The magnitude and causes of agglomeration economies∗

Diego Puga§
IMDEA Social Sciences and CEPR

July 2009

Abstract: Firms and workers are much more productive in large

and dense urban environments. There is substantial evidence of
such agglomeration economies based on three aproaches. First,
on a clustering of production beyond what can be explained by
chance or comparative advantage. Second, on spatial patterns in
wages and rents. Third, on systematic variations in productivity
with the urban environment. However, more needs to be learned
about the causes of agglomeration economies. We have good
models of agglomeration through sharing and matching, but not
a deep enough understanding of learning in cities. Despite re-
cent progress, more work is needed to distinguish empirically
between alternative causes.

Key words: agglomeration economies.

jel classification: r30

∗ This paper has been written for the Journal of Regional Science’s 50th
Anniversary Symposium. The author
thanks Gilles Duranton for helpful comments.
§ Madrid Institute for Advanced Studies (imdea) Social Sciences, Antiguo pabellón central del Hospital

de Cantoblanco, Carretera de Colmenar Viejo km. 14, 28049 Madrid, Spain (e-mail: [email protected];
website: http://diegopuga.org).
1. Introduction

The concentration of firms and people in cities creates congestion and bids up the price of
land, and yet around half of the world population live in cities. What are the advantages of
cities, that are able to offset the obvious costs and attract so many enterprises and workers
to them? How large are these advantages and why do they arise? These are among
the most fundamental questions in urban economics, since without answering them we
cannot understand the very existence of cities. This paper reviews what we know about
the magnitude and causes of the productive advantages of cities and also tries to identify
the largest gaps in our knowledge of agglomeration economies.
Firms and workers are much more productive in large and dense urban environments
than in other locations. It is also in large cities where the vast majority of substantial
innovations emerge. The productivity advantages of cities and urban clusters with a high
density of firms and workers have been perceived for a long time, and already received
the attention of Adam Smith (1776) and Alfred Marshall (1890). Over the past thirty
years, urban economists have been rather successful at documenting and quantifying
these advantages. They have done so with three broad approaches. First, by showing
that productive activities are much more clustered than would be expected if location was
the result of a random outcome or merely reflected underlying differences across space
leading to comparative advantage. Second, by studying spatial patterns in wages and
land rents. If firms and workers are mobile and wages and land rents differ across space,
higher wages and land rents in large and dense urban environments must reflect some
productive advantage. Third, by looking at systematic variations in productivity across
space. This last and most direct approach been particularly fruitful in recent years, thanks
in part to the increasing availability of spatial data at the level of individual plants and
workers.
Stories about the causes of agglomeration economies are as old as the realization that
such advantages exist. The aforementioned classic works by Smith and Marshall contain
frequently cited discussions of the advantages arising from the greater specialization
made possible by larger markets, from sharing intermediate suppliers, from pooling in
labour markets, or from the localized transmission of ideas. However, understanding the
causes of agglomeration economies requires going much deeper than these classic stories
or the many more that have followed. For instance, we may perceive some advantages
from having a large local labour market, but what is the precise channel through which
such advantages operate? Is it because a larger labour market improves matching between
employers and employees? Or is it because large concentrations of employment iron out
idiosyncratic shocks and improve establishments’ ability to adapt their employment to
good and bad times? Or perhaps because larger markets allow workers to specialize

1
in a narrower set of activities and improve their performance? And how important are
these advantages relative to alternative sources of agglomeration economies not operating
through the labour market? To answer such questions we need good models that formal-
ize the microeconomic foundations of urban agglomeration economies, as well as detailed
empirical work able to identify and quantify the precise mechanisms at work. This is an
area where there has also been much recent progress. However, as we shall discuss in
detail below, there are very substantive open questions that forthcoming research ought
to address.

2. Evidence and magnitude of agglomeration economies

Excessive localization as a sign of agglomeration economies

One of the fundamental results in spatial economics is Starrett’s (1978) spatial impossibil-
ity theorem. This states that, once we abstract from the heterogeneity of the underlying
space, and without indivisibilities or increasing returns, any competitive equilibrium in
the presence of transport costs will feature only fully autarchic locations where every good
will be produced at small scales (see Ottaviano and Thisse, 2004, for a detailed discussion).
Thus, substantial localization or spatial concentration of economic activity may be seen as
a sign of agglomeration economies.
Starrett’s (1978) theorem already points to two reasons why, even in the absence of
scale economies at the level of cities, one would not expect a uniform distribution of
economic activity. First, because there is some inevitable lumpiness from small-scale indi-
visibilities in any production process. For instance, most technologies require production
establishments to be within a certain size range. If we were to pin a map on the wall
and, blindfolded, throw one dart at it for each establishment that exists in a sector, some
areas would inevitably get several darts and others none. The ‘dartboard approach’ of
Ellison and Glaeser (1997) yields measures of geographic concentration that correct for
both the size distribution of plants and for differences in the size of geographic areas.
The second reason why, even without agglomeration economies, industries would not be
uniformly distributed across space is that space itself is not uniform. Some areas are too
arid or too rugged to be used. Furthermore, other activities are closely tied to geographic
features such as natural resources and ports. However, Ellison and Glaeser (1999) find
that natural advantages, even when very widely defined, can predict only about 20% of
industrial localization. Panel-data studies using area fixed-effects to capture any sort of
localized advantage (e.g., Henderson, 1997) also find that such permanent advantages
leave substantial agglomeration effects unexplained.

