution

Variable Coeficient P-valor

Coeficiente 0.7827073 0.00234

H 6.073659e-06 0.0003

Y 2.022911 0.0000

LAG 0.5638548 0.0000

 Jarque-Bera (Normalidad) 0.29272

Test de Breusch-Pagan 0.58351

(Heteroscedasticidad)

Test de Moran sobre los 0.11955

Residuales (dependencia
espacial residual)

Source: Own calculations based on DANE and DNP budget executions

According to the results by the model of spatial lag, can be concluded the following
aspects: first, judging by the sign of the coefficient of the variable H concerning the
amount of property tax collection, it can be concluded that the central hypothesis is
confirmed that supports the intuition of Helpman-Hanson model, namely, that in general
the municipalities experiencing higher incomes have centrifugal forces characterized by
the increased value of the immovable property, in this case for the value of the property
given by the underlying competition for land. On the other hand, the level of economic
activity reflected by the variable "Y" is a statistically significant in explaining regional
differences in the amount of collection of industry and commerce.
Once estimated a spatial lag model, it can also concluded an important fact about the
possibilities of agglomerations and potential contagion effects of economic activity at the
municipal level, usually considered in the literature as spatial spillovers effects: the
coefficient of the spatial lag for the dependent variable is statistically significant to
explaining the spatial pattern; therefore, it can be concluded that the level of collection of
industry and commerce of a municipality depends directly on the amount of collection of
neighboring municipalities. This conclusion indicates the existence of a local spatial
spillover effect in the collection of industry and commerce for the Antioquia department.
As can be judged in the table (2), the model meets the necessary diagnostic concerning
the normality of errors, homoskedasticity of them and also the Moran test residual
indicates that the model fully explains the spatial pattern of the dependent variable
analyzed.
An interesting aspect in relation to the spatial pattern found, involves the application of a
bivariate MORAN test to know concretely the characteristics of spatial correlation
between the level of collection property tax and economic activity of the department's
municipalities. The map (3) corresponds to bivariate LISA cluster map on these two
variables. As can be seen, the municipalities in red make up the high-high cluster, which
means that the municipalities that have a high level of collection and property tax are
accompanied by neighbors who have a high level of income or economic activity; these
municipalities related mainly on the municipalities of the Aburrá Valley and nearby
municipalities in the subregion of the Near East. On the other hand, the low-low cluster in
blue corresponds to the municipalities that have on the contrary a low level of collection
and property tax also are immersed in a neighborhood of low economic activity; This
cluster is made up of municipalities of Southwest and North subregions, fundamentally in
the municipality of Caucasia in the Bajo Cauca, which represents a cluster itself and the
municipality of Arboletes in the Urabá subregion.
Finally, the high-low cluster, in pink, which consists of municipalities that have a high level
of property tax collection, and in turn are accompanied by a neighborhood of low level of
economic activity, is presented. As can be seen, these municipalities belong more to the
south of the Urabá subregion, and the municipality of Santa Fe de Antioquia in the West
subregion. It can be concluded from the subregional distribution of these clusters,
especially those belonging to the cluster of low collection (high-low cluster), That
especially the Urabá subregion does not adaps to the spatial pattern suggested by the
central hypothesis of Helpman-Hanson model and could be expected, as a policy
recommendation that an economic improvement in the municipalities of sub nearby
regions will eventually result in an economic improvement in the subregion; this
conclusion being justified in the statistically significant coefficient of the spatial lag model,
which denotes a spillover effect of the underlying municipal economic activity.
Map (3). Map of bivariate cluster for property tax collection

Source: Own calculations based on DANE and DNP budget executions

B. Estimation of the central hypothesis Helpman-Hanson model from market potential.

Table (3) shows the results of estimation by nonlinear least squares structural parameters
𝑘1 and 𝑘2 of equation (21).

