Transcript Oort Pecs 2009 (1)

Empirical Methodologies to Research
Agglomeration Externalities
Frank van Oort
02-07-2009
1
Content
Empirical methodologies of agglomeration externalities
1.
Growth in cities
Spatial dependence
Spatial heterogeneity
Related variety
2.
Employment-population
Causality
3.
Scale issues
MAUP
Multilevel issues
4.
Contexts
Locations or networks
---------------------------------------------------------------------5.
Policy analysis
Indicators of clusters
2
Content
Empirical methodologies of agglomeration externalities
1.
Growth in cities
Spatial dependence
Spatial heterogeneity
Related variety
2.
Employment-population
Causality
3.
Scale issues
MAUP
Multilevel issues
4.
Contexts
Locations or networks
---------------------------------------------------------------------5.
Policy analysis
Indicators of clusters
3
Basic principles – externalities in cities
“My purpose is to show that cities are primary economic
organs” (Jacobs 1969, p.6).
“Development is a process of continuously improving in a
context that makes injecting improvisations feasible. Cities
create that context. Nothing else does” (Jacobs 1984, p.155).
“The city is not only the place where growth occurs, but also
the engine of growth itself” (Duranton 2000, p.291-292).
“Large cities have been and will continue to be an important
source of economic growth” (Quigly 1998, p.137).
“Agglomeration can be considered the territorial counterpart of
economic growth” (Fujita and Thisse 2002, p.389).
4
Growth and innovation
externalities
Spillovers
Agglomeration
Clusters
Regional Innovation
System
Knowledge Economy
Knowledge Production
Function
5
Knowledge spillovers are:
• unpaid externalities,
•
•
•
•
•
•
•
•
•
•
•
causes of (cumulative) economic growth,
causes of innovation-diffusion,
in urban contexts: agglomeration-economies,
according to NEG-economists, not measurable,
according to many, not measurable in space,
according to some, related to specialised,
diversified and urban localised industries,
according to many scale-free,
embedded in endogenous growth models,
embedded in evolutionary economic theory
often related to innovative firms,
knowledge institutions, growing firms or the
emergence of new firms,
often interpretable as location factors of firms (clusters),
and hence interesting for policy
6
Example: the Dutch ICT-sector
Spatial factors (externalities) that econometrically
relate to firm growth in ICT-firms
Theory: endogenous growth theory &
externalities, evolutionary economic geography
Hypotheses on agglomeration economies:
specialisation, diversity, competition (Glaeser et
al. 1992)
580 municipalities
7
Dynamic externalities
Glaeser et al. (1992) – Growth in cities. JPE.
Henderson et al. (1995) – Industrial development in
cities. JPE.
Their research results are highly suggestive for the
relevance of urban environments for economic
growth processes.
Debate:
localisation (Marshallian) economies
versus
urbanisation
(Jacobs)
economies
BT-BS (63%),
DV (9%),
OV (11%),
SCV (17%)
8
Agglomeration hypotheses
9
Many Empirical Studies, Different Findings
Source: Rosenthal & Strange (2004)
Lack of robustness across studies implies that different economies
can exist next to each other and that not per definition one type of
agglomeration externality leads to more concentration or economic
growth than the other.
10
ICT-sector Netherlands:
11
Localised spillovers: spatial dependence
Spatial proximity based clustering
also called spatial autocorrelation
or contiguous spatial dependence
Spatial regimes on urban structures
also called spatial heterogeneity or
non-contiguous spatial dependence
12
Spatial
proximity:
Moran 1 employment
function ICTfirms
(location
quotients
1996)
13
Spatial
proximity:
Moran 2 - log
employment
growth all
ICT-firms
(1996-2000)
14
Spatial
proximity:
Moran 3 - log
new firm
formation
(1996-2000)
15
urban
regimes 1:
national
zoning
16
urban
regime
2:
labourmarkets
17
urban
regimes 3:
Municipal
size
18
19
Conclusions: agglomeration hypotheses
Growth in the Dutch ICT sector tends to be concentrated in
urban areas that are already relatively specialized in this
sector and that are relatively rich in the presence of other
industries. These outcomes do not fully support or
contradict any of the four theories of localized knowledge
spillovers (MAR, Porter, Jacobs)
Spatial policy should not be restricted to the local or regional
environment alone. Spatial externalities relevant in local
contexts can be at work at higher levels. Local policy
makers should be open to the argument of spillover
effects from nearby (not necessarily adjacent)
agglomerations instead of promoting ‘own’ ICT-clusters
20
Summary Van Oort (2007)
21
Why, where and when does it matter?