2
 Duranton and Overman (2005) note three desirable properties of the spatial concen-
tration index of Ellison and Glaeser (1997): it is comparable across industries, it controls
for the concentration of overall economic activity when looking at individual sectors, and
it controls for the distribution of plant sizes. They also suggest two additional desirable
properties for a localization measure. One is to be able to accompany measures of localiza-
tion with measures of statistical significance. Another desirable property is to avoid the
bias created by aggregating data into large areas (boxes) instead of individual locations
(points) — the so-called modifiable area unit problem. If a cluster of firms spreads across
the border between two statistical units for which data is available, the close proximity of
firms will be missed at least in part by localization indices based on area aggregates. We
will see firms split across the two areas but not that firms are located within those areas so
that they are close to each other — or even that the areas share a border, since areas are typ-
ically treated symmetrically. To get around this problem, they propose a distance-based
approach that looks at the entire distribution of pairwise distances between plants and
compares them to the distribution resulting from random allocations of plants. They find
using data for the United Kingdom that about half of all 4-digit sectors are too localized at
a 5% confidence level for their clustering to reflect merely a random outcome. In addition,
localization mostly takes place within close distances — less than 50 kilometres.
Instead of looking at location in general, Rosenthal and Strange (2003) focus on the lo-
cation decisions of new establishments, which are less constrained by past characteristics.
Looking at the production environment in concentric rings of varying size around the
zip code of new establishments, they also find that agglomeration effects decrease quite
rapidly with distance.

Quantifying agglomeration economies through wages and rents

Comparing wages across areas provides more direct evidence of the existence and magni-
tude of agglomeration economies (see, e.g., Glaeser and Maré, 2001, Wheaton and Lewis,
2002, Mion and Naticchioni, 2005, Combes, Duranton, and Gobillon, 2008, Combes, Du-
ranton, Gobillon, and Roux, 2010). A common justification for this approach is that in
competitive markets labour is paid the value of its marginal product. However, even if
labour markets are not perfectly competitive, higher wages in large/dense urban areas
can be seen as evidence of higher productivity. For workers, higher wages may be offset
by larger commuting and housing costs. However, higher wages and land rents in large
cities would lead firms to relocate elsewhere unless there were some significant productive
advantages.
A key concern when interpreting the urban wage premium as evidence of agglomer-
ation economies is that the ability of workers may also vary across cities. If more able

3
workers sort into larger cities, then the urban wage premium may reflect their greater
abilities instead of any intrinsic advantage to urban location. Glaeser and Maré (2001)
discuss this issue at length and explore a number of solutions: controlling for observable
skills, instrumenting for urban residence using parental background, and finally exploit-
ing the panel dimension of the data to include individual worker fixed-effects. They find
that, even after all these corrections, there is significant wage premium associated with
living and working in dense cities, although smaller in magnitude than before taking
unobserved ability into account. Working with French data, Combes et al. (2010) find that
the elasticity of the wage premium associated with each city with respect to urban density
is cut by about one half when worker fixed-effects are introduced in the estimation. They
interpret this as evidence of substantial sorting, implying that workers in denser cities
tend to have greater unobserved skills. Since introducing worker fixed-effects makes the
estimation of agglomeration economies be based exclusively on migrants, Glaeser and
Maré (2001) argue instead that the drop in magnitude may be partly due to the fact that it
takes time for workers to fully reap the benefits of locating in big cities.
A related approach is to study the spatial variation of rents. If firms are willing to
pay higher rents in big cities it must be because there is some compensating productive
advantage. A difficulty with this stretegy is finding the required data on comercial rents,
so residential rents are sometimes used as a proxy (Dekle and Eaton, 1999).
Following Rosen (1979) and Roback (1982), a particularly fruitful approach is to com-
bine data on wages and rents. This helps disentangle the consumption amenities from
the productive advantages of big cities. For workers, higher wages make them better off
whereas higher rents make them worse off. Thus, greater consumption amenities in a
city will make workers willing to accept both lower wages and higher rents. For firms,
both higher wages and higher rents mean increased costs. Thus, localized productive ad-
vantages will make firms willing to accept higher wages and higher rents. Consequently,
both consumption amenities and productive advantages should be associated with higher
rents. However, consumption amenities should be associated with lower wages whereas
productive advantages should be associated with higher wages. Note, however, that this
raises an additional concern when looking for agglomeration effects through wages. If big
cities are associated with both better amenities and higher productivity, the net effect on
wages may be ambiguous.

Productivity evidence of local increasing returns: more output out of the same inputs

By definition, local external scale economies imply that plants are able to produce more
output with the same inputs in larger, denser, urban environments. Thus, perhaps the
most natural way to directly measure the magnitude of agglomeration economies is to