Table (3). Least Squares Estimation of Nonlinear Market Potential Ecuation

Structural parameter Coeficient p-valor

𝑘1 1.438e-03 1.77e-14

𝑘2 9.581e+01 2e-16

Number of iterations to
convergence: 8
Achieved convergence
tolerance: 6.248e-09

Source: Own calculations based on DANE and DNP budget executions

According to the results, the version of market potential equation Helpman-Hanson
(1998), proposed by Brakman et.al (2002), allows us to conclude two important facts: first,
similar to what had thrown the spatial lag model, the hypothesis that economic
agglomerations impose a cost, a centrifugal force, reflected in the price of housing or, in
our case, the level of municipal property tax collection is confirmed; This fact confirms the
positive character of 𝑘1 parameter equation (20). Second, the 𝑘2 parameter of the
equation allows reflect the average weight of the distance in the calculation of market
potential, i.e. the relative weight of the impedance of the agglomeration at the
department.
VIII. CONCLUSIONS
The combination of factors such as scale economies, market imperfections, friction or
transportation costs, and endowments, have enabled to the New Economic Geography
develops a series of models of general equilibrium highly specified, able to address the
phenomenon of economic agglomeration strictly for determine economic activity.
However it has been widely recognized by literature the difficulty of empirically assess of
these microfounded models.
From the exploratory analysis of the variables it can be concluded that the behavior of
municipalities is conditioned by the behavior of your average neighborhood. The statistics
indicate that there is an underlying spatial pattern in the variables under analysis.
Confirmatory analysis of spatial data using basic estimation models of spatial
econometrics by using a spatial lag model, can conclude that the central assumptions
behind the model of Helpman-Hanson is confirmed, in general the municipalities that
experience higher level of income have centrifugal forces characterized by the increased
value of consumer immobile factors, in this case for the value of the property, given the
demand for land in agglomerations. On the other hand, the level of economic activity
reflected in the amount of collection of industry and tcommerce is a statistically significant
determinant on explaining regional disparities in income and patterns according spillover
effects, which is in good agreement with the spirit of the New Economic Geography. From
the estimated market potential hypothesis it is confirmed according to which economic
agglomerations impose a cost, a centrifugal force, reflected in the price of housing in
addition to confirming the significance of the costs of spatial friction by average weight of
distance. This gives us clues from an empirical perspective on the equilibria suggested by
the NEG and as, somehow the spatial distribution of economic activities in Antioquia are
the result of the trade off of these equilibria. Finally the results were consistent with the
main assumptions of the model Helpman-Hanson.
Subsequent investigations could estimate the model Helpman-Hanson for Antioquia,
hopefully from panel data estimation to observe the evolution of the structural variables
of agglomeration, and incorporate the effect of some local shocks and their potential
impacts on the dynamics of Antioquian space economy.

REFERENCES

Acevedo Bohórquez, I., & Velásquez Ceballos, E. (2008). Algunos conceptos de la econometría
espacial y el análisis exploratorio de datos espaciales. Ecos de economía(27), 9-34.

Anselin, L. (1995). Local Indicators of Spatial Associations - LISA. Geographical Analysis, 27(2), 93-
115.

Anselin, L. (1996). The Moran Scatterplot as an ESDA Tool to Assess Local Instability in Spatial
Association. Spatial Analytical Perspectives on GIS, 111-125.

Anselin, L. (2001). A Companion to Theoretical Econometrics. (B. H. BALTAGI, Ed.) Malden,

Massachusetts , USA: Blackwell Publishing Ltda.

Aroca, P. (2000). ECONOMETRÍA ESPACIAL: Una Herramienta Para el Análisis de la Economía

Regional. IDEAR.

Belmar, C. (03 de Abril de 2014). CIUIDADES COMPETITIVAS. Recuperado el 29 de 10 de 2015, de

LA GEOGRAFIA ECONOMICA Y ECONOMIAS DE AGLOMERACION: UNA MIRADA A LA
 LITERATURA:
http://www.ciudadescompetitivas.org/page54.html#ralacionaglomeraciontamanodelasciu
dades

Boudeville, J. (1970). Les Espaces Économiques. París .

Brakman, S., Garretsen, H., & Schramm, M. (January de 2002). The Spatial Distribution od Wages
And Employment: Testing the Helpman-Hanson model for germany. 29.

Calle, Correa, & López. (2008). La Aglomeración en Antioquia (1968-2005): Un análisis de la

Dependencia Espacial. 24. Medellin, Antioquia, Colombia.

Combes, P., & Lafurcade, M. (2001). Transport Costs Decline and Regional Inequalities: Evidence
from France. CEPR Discussion Papers(2894).

DEPARTAMETO ADMINISTRATIVO NACIONAL DE ESTADÍSTICA. (s.f.). COLOMBIA. PROYECCIONES

DE POBLACIÓN MUNICIPALES POR ÁREA 2005-2020. Oficial, DANE. Recuperado el Febrero
de 2016, de http://www.dane.gov.co/

DEPERTAMENTO NACIONAL DE PLANEACIÓN. (2013). Ejecuciones presupuestales de cada

municipio y cada departamento para el período 2000-2012. Oficial. Recuperado el 01 de
2016, de https://www.dnp.gov.co/programas/desarrollo-territorial/Paginas/ejecuciones-
presupuestales.aspx

Dixit, A. K., & Stiglitz, J. E. (1977). Monopolistic Competition and optimal Product diversity.
American Economic Review, 67(3), 297-308.