Duranton and Puga (2000) paper
•
Specialized and diversified cities co-exist
•
Larger cities tend to be more diversified
•
The distribution of city-sizes and specializations tend to be
stable over time
•
City growth is related to specialization and diversity
•
Relocations are from diversified to specialized cities
•
Assumptions: crowding, agents, labour mobility,
(endogenous) self-organisation, path-dependency,
systems of cities (policentricity).
22
Do we measure knowledge spillovers?
•
Jacobs’ externalities  related variety
•
Space, knowledge and growth are more complex related
than often thought: spatial dependence, spatial
heterogeneity, scale, measurement units, timeframe,
definitions
•
Localization versus urbanization economies too simple?
•
Innovation as source of growth (KPF)
•
We did not measure transactions or linkages (yet)!
•
Micro-foundations of growth
•
Causality
•
MAUP
•
Contexts of firms (networks, sectors, location)
23
Related and unrelated variety
24
Variety and urban economic growth
• source externalities = localization:
similar firms & products, intra-industry,
incremental innovation -> productivity growth
• source externalities = Jacobs externalities:
inter-industry, radical innovation, new markets ->
employment growth
25
Hypotheses
1. Jacobs externalities are positively related to
employment growth
2. Localisation economies are positively related to
productivity growth
3. Unrelated variety is negatively related to
unemployment growth
Analysis:
• COROP (functional) regions
• Netherlands – natural control location factors
• 1996-2002 – base year approach
• control variables
• sensitivity analysis period
• standardised variables
• spatial econometrics (lag/error and regimes)
26
Related and unrelated variety
27
Employment growth
28
Conclusions
Hypothesis 1 (related/Jacobs/employment) – confirmed
Hypothesis 2 (localisation/productivity) – unconfirmed
(traditional)
Hypothesis 3 (unrelated/unemployment) – confirmed (but
sensitive)
Classic studies (Glaeser cs.) measure unrelated variety and
conclude on related variety
Agglomeration or cities (urbanisation) per se are not enough
for stimulating economic growth
Spatial regimes more important than spatial aurocorrelation
patterns
29
Content
Empirical methodologies of agglomeration externalities
1.
Growth in cities
Spatial dependence
Spatial heterogeneity
Related variety
2.
Employment-population
Causality
3.
Scale issues
MAUP
Multilevel issues
4.
Contexts
Locations or networks
---------------------------------------------------------------------5.
Policy analysis
Indicators of clusters
30
Employment-population dynamics
in the Dutch Randstad
Frank van Oort
Pecs, 2-7-2009
Urbanisation Act (1976, p.132)
“Little is known on the way spatial development and
social-economic development influence each other.
There exists uncertainty especially on the mutual
dependence of choices for places of living (of people)
and working (of firms)”
Introduction
 Employment-population dynamics:
 “Do people follow jobs, or do jobs follow people?”
 International research:
 Mixed results
 Limited Dutch research:
 “jobs follow people”
 Levels of analysis: NUTS3 and zip-codes
International research
 Marlon Boarnet (1994):
 “The monocentric model and employment location”
 Employment is endogenous to population changes
 Donald Steinnes (1977):
 “Causality and intraurban location”
 Causality runs from residence to employment
 Donald Steinness (1982):
 “Do people follow jobs” A causality issue in urb. ec.
 Measurement issues!!
Research questions
 Research question:
 What is the relation between population growth and
employment growth at the level of Dutch municipalities?
 Special attention for:
 Industrial composition
 Influence of policy
 Spatial differentiation: Randstad Holland
Why important for policy?
 Population-employment dynamics is primarily an issue on the municpal
level (Nota Ruime)
 ‘For the accommodation of space for population and employment over
Dutch municipalities, the Dutch government wants municipalities to offer
space for the existing population and its growth, as well as the existing
population of local firms and their growth potential’.
 ‘Other services related to spatial planning (governmental, retail, etc.) should
be accommodated timely and in the right numbers, related to the local
demand of citizens and entrepreneurs’.
 No policy regarding spatial heterogeneity:
 Does the Randstad encounter the same dynamics as municipalities in the
Intermediate Zone and in the National Periphery?
Model
•dPi,t =Xβ + dWEi,t + WEi,t-1 – λP Pi,t-1 + other P-factors
•dEi,t =Yδ + dWPi,t + WPi,t-1 – λE Ei,t-1 + other E-factors
What follows what?