4
use data on outputs and inputs to estimate how productivity varies across space. The
first influential modern study to do this, by Sveikauskas (1975), regressed log output
per worker in a cross-section of city-industries on log city population and found that a
doubling of population increases output per worker by about 6 percent. After dealing
with various concerns discussed below, more recent productivity studies suggest that a
doubling of city size increases productivity by between 3 and 8 percent for a large range
of city sizes (Rosenthal and Strange, 2004).
Output per worker, as used by Sveikauskas (1975), may not be the best measure of
productivity in this context. As Moomaw (1983) points out, if capital is used more inten-
sively in large cities, then estimating agglomeration economies using output per worker
will lead to an upward bias in the estimated effects. For this reason the literature has
moved towards using total factor productivity, calculated at the aggregate level for each
area being considered or, more recently, at the plant level. A fundamental contribution
to quantifying agglomeration economies through individual productivity estimates is
Henderson (2003), who uses plant level data in high-tech and machinery sectors for the
United States and is particularly careful in dealing with potential biases.
Another problem raised by early estimates of the magnitude of agglomeration
economies, also discussed by Moomaw (1981), is that higher output per worker may
be not so much a consequence of higher local employment or density but its cause. If
a location has an underlying productive advantage, then it will tend to attract more
firms and workers and become larger as a result. Following Ciccone and Hall (1996),
the standard way to tackle this issue is to instrument for current size/density. Since
there is substantial persistence in the spatial distribution of population but the drivers
of high productivity today differ from those in the distant past, the usual instruments are
historical data for size/density as well as characteristics that are thought to have affected
the location of population in the past but that are mostly unrelated to productivity today.
Ciccone and Hall (1996) find that reverse causality on this matter is only a minor issue.
This conclusion has been confirmed by much of the subsequent literature. In a recent
contribution, Combes et al. (2010), use a wide range of historical and geological instru-
ments — the latter justified because fertile soil was an important attraction for population
in the past, cities with large historical populations tend to remain large today, but fertile
soil no longer matters much for local productivity. They conclude that instrumenting
only reduces the estimated magnitude of agglomeration economies by about one fifth. An
alternative strategy to deal with a potential endogeneity bias is to exploit the time dimen-
sion of panel data. In particular, Henderson (2003) includes city-time fixed effects when
estimating plant-level productivity, to capture any unobserved attributes that may have
attracted more entrepreneurs to a given city. Greenstone, Hornbeck, and Moretti (2008)
tackle this same issue by comparing outcomes in us counties chosen by large new plants

5
with outcomes in runner-up counties that were competing for the same plants. They find
that, after the new plant opening, incumbent plants in chosen counties experience a sharp
increase in total factor productivity relative to incumbent plants in runner-up counties.
Finally, productivity estimates are also subject to the caveats discussed above in relation
to sorting. A difficult question that arises upon finding significant advantages of agglom-
eration is why, then, do some firms choose to remain relatively isolated (see Holmes, 2010,
for a discussion). A partial answer is that higher wages and land prices in denser areas
help justify location in more and in less dense areas as a spatial equilibrium. Nevertheless,
this does not resolve the important concern that isolated plants may be substantially
different from plants in denser locations. When looking at workers, migrants in panel
data help adress those concerns. However, when looking at firms, this is much more
difficult.

Agglomeration or survival of the fittest?

While the productive advantages of large cities have usually been attributed to agglomera-
tion economies (i.e., larger cities promote interactions that increase productivity), recently
an alternative explanation has been offered based on ‘firm selection’ (i.e., larger cities
toughen competition allowing only the most productive firms to survive). Melitz and
Ottaviano (2008) model this argument by incorporating endogenous price-cost mark-ups
in the framework of Melitz (2003) (see also Syverson, 2004, who studies the ready-made
concrete industry, and Baldwin and Okubo, 2006, for a related approach focusing on firm
sorting). They show that the presence of more firms in larger markets makes competition
tougher and this leads less productive firms to exit. Thus, higher average productivity in
larger cities could result from a survival of the fittest rather than from a productivity boost
based on agglomeration economies.
In a recent paper, Combes, Duranton, Gobillon, Puga, and Roux (2009) develop a
framework to distinguish between agglomeration and firm selection in explaining why
average productivity is higher in larger cities. They nest a generalized version of the firm
selection model of Melitz and Ottaviano (2008) (freed of distributional assumptions and
extended to many cities) and a simple model of agglomeration in the spirit of Fujita and
Ogawa (1982) and Lucas and Rossi-Hansberg (2002). This nested model enables them to
parameterise the relative importance of agglomeration and selection.
The main prediction of the nested model is that, while selection and agglomeration
effects both make average firm log productivity higher in larger cities, they have different
predictions for how the shape of the log productivity distribution varies with city size. In
particular, stronger selection effects in larger cities should lead to a greater left truncation
of the distribution of firm log productivities in larger cities, as the least productive firms

6
exit. Stronger agglomeration effects in larger cities should lead instead to a greater right-
wards shift of the distribution of firm log productivities in larger cities, as agglomeration
effects make all firms more productive.
While most models of agglomeration feature a representative firm, the model of
Combes et al. (2009) features heterogeneous firms and allows the benefits of agglomeration
to vary systematically across firms within each city. In particular, through a simple techno-
logical complementarity between the productivity of firms and that of workers, firms that
are more productive per se are also better at reaping the benefits of agglomeration. The
model predicts that if this additional effect holds in practice then agglomeration should
lead not only to a rightwards shift but also to an increased dilation of the distribution of
firm log productivities in larger cities.
To implement this model on exhaustive French establishment-level data, Combes et al.
(2009) develop a new quantile approach that allows estimating a relative change in left
truncation, shift, and dilation between two distributions. Their main finding is that
productivity differences across urban areas in France are mostly explained by agglomera-
tion. The distribution of firms’ log productivities in large French cities (population above
200,000) is remarkably well described by taking the distribution of firms’ productivities in
small French cities, dilating it, and shifting it to the right. This holds for the productivity
distributions of firms across all sectors as well as most two-digit sectors when considered
individually. For all manufacturing and business service sectors combined, relative to the
rest of the country, the distribution of firms log productivities in large cities is shifted to
the right by 0.09 and dilated by a factor of 1.22. Firms in large cities are thus on average
about 9 percent more productive than in small cities. Because of dilation, this productivity
advantage is only of 5 percent for firms at the bottom quartile and 14 percent for firms at
the top quartile. On the other hand, they find no differences between small and large cities
in terms of selection-driven left truncation of the log productivity distribution.