Fujita, M., & Thisse, J. F. (2013). Economics of Agglomeration: Cities, Industrial Location and
globalization (Second Edition ed.). New York: Cambridge University Press.

Fujita, M., Krugman, P., & Venables, A. (1999). The Spatial Economy: Cities, Regions, and
International Trade. Cambridge, Massachusetts: The MIT Press.

Fujita, M., Krugman, P., & Venables, A. (2002). The Spatial Economy; Cities, Regions, and
international trade. Cambridge, Massachusetts: The MIT Press.

Galeano Duque, V. (2013). Localización Espacial de la Actividad Económica en Medellín, 2005-2010

Un Enfoque de Economía Urbana. Ensayos Sobre Política Económica, 31(70), 216-266.

Garretsen, H., Schramm, M., & Brakman, S. (2003). The Spatial Distribution of Wages: Estimating
the Helpman-Hanson model for Germany. Tjalling C. Koopmans Research Institute.

Hanson, G. (1998). Market potential, increasing returns, and geographic concentration. NBER
Working Paper Series(6429).

Harris, C. (1954). The market as a factor in the localization of industry in the United States. Annals
of the Association of American Geographers, 315-348.

Helpman, E. (1998). The Size of Regions. Topic in public economics, 33-54.

Kiso, T. (2005). Does New Economic Geography Explain the Spatial Distribution Of Wages in Japan?
Krugman, P. (1991). Increasing Returns and Economic Geography. Journal of Political Economy, 99,
483-499.

Le Gallo, J. (2004). Space-time analysis of GDP disparities among European regions: A Markov
chains approach. (S. Publications, Ed.) International Regional Science Review, 27(2), 138-
163.

Loaiza Quintero, Ó. L., Moncada Mesa, J., Galeano Duque, V., Vélez Villegas, J. E., & Restrepo
Estrada, C. E. (2015). Un análisis cuantitativo de las disparidades regionales en Colombia
1985 y 2010: Una aplicación de las herramientas de la estadística espacial. Medellin:
Universidad de Antioquia.

Mion, G. (2003). Spatial Externalities and Empirical Analysis: The Case of Italy. Zentrum für
Europäische Wirtschaftsforschung GmbH.

Mion, G. (2004). Spatial Externalities and Empirical Analysis: The case of Italy. Journal of urban
Economics, 56(1), 97-118.

Moreno Jimenez , A. (1991). Modelización cartográfica de densidades mediante estimadores

Kernel. (U. A. Madrid, Ed.)

Polèse, M., & Rubiera, F. (2009). Economía urbana y regional. Thomson Reteurs.

Silverman, B. W. (1989). Density estimation for statics and data analysis.

Vélez Villegas, J. E. (2008). Los Procesos de Aglomeración en Colombia a la luz de la Nueva

Geografía Económica. Ensayos sobre Politica economica, 27(58), 106-139.
APPENDIX

A1. Descriptive statistics

Variable Media Variance Lower Quartile Upper Quartile

W 0,04028305 0,003506205 0,000350545 0,391219095

Y 0,148246736 0,020116649 0,026194639 1,047353858

H 0,049803525 0,003084608 0,003236874 0,375094853

Source: Own calculations based on DANE and DNP budget executions
W Collection by way of Industry and trade average per capita.
Y Total collection average per capita.
H Property tax average per capita.
All variables are for 2011.

A.2. MORAN statistics on the variables under analysis

Variable Statistical p-value

W 6.9659 1.632e-12

H 6.3736 9.231e-11

Y 6.4709 4.869e-11
Source: Own calculations based on DANE and DNP budget executions
 A.3 Spatial Distribution of collections by Predial Tax in the municipalities of the
department (2011)

Quantile Map Cluster Map

Source: Own calculations based on DANE and DNP budget executio

A.4 List of cluster municipalities

High-High Cluster Low-Low Cluster

municipalities municipalities

Medellín Vigía del Fuerte

Bello Urrao
Itagüí Frontino
Envigado Dabeiba
Sabaneta Cañas Gordas
La Estrella Pequé
Caldas Buriticá
Rionegro Sabana Larga
El Retiro Liborina
Guarne Giraldo
El Carmen del Viboral Armenia
Girardota Sopetrán
Copacabana
Don Matías
Apartado
Carepa