Jobs follow People
 cf. earlier research
 But:
 Especially personal
services follow popualtion
 Industrial sectors and
business services to a
much lesser extent
 Distribution does not
follow poulation
 C.P. other local growth
factors population
(environment) and jobs
growth (clusters)
Spatial differentiation
Spatial differentiation
Spatial differentiation
 National population-employment dynamics is
mainly determined by dynamics in the Randstad
(‘jobs follow people’)
 In the Intermediate Zone complex dynamics
(‘jobs follow people’ and ‘people follow jobs’)
 Little dynamics in the National periphery (and
when so, ‘people follow jobs’)
Conclusions
 In general:
 ‘jobs follow people’
 However, differences for industries, policy and
density zones:
 Personal service jobs follow people
 Jobs in the North-wing of the Randstad follow people
 In VINEX-municipalities jobs follow people
 Population-employment dynamics in the Netherlands:
 Restrictions on location choice of people (Vinex
municipalities in the North-wing Randstad)
 Outside the Randstad more opportunities for choice
(where people follow jobs more easily)
Conclusions
Dutch jobs follow people; but only
because the latter have no choice
Policy implications
• Attracting population outside the North-wing of the
Randstad is no guarantee for job growth.
• In the National Periphery, policy aiming at only
employment growth, e.g. by providing business sites,
only to a very limited (to no) extent leads to
population growth. Policy aimed at only population
growth does not lead to employment growth.
• How about shrinking regions!
Content
Empirical methodologies of agglomeration externalities
1.
Growth in cities
Spatial dependence
Spatial heterogeneity
Related variety
2.
Employment-population
Causality
3.
Scale issues
MAUP
Multilevel issues
4.
Contexts
Locations or networks
---------------------------------------------------------------------5.
Policy analysis
Indicators of clusters
45
Scale-dependency
• Traditional: functional region unit of observation
• The choice of this level as spatial unit of analysis
is arbitrary and foremost a result of data
limitations.
• Other problems:
– Agglomeration externalities may well reach
beyond the regional level or be present at a
lower scale
– Most often agglomeration externalities are
treated as spatially fixed (agglomeration
externalities as club good); this is
unsatisfactory
46
Analysis
• Accordingly, the spatial scope of agglomeration
externalities remains opaque
• Moreover, their effects seem to depend on the
spatial scale they are studied
• It is the geographical scale and scope of
agglomeration externalities that will be the focus of
our analysis (holding sector, time, area, and
measurement constant)
47
Data
• Aggregated plant-level data on employment
(N=647000, 5-digit sectors)
• Three spatial levels of analysis
– Neighborhood (N=3957, +/- 9 km2)
– Municipality (N=483, +/- 70 km2)
– Functional Region (N=40, +/- 850 km2)
• Spatial autoregression to evaluate scale and
scope of agglomeration externalities
48
Very preliminary outcomes
Dependent Variable: Absolute Employment Growth (1996-2004),
estimated with constant, control variables (wage, competition,
investments), controlled for fixed and random effects, and spatial
dependency. To control for endogeneity, we used lagged levels of
past conditions.
Textiles, Apparel & Leather Industry
Municipality
Functional Region
Localization Economies
++
--
Urbanization Economies
--
+
Jacobsean Economies
0
0
Municipality
Functional Region
Localization Economies
++
--
Urbanization Economies
--
+
Jacobsean Economies
0
0
Consumer Electronics
49
Synthesis
If agglomeration externalities turn out to be scaleand scope-dependent:
• Question the external validity of past studies on
agglomeration externalities
• Focus on the micro-foundations of agglomeration
externalities
• Take the firm or plant seriously in analysis by
taking it as unit of analysis.
50
Content
Empirical methodologies of agglomeration externalities
1.
Growth in cities
Spatial dependence
Spatial heterogeneity
Related variety
2.
Employment-population
Causality
3.
Scale issues
MAUP
Multilevel issues
4.
Contexts
Locations or networks
---------------------------------------------------------------------5.
Policy analysis
Indicators of clusters
51
Agglomeration economies and the
performance of new firms
Frank van Oort, Utrecht University
Martijn Burger, Erasmus University Rotterdam
Pecs, 2-7-2009
Agglomeration economies: definition
• Dates back to the work of Marshall (1890), Hoover
(1948) and Isard (1956).
• Cost-saving benefits or productivity gains external to a
firm, from which a firm can benefit by being located at
the same place as one or more other firms.
• Uncontrollable for a single firm.
• Immobile or spatially constrained.