3. The causes of agglomeration economies

While there are solidly established conclusions regarding the existence of agglomeration
economies, and a reasonably narrow range for their estimated magnitude, identifying the
causes of agglomeration economies is proving more difficult.
There is a large theoretical literature that develops three broad classes of mechanisms
to explain the existence of urban agglomeration economies (this classification follows
Duranton and Puga, 2004, who cover the theoretical literature in detail). First, a larger
market allows for a more efficient sharing of local infrastructure and facilities, a variety
of intermediate input suppliers, or a pool of workers with similar skills. Second, a larger
market also allows for a better matching between employers and employees, buyers and

7
suppliers, or business partners. This better matching can take the form of improved
chances of finding a suitable match, a higher quality of matches, or a combination of
both. Finally, a larger market can also facilitate learning, for instance by promoting the
development and widespread adoption of new technologies and business practices.
Going from modelling these mechanisms to identifying them empirically is not
straightforward because they all share the prediction that productivity increases with
the scale of an activity at a location. This ‘Marshallian equivalence’ (Duranton and
Puga, 2004) makes it very difficult to distinguish the main causes of the productivity
advantages of cities. There are some clues in the aggregate estimates. For instance, the
steep spatial decay of agglomeration economies (Rosenthal and Strange, 2003, Henderson,
2003, Desmet and Fafchamps, 2005) points to the importance of local interactions, and the
rising wage profile following relocations to big cities (Glaeser and Maré, 2001) suggests
an important dynamic component. However, to identify specific mechanisms, we must
understand their microeconomic foundations and look for features that distinguish each
particular mechanism. We now turn to reviewing the different mechanisms, providing
a brief description of each theoretical argument and a review of the available empirical
evidence supporting it.

Sharing facilities

Perhaps the simplest way to model localized increasing returns is to point to indivisi-
bilities in the provision of certain goods or facilities, such as local infrastructure. Once
the large fixed cost associated with a shared facility has been incurred, the larger the
population that shares the facility the lower the cost per user. At the same time, congestion
of the facility and crowding of the land surrounding it will place limits to growth of the
user base (see Buchanan, 1965, for a pioneering discussion of the problems associated
with the provision of shared facilities, and Scotchmer, 2002, for a detailed review of the
literature).
While simple from a modelling perspective, this mechanism can be of practical impor-
tance in certain contexts. For instance, in their study of urban sprawl using remote-sensing
data, Burchfield, Overman, Puga, and Turner (2006) find that residences are closer to
each other in cities where water provision relies on shared public facilities whereas urban
development is more scattered in areas cities aquifers make individual household wells
viable.

Sharing suppliers

A more sophisticated model of urban agglomeration economies through sharing is due

to Abdel-Rahman and Fujita (1990). They capture the advantages for final producers of

8
being able to share a larger common base of suppliers in larger and more specialized cities.
In their model, perfectly competitive final-goods firms use sector-specific intermediate
inputs. These inputs are produced by a monopolistically competitive industry featuring
Ethier’s (1982) production-side version of the monopolistic competition framework of
Dixit and Stiglitz (1977). Final goods are freely tradable while intermediates are only sold
locally. Cities are subject to housing and commuting costs that increase with their pop-
ulation. In equilibrium, aggregate production at the city-sector level exhibits increasing
returns, despite constant returns to scale in perfectly-competitive final production. The
reason is that an increase in final production, by virtue of expanding input sharing across
a wider variety of suppliers, requires a less-than-proportional increase in primary factors.
Rosenthal and Strange (2001) study the empirical importance of sharing a common
base of suppliers relative to other sources of agglomeration. They do so by regressing
geographic concentration for each sector (measured by the index of Ellison and Glaeser,
1997) on proxies for different agglomeration motives computed also at the sector level
(see also Audretsch and Feldman, 1996, for an earlier example of this approach using
cross-sectional variation for identification). They measure the importance of input sharing
for each sector using the purchase of manufactured and non-manufactured inputs as
a share of value added. They find effects that are weak relative to other motives for
agglomeration, with coefficients that are often statistically insignificant.
Overman and Puga (2010), while studying the importance of the labour pooling mecha-
nism discussed below using uk data, control for the importance of input sharing following
the approach of Rosenthal and Strange (2001). When they simply use input purchases
relative to value added as a proxy for the importance of input sharing, they find no
support for this being an important determinant of spatial concentration. However, they
then argue that, when an industry buys a lot from other industries, the importance of
input sharing for its concentration will depend on whether those other industries are, in
turn, spatially concentrated or dispersed. For instance, the meat processing industry is
a large buyer of inputs from farms and from the plastic film industry. However, farms
are very dispersed across the country and so is the plastic film industry, since it supplies
many other sectors located in different places in addition to meat processing. Hence, the
meat processing industry has no reason to concentrate spatially even if it makes large
intermediate purchases: it can easily find its inputs everywhere. For a sector to cluster
to share intermediate suppliers, it must be the case not only that the sector makes large
purchases of intermediates but also that those intermediates are supplied by industries
that are themselves very spatially concentrated. Following this line of reasoning, to better
capture the importance of vertical linkages for a particular industry, Overman and Puga
(2010) calculate sum of the Ellison and Glaeser (1997) index across all other industries
weighted by the share each represents in the input shares of the industry being considered.

9
This more sophisticated measure yields strong support for input sharing as a motive for
agglomeration at the sector level. Amiti and Cameron (2007) exploit detailed information
on the location of input suppliers in Indonesia to show that the benefits of proximity to
suppliers fall rapidly with distance.
Instead of trying to disentangle the relative importance of various potential causes
of agglomeration by looking at which industries are more and less agglomerated (like
Audretsch and Feldman, 1996, Rosenthal and Strange, 2001, and others), Ellison, Glaeser,
and Kerr (2010) study co-agglomeration patterns. The underlying idea is that, while plants
in any given industry are similar in many dimensions, when we look across industries
plants are similar in some dimensions and not in others. For example, some industry pairs
exchange intermediate inputs but employ very different workers. Others employ similar
workers but do not exchange inputs. Hence, by looking at which similarities across indus-
tries help to predict better which industry pairs are co-agglomerated, one can gain insight
into the relative importance of different motives for agglomeration. Ellison et al. (2010)
find that sectors buying similar intermediates tend to co-agglomerate the most, followed
by sectors that employ similar workers. To address reverse causality (i.e., the possibility
that certain industry pairs end up buying intensively from each other precisely because
they are agglomerated, they use uk measures to instrument for us industry characteristics.
In Abdel-Rahman and Fujita (1990), compared to a firm in a dense location, a relatively
isolated firm produces using a narrower input mix purchased from outside suppliers.
Alternatively, a relatively isolated firm might produce with a similar input mix but pro-
duce more of these inputs in-house. This possibility is discussed by Stigler (1951) and
investigated empirically by Holmes (1999). He combines detailed plant level data for the
United States with spatial data on input purchases. Using this, he first shows that the most
concentrated industries buy more inputs from outside suppliers in locations where they
are clustered than in the rest of the country. Then Holmes (1999) regresses local purchased
input intensity (differenced from the industry mean) on same-industry employment in
the establishment’s own county and in other counties within fifty miles (again differenced
from the industry mean). The results indicate that the purchased-inputs intensity of a
plant increases with the level of employment of neighboring plants in the same industry.
This provides arguably the strongest empirical support to date for the importance of input
sharing.

Sharing the gains from individual specialization

Adam Smith’s (1776) famous pin factory example suggests that perhaps the presence of
more workers in a given activity within a city increases output more than proportionately
not just because extra workers can carry new tasks but because it allows existing workers

10
to specialise on a narrower set of tasks. Several papers (Baumgardner, 1988, Becker and
Murphy, 1992, Duranton, 1998, Becker and Henderson, 2000, Henderson and Becker, 2000)
develop models exploring the idea that a larger market fosters specialization.
On the empirical side, the main evidence in favour of increasing specialization in larger
markets comes from looking at professionals. For instance, Baumgardner (1988) shows
that physicians perform a narrower range of activities in large markets. The work of
Holmes (1999) discussed above can also be seen as supporting greater specialization in
dense markets, although in this case it is greater specialization across firms as in-house
input production is transferred to outside suppliers.

Sharing a labour pool

While there are various interpretations of labour market pooling as a source of agglom-
eration economies, some of them reviewed below, Marshall emphasized that “a localized
industry gains a great advantage from the fact that it offers a constant market for skill”
(Marshall, 1890, p. 271). Krugman (1991) formalizes this reasoning by considering a series
of sectors where establishments experience idiosyncratic shocks. Individual profits are
convex in the establishment-specific shock, since each establishment responds to the shock
by adjusting its levels of both production and employment. However, changes in the
establishment’s employment affect local wages, and the effect is greater the more isolated
the establishment is from other establishments in the same sector or using similar workers.
If wages are higher when the establishment wants to expand production in response to a
positive shock and lower when it wants to contract production in response to a negative
shock, this limits the establishment’s ability to adapt its employment level to good and
bad times. Consequently, establishments that tend to experience substantial changes in
their employment relative to other establishments using workers with similar skills will
find it advantageous to locate in places where there is a large number of workers with
such skills. Here agglomeration arises because large concentrations of employment iron
out idiosyncratic shocks and facilitate the transfer of labour from low to high productivity
establishments.
Overman and Puga (2010) look at this mechanism empirically. They measure the
likely importance of labour pooling by calculating the fluctuations in employment of
individual establishments relative to their sector and averaging by sector. They finds that
sectors whose establishments experience more idiosyncratic volatility are more spatially
concentrated, even after controlling for a range of other industry characteristics. Overman
and Puga (2010) study labour pooling looking at each sector separately. However, labour
pooling could work across sectors if these use workers with similar skills. Indeed, Ellison

11
et al. (2010) find that industries with similar labor mixes tend to co-aglomerate, which is
indicative of labour pooling mattering also for bringing sectors together.

Better matching

Another advantage of thick labour markets, in addition to the labour pooling argument
discussed above, is that they lead to better matching between employers and employees
— note that a similar argument can be made about matching between buyers and sup-
pliers, or between business partners. Helsley and Strange (1990) formalize this argument
by considering firms with heterogenous skill requirements represented by equally-spaced
points on a circumference. Workers also have differentiated skills uniformly spread over
the same circumference, and must incur more costly retraining the greater the difference
between their skill and the skill required by their employer. A larger city allows for the
skill space to be more densely covered by firms, and thus reduces the average cost of
mismatches.
Better matching can also take the form of improved chances of finding a suitable match.
The frictional search literature often works with an aggregate matching function that
yields the number of matches as a function of vacancies and job seekers. However, mi-
croeconomic foundations for such a matching function typically yield constant instead of
increasing returns to scale. An important exception is the stock-flow matching framework
of Coles (1994) and Coles and Smith (1998). Consider an unemployed worker who can
simultaneously apply to all job vacancies that may suit her. In the first instance, the
worker applies to the entire stock of available vacancies. If none of the applications are
successful, the worker subsequently only applies to newly posted vacancies. Similarly, a
new vacancy receives applications from the entire stock of workers, but if none of these
initial applications result in a suitable pairing, from then on the vacancy only receives
applications from newly unemployed workers. The total number of matches is then the
sum of matches between the flow of vacancies and the stock of unemployed workers and
of matches between the flow of unemployed workers and the stock of vacancies. This
yields naturally a matching function that exhibits increasing returns to scale. The intuition
is simple: in a market with more job opportunities that can be explored simultaneously it
is less likely that none of them work out.
Finally, there are interesting interactions between better matching in terms of improved
chances of finding a suitable match and in terms of a higher quality of matches. A higher
probability of matching in thicker markets allows firms and workers to be more “choosy”,
increasing the average quality of matches but somewhat reducing the higher probability
of matching (Berliant, Reed, and Wang, 2000).