Agglomeration economies: definition
• Localization Economies
– Agglomeration economies stemming from
concentration of specialized economic
activities, internal to industry
• Urbanization Economies
– Agglomeration economies stemming from
market size, external to industry
Empirical literature on agglomeration
• Empirical studies show that the elasticity of productivity
to city and local industry size typically ranges between
3% and 8% (Rosenthal and Strange, 2004)
• However, results vary over the sectors, regions and time
period under observation.
• Agglomeration externalities may differ with respect to
their reach and the scale on which they are present.
The firm in Agglomeration Economics
• At the same time, little is known about the importance of
agglomeration economies for the performance of firms.
• Most empirical research on agglomeration uses
aggregated data with cities or city-industries as basic
reference unit.
• These studies provide only limited insights and weak
support for the effect of agglomeration economies on
firm performance due to ecological fallacy and spatial
selection.
The firm in Agglomeration Economics
• Lack of firm-level evidence in the agglomeration
economics literature can mainly be ascribed to data
limitations and confidentiality restrictions
• Absence is nevertheless ‘disturbing’ because the
theories that underlie agglomeration economies are
microeconomic in nature
• Agglomeration economies do not directly foster regional
economic growth, but only indirectly through their effect
on firm performance.
The firm in Agglomeration Economics
•Regional
Circumstances
•Regional
Economic Growth
•1
•3
•2
•Firm
•Orientations
• Firm
Performance
•
The region in which a firm is embedded generates opportunities and economic
constraints for firms located in that region, e.g. through agglomeration economies
and agglomeration diseconomies.
•
Firms with more economic opportunities and less economic constraints
(proposition 1) tend to perform better in terms of survival chances and productivity
growth.
•
Regions with better performing firms (proposition 1 and 2) exhibit higher economic
growth. Regional performance is here conceptualised as the weighted sum of the
firms’ performances.
The firm in Agglomeration Economics
• Regional growth is a byproduct of interest-seeking firms
trying to optimise their own performance
• The external environment of the firm not only consists of
the location of the firm (physical environment), but also
of other components such as the sector in which the firm
is embedded
• Not all opportunities and constraints of firms are related
to macro-level properties
• However, even when constraints and resources are firmbased, it often remains debatable to what extent their
effect is independent of region and/or sector
From theory to empirics: questions
• We need a model that looks both at micro and macro-level
characteristics when the link between micro and macro levels is
existent – is that the case?
• New types of general questions:
– How is important is the external environment of firms for firm
performance?
– Which types of firms draw more on their external environment?
• New types of more specific questions:
– How important are agglomeration economies for firm
performance?
– Which types of firms profit more from agglomeration economies?
From theory to practice: empirical model
• One way to tackle these questions is by means of
random effects (multilevel analysis), which offers a
natural way to assess to what extent a link between the
micro-level and macro-level is existent
• Two important assets of multilevel analysis:
– Modelling contextuality by means of variance
partitioning
– Modelling heterogeneity by allowing relations to
vary across environments (random slopes).
From theory to practice: empirical model
• We focus on a model of firm survival and employment
growth that is specific to characteristics of the internal
and external environment of the firm
• This external environment may consists of several
components, such as the firm’s location, sector or club
(location-by-sector)
• These environments should simultaneously be assessed
in order to avoid underspecification of the model
• Main interest: effect of agglomeration economies on firm
survival and employment growth.