12
 On the empirical side, Gan and Li (2004), after presenting a model where, as in Coles
(1994) and Coles and Smith (1998), the probability of matching increases with the thickness
of the market, test its predictions empirically. They compare fields of different size in
the academic recruitment market for new PhDs in Economics, and find that a field of
specialization with more job openings and more candidates offers a higher probability of
matching. Another piece of evidence in favour of larger cities facilitating matching is the
work of Costa and Kahn (2000). They show that couples in which both spouses have col-
lege degrees are increasingly likely to be located in the largest metropolitan areas, and not
just because they meet there. One explanation is that college-educated couples are more
likely to face a co-location problem and moving to big cities increases the chances that both
find suitable matches. Gathering evidence on the other matching mechanism discussed,
that thicker markets improve the quality of matches, is more complicated because of the
intrinsic difficulties of measuring match quality. Hopefully, the increasing availability
of matched employer-employee micro-data will encourage more work on agglomeration
through matching.

Learning

Agglomeration mechanisms directly dealing with learning have received much less at-
tention in the theoretical literature than the sharing and matching mechanisms discussed
above. There are, however, some noteworthy exceptions. Glaeser (1999) develops a model
in which young workers migrate to big cities because interactions with experienced work-
ers helps them acquire valuable skills and experienced workers remain in cities to share
the rents of this learning process. Besides this purposeful transmission of knowledge, the
informal literature on learning in cities has also emphasized the casual and unintended
flows of information facilitated by big cities. However, despite being often motivated by
examples that are specifically spatial in nature, the literature on social learning (see Vives,
1996, for a review) has so far not produced models capturing this microfoundation of
urban agglomeration economies. Nevertheless, spatial interactions motivate the external-
ity used in the richest models studying the spatial allocation of production and housing
within a city (Fujita and Ogawa, 1982, Ota and Fujita, 1993, Lucas and Rossi-Hansberg,
2002).
In addition to facilitating the transmission of knowledge, cities are also seen as pro-
moting the creation of new knowledge. The work of Jacobs (1969) is often associated
with the idea that diversified urban environments facilitate search and experimentation
in innovation. Duranton and Puga (2001) develop microeconomic foundations for such
a role. In their model, a young firm needs a period of experimentation to realise its full
potential — the entrepreneur may have a project, but may not know all the details of the

13
product to be made, what components to use, where to source them, which workers to
hire, and how to finance the venture. A diverse city provides many alternatives to try
without having to relocate and this creates dynamic advantages to urban diversity. When
combined with more standard static agglomeration economies and urban congestion
costs, this justifies the coexistence of diversified and specialised cities and the agglom-
eration of firms at different stages of their life-cycle in cities of each type. Young firms
locate in more diverse urban environments and, when their products mature, relocate to
more specialized towns. Looking at establishment relocations across France, Duranton
and Puga (2001) find evidence of the relocation pattern predicted by their model.
On the empirical side, there is strong support for the idea that cities facilitate innova-
tion, the diffusion of knowledge, and the acquisition of skills. Audretsch and Feldman
(1996) show that innovative activity, as measured by significant new product introduc-
tions, tends to cluster geographically to a greater extent in industries where new economic
knowledge plays a more important role. This greater spatial concentration of innovation
holds even after controlling for the concentration of production.
A priori, the diffusion of knowledge might seem hard to trace. However, Jaffe, Trajten-
berg, and Henderson (1993) track knowledge flows by looking at patent citations. They
show that inventors are much more likely to cite prior patents with inventors from the
same city than a randomly drawn control sample of cited patents. No (2003), using data
on the adoption of advanced manufacturing technologies in Canada, finds that adoption
is more likely in locations with more prior adopters, particularly if they use similar tech-
nologies but do not compete in the same detailed sector. Charlot and Duranton (2004)
use survey data on communication between workers to show that workplace communi-
cation is more extensive in urban areas. In addition, more workplace communication is
associated with higher wages.
Some results from studies quantifying agglomeration economies already discussed
above Rosenthal and Strange (2003), Henderson (2003), Glaeser and Maré (2001) might
also be interpreted as supporting the relevance of learning mechanisms. In particular,
Glaeser and Resseger (2010) and others have argued that there are strong complementari-
ties between skills and agglomeration, so that skills amplify the benefits of agglomeration
and agglomeration facilitates the accumulation of skills.

4. Conclusions

Despite the broad agreement on the magnitude of agglomeration economies at the urban
level, the literature has been far less successful at distinguishing between the possible
sources. This requires models that work out the microfoundations to help identify dis-
tinguishing features and empirical work that carefully exploits these for identification.

14
On the theoretical side, we have good models of agglomeration through sharing and
matching, but not a deep enough theoretical understanding of learning in cities. On
the empirical side, evidence of matching as a source of agglomeration is perhaps most
needed. However, despite several notable existing contributions, there is room for much
more work able to credibly claim identification of a particular driver of agglomeration.

References
Abdel-Rahman, Hesham M. and Masahisa Fujita. 1990. Product variety, Marshallian
externalities, and city sizes. Journal of Regional Science 30(2):165–183.

Amiti, Mary and Lisa Cameron. 2007. Economic geography and wages. Review of Eco-
nomics and Statistics 89(1):15–29.

Audretsch, David B. and Maryann P. Feldman. 1996. R&D spillovers and the geography
of innovation and production. American Economic Review 86(3):630–640.

Baldwin, Richard E. and Toshihiro Okubo. 2006. Heterogeneous firms, agglomeration

and economic geography: spatial selection and sorting. Journal of Economic Geography
6(3):323–346.

Baumgardner, James R. 1988. The division of labor, local markets, and worker organiza-
tion. Journal of Political Economy 96(3):509–527.

Becker, Gary S. and Kevin M. Murphy. 1992. The division of labor, coordination costs, and
knowledge. Quarterly Journal of Economics 107(4):1137–1160.

Becker, Randy and J. Vernon Henderson. 2000. Intra-industry specialization and urban
development. In Jean-Marie Huriot and Jacques-François Thisse (eds.) Economics of
Cities: Theoretical Perspectives. Cambridge: Cambridge University Press, 138–166.

Berliant, Marcus, Robert R. Reed, III, and Ping Wang. 2000. Knowledge exchange, match-
ing, and agglomeration. Discussion Paper 135, Federal Reserve Bank of Minneapolis.

Buchanan, James M. 1965. An economic theory of clubs. Economica 32(125):1–14.

Burchfield, Marcy, Henry G. Overman, Diego Puga, and Matthew A. Turner. 2006. Causes
of sprawl: A portrait from space. Quarterly Journal of Economics 121(2):587–633.

Charlot, Sylvie and Gilles Duranton. 2004. Communication externalities in cities. Journal
of Urban Economics 56(3):581–613.

Ciccone, Antonio and Robert E. Hall. 1996. Productivity and the density of economic
activity. American Economic Review 86(1):54–70.

Coles, Melvyn G. 1994. Understanding the matching function: The role of newspapers
and job agencies. Discussion Paper 939, Centre for Economic Policy Research.

15
Coles, Melvyn G. and Eric Smith. 1998. Marketplaces and matching. International Economic
Review 39(1):239–255.
Combes, Pierre-Philippe, Gilles Duranton, and Laurent Gobillon. 2008. Spatial wage
disparities: Sorting matters! Journal of Urban Economics 63(2):723–742.
Combes, Pierre-Philippe, Gilles Duranton, Laurent Gobillon, Diego Puga, and Sébastien
Roux. 2009. The productivity advantages of large cities: Distinguishing agglomeration
from firm selection. Discussion Paper 7191, Centre for Economic Policy Research.
Combes, Pierre-Philippe, Gilles Duranton, Laurent Gobillon, and Sébastien Roux. 2010.
Estimating agglomeration effects with history, geology, and worker fixed-effects. In
Edward L. Glaeser (ed.) Agglomeration Economics. Chicago, il: Chicago University Press,
forthcoming.
Costa, Dora L. and Matthew E. Kahn. 2000. Power couples: Changes in the locational
choice of the college educated, 1940-1990. Quarterly Journal of Economics 115(4):1287–
1315.
Dekle, Robert and Jonathan Eaton. 1999. Agglomeration and land rents: Evidence from
the prefectures. Journal of Urban Economics 46(2):200–214.
Desmet, Klaus and Marcel Fafchamps. 2005. Changes in the spatial concentration of
employment across us counties: A sectoral analysis 1972–2000. Journal of Economic
Geography 5(3):261–284.
Dixit, Avinash K. and Joseph E. Stiglitz. 1977. Monopolistic competition and optimum
product diversity. American Economic Review 67(3):297–308.
Duranton, Gilles. 1998. Labor specialization, transport costs, and city size. Journal of
Regional Science 38(4):553–573.
Duranton, Gilles and Henry G. Overman. 2005. Testing for localization using micro-
geographic data. Review of Economic Studies 72(4):1077–1106.
Duranton, Gilles and Diego Puga. 2001. Nursery cities: Urban diversity, process innova-
tion, and the life cycle of products. American Economic Review 91(5):1454–1477.
Duranton, Gilles and Diego Puga. 2004. Micro-foundations of urban agglomeration
economies. In Vernon Henderson and Jacques-François Thisse (eds.) Handbook of Re-
gional and Urban Economics, volume 4. Amsterdam: North-Holland, 2063–2117.
Ellison, Glenn and Edward L. Glaeser. 1997. Geographic concentration in us manufactur-
ing industries: A dartboard approach. Journal of Political Economy 105(5):889–927.
Ellison, Glenn and Edward L. Glaeser. 1999. The geographic concentration of industry:
Does natural advantage explain agglomeration? American Economic Review Papers and
Proceedings 89(2):311–316.
Ellison, Glenn, Edward L. Glaeser, and William Kerr. 2010. What causes industry ag-
glomeration? Evidence from coagglomeration patterns. American Economic Review
:forthcoming.

16
Ethier, Wilfred J. 1982. National and international returns to scale in the modern theory of
international trade. American Economic Review 72(3):389–405.

Fujita, Masahisa and Hideaki Ogawa. 1982. Multiple equilibria and structural transition of
non-monocentric urban configurations. Regional Science and Urban Economics 12(2):161–
196.

Gan, Li and Qi Li. 2004. Efficiency of thin and thick markets. Working Paper 11813,
National Bureau of Economic Research.

Glaeser, Edward L. 1999. Learning in cities. Journal of Urban Economics 46(2):254–277.