Data
• Data from employment register (LISA) on of 46,000 new
firms in 2000/2001 in the advanced producer services
sector in the Netherlands
• New firms less constrained by previous decisions, which
influences how they value the marginal worker and
whether new employment is created
• Less influence of a possible spatial sorting effect
Data
• These 46,000 firms are divided over 40 regions (LMA’s)
and 19 subsectors in the advanced producer services
• Dependent variables: firm survival (yes/no) and
employment growth (yes/no) in the first five years of a
firm’s existence
• Independent variables related to the internal and
external environment of the firm
Variables
• Firm: firm survival and employment growth, initial firm
size
• Region-by-Sector (Club): localization economies (own
sector density), competition (turnover)
• Regional: urbanization economies (population density),
R&D expenditures, human capital
• Fixed effects at the sector level
From theory to practice: model
•Regions
•(k1)
•Sectors
•(k2)
•Regional
Sectors (j)
•Mixed hierarchical and
cross-classified probit
model
•DV: Survival /
Employment growth (1 if
Yes)
•Four classifications:
- Firm
- Regions
•New
•Firms (i)
- Sectors
- Regions-by-Sectors
(Club)
(3)
probit ( ij ( k1,k 2) )  X ij ( k1,k 2)   u0(1)j  v0(2)

v
k1
0 k 2  e0ij ( k1,k 2)
•Intercept reflecting
average probability of
firm survival
•Differential
intercepts for clubs,
regions and sectors;
higher level residuals
•Remaining firm
differential; firm
level residual
•Notation: ith firm in the jth club, that is nested in region k1 and sector k2
Model estimation
- Three-level probit model (with four classifications) with a
random intercept for firms at the lowest level and random
intercepts for regions, sectors-by-regions, and sectors at
the higher levels
- Estimated by means of Restricted Iterated General Least
Squares using the MLWIN statistical package
- For survival and new firm employment growth models a
Mundlak correction to account for endogeneity bias
Model estimation
- First analysis: gives an indication to what extent location
matters by explicitly disentangling the between location
variance from the between firm and between sector
variance
- Second analysis: assesses the effect of agglomeration
economies on new firm survival and growth
- Third analysis: assesses whether the effect of
agglomeration economies on new firm survival and
growth varies across firms
Variance Partition Coefficient
• Variance partition coefficients (VPC) can say something
about the importance of the context
• VPC measures the extent to which the y-values of new
firms in the same club/region/sector resemble each other
as compared to those from new firms in different
clubs/regions/sector
• It may also be interpreted as the proportion of the total
residual variation that is due to differences between
clubs/regions/sectors
VPC’s - survival
•3.3
%
•Regions
•(k1)
•1.3
%
•Sectors
•(k2)
•Regional Sectors /
Clubs (j)
2
VPC j   (1)
u0 j
•New
•Firms (i)
•
•
•
•
•4.5
%
•90.9
%
2
/( (1)
u0 j
2
  (2)
v0 k1
2
  (3)
v0 k 2
 1)
•Probit distribution for
the firm-level residual
implies a variance of 1
90.9% of the variance is between-firm variance.
1.3% of the variance is between-club variance
3.2% of the variance is between-location variance
4.5% of the variance is between-sector variance
VPC’s – employment growth
•2.5
%
•Regions
•(k1)
•0.8
%
•Sectors
•(k2)
•Regional Sectors /
Clubs (j)
•New
•Firms (i)
•
•
•
•
•3.0
%
2
VPC j   (1)
u0 j
•93.7
%
2
/( (1)
u0 j
2
  (2)
v0 k1
2
  (3)
v0 k 2
 1)
•Probit distribution for
the firm-level residual
implies a variance of 1
90.9% of the variance is between-firm variance.
1.3% of the variance is between-club variance
3.2% of the variance is between-location variance
4.5% of the variance is between-sector variance
Adding predictor variables
• As yet, we only have partitioned the variability in survival
and employment growth of new firms over areas,
sectors-by-areas, sectors, and firms
• However, we can add predictor variables for these
classifications, in order to see to what extent they explain
the partitioned variability
• The predictors we add contain measures related to the
firm characteristics, agglomeration externalities and
sectoral externalities
Adding predictor variables
•Regions
•(k1)
•Sectors
•(k2)
•Mixed hierarchical and
cross-classified probit
model with
• firm-level variables
•Regional
Sectors (j)
• club-level variables
• regional-level variables
• sector-dummies
•New
•Firms (i)
q
probit ( ijk1 )  X ijk1 0   10 X pijk1    q 0 X q jk1 
j 1
v1k1 X pijk1  u0 jk1  v0k1
r
 r 0 X rk1   k 2  u1 jk1 X pijk1
k11
Empirical results
• Localization economies have small positive effect on
new firm survival, while it has no impact on employment
growth
• Urbanization economies have a positive effect on both
the survival opportunities and employment growth of new
firms
Adding cross-level interactions
•Regions
•(k1)
•Sectors
•(k2)
•Mixed hierarchical and crossclassified probit model with
• firm-level variables
• club-level variables
•Regional
Sectors (j)
• regional-level variables
• sector-dummies
• cross-level interactions
between firm size and different
types of agglomeration
economies
•New
•Firms (i)
q
probit ( ijk1 )  X ijk10   10 X1ijk1    q0 X q jk1 
j 1
r
r
q
k11
j 1
 r 0 X rk1   k 2   q10 X1ijk1 X qjk1 
 r10 X1ijk1 X rk1  u1 jk1 X1ijk1  v1k1 X1ijk1  u0 jk1  v0k1
k11
Empirical results
• With respect to surviving, larger start-ups profit from
localisation economies and urbanisation economies and
not their smaller counterparts
• With respect to employment growth, smaller start-ups
profit more from localisation economies, and not their
larger counterparts.