Glaeser, Edward L. and David C. Maré. 2001. Cities and skills. Journal of Labor Economics
19(2):316–342.

Glaeser, Edward L. and Matthew G. Resseger. 2010. The complementarity between cities
and skills. Journal of Regional Science :forthcoming.

Greenstone, Michael, Richard Hornbeck, and Enrico Moretti. 2008. Identifying agglomer-
ation spillovers: Evidence from million dollar plants. Working Paper 13833, National
Bureau of Economic Research.

Helsley, Robert W. and William C. Strange. 1990. Matching and agglomeration economies
in a system of cities. Regional Science and Urban Economics 20(2):189–212.

Henderson, J. Vernon. 1997. Externalities and industrial development. Journal of Urban

Economics 42(3):449–470.

Henderson, J. Vernon. 2003. Marshall’s scale economies. Journal of Urban Economics

53(1):1–28.

Henderson, J. Vernon and Randy Becker. 2000. Political economy of city sizes and forma-
tion. Journal of Urban Economics 48(3):453–484.

Holmes, Thomas J. 1999. Localization of industry and vertical disintegration. Review of

Economics and Statistics 81(2):314–325.

Holmes, Thomas J. 2010. Structural, experimentalist, and descriptive approaches to em-

pirical work in regional economics. Journal of Regional Science :forthcoming.

Jacobs, Jane. 1969. The Economy of Cities. New York: Random House.

Jaffe, Adam B., Manuel Trajtenberg, and Rebecca Henderson. 1993. Geographic local-
ization of knowledge spillovers as evidenced by patent citations. Quarterly Journal of
Economics 108(3):577–598.

Krugman, Paul R. 1991. Geography and Trade. Cambridge, ma: mit Press.

Lucas, Robert E., Jr. and Esteban Rossi-Hansberg. 2002. On the internal structure of cities.
Econometrica 70(4):1445–1476.

Marshall, Alfred. 1890. Principles of Economics. London: Macmillan.

17
Melitz, Marc and Gianmarco I. P. Ottaviano. 2008. Market size, trade and productivity.
Review of Economic Studies 75(1):295–316.
Melitz, Marc J. 2003. The impact of trade on intra-industry reallocations and aggregate
industry productivity. Econometrica 71(6):1695–1725.
Mion, Giordano and Paolo Naticchioni. 2005. Urbanization externalities, market potential
and spatial sorting of skills and firm. Discussion Paper 5172, Centre for Economic Policy
Research.
Moomaw, Ronald L. 1981. Productivity and city size? a critique of the evidence. Quarterly
Journal of Economics 96(4):675–688.
No, Joung Yeo Angela. 2003. Agglomeration effects in the diffusion of advanced manu-
facturing technologies. Processed, University of Toronto.
Ota, Mitsuru and Masahisa Fujita. 1993. Communication technologies and spatial orga-
nization of multi-unit firms in metropolitan areas. Regional Science and Urban Economics
23(6):695–729.
Ottaviano, Gianmarco I. P. and Jacques-François Thisse. 2004. Agglomeration and eco-
nomic geography. In Vernon Henderson and Jacques-François Thisse (eds.) Handbook of
Regional and Urban Economics, volume 4. Amsterdam: North-Holland, 2563–2608.
Overman, Henry G. and Diego Puga. 2010. Labour pooling as a source of agglomera-
tion: An empirical investigation. In Edward L. Glaeser (ed.) Agglomeration Economics.
Chicago, il: Chicago University Press, forthcoming.
Roback, Jennifer. 1982. Wages, rents, and the quality of life. Journal of Political Economy
90(6):1257–1278.
Rosen, Sherwin. 1979. Wage-based indexes of urban quality of life. In Peter N. Miezkowski
and Mahlon R. Straszheim (eds.) Current Issues in Urban Economics. Baltimore, md: Johns
Hopkins University Press, 74–104.
Rosenthal, Stuart S. and William Strange. 2004. Evidence on the nature and sources of
agglomeration economies. In Vernon Henderson and Jacques-François Thisse (eds.)
Handbook of Regional and Urban Economics, volume 4. Amsterdam: North-Holland, 2119–
2171.
Rosenthal, Stuart S. and William C. Strange. 2001. The determinants of agglomeration.
Journal of Urban Economics 50(2):191–229.
Rosenthal, Stuart S. and William C. Strange. 2003. Geography, industrial organization,
and agglomeration. Review of Economics and Statistics 85(2):377–393.
Scotchmer, Suzanne. 2002. Local public goods and clubs. In Alan J. Auerbach and Martin
Feldstein (eds.) Handbook of Public Economics, volume 4. Amsterdam: North-Holland,
1997–2042.
Smith, Adam. 1776. An Inquiry into the Nature and Causes of the Wealth of Nations. London:
Printed for W. Strahan, and T. Cadell.

18
Starrett, David A. 1978. Market allocations of location choice in a model with free mobility.
Journal of Economic Theory 17(1):21–37.

Stigler, George J. 1951. The division of labor is limited by the extent of the market. Journal
of Political Economy 59(3):185–193.

Sveikauskas, Leo. 1975. Productivity of cities. Quarterly Journal of Economics 89(3):393–413.

Syverson, Chad. 2004. Market structure and productivity: A concrete example. Journal of
Political Economy 112(6):1181–1222.

Vives, Xavier. 1996. Social learning and rational expectations. European Economic Review
40(3-5):589–601.

Wheaton, William C. and Mark J. Lewis. 2002. Urban wages and labor market agglomer-
ation. Journal of Urban Economics 51(3):542–562.

19