Upload
juan-mateos-garcia
View
255
Download
0
Embed Size (px)
Citation preview
Complex places for complex times
An analysis of the complexity of local
economies in the UK
Juan Mateos-Garcia ([email protected])
James Gardiner ([email protected] )
Birmingham, 16 November 2016
Renewed, urgent need to
understand the drivers of local
economic development
Demand pull: Political upheaval
linked to regional economic
disparities
Supply push: Policies to spur
regional economic
development (smart
specialisation, UK industrial
strategy)
Context
Literature
The ‘product’ space as a network of interrelated capabilities (Hausman & Hidalgo, 2011)
The idea of economic complexity
(EC) is receiving increasing
attention
Captures the industrial
composition of a country
Variety of the products it
exports, and their rarity
EC linked to GDP growth, lower
inequality. Why?
Jacobs spillovers
Resilience
Hausman & Hidalgo (2011), Hartmann et al (2015)
Complexity...is complicated
Video games Tourism PR Industry M
Coding Arts Capability NSelling stuffCapabilities
Industries
Economic complexity
Takes into account the uniqueness of sectors in a local economy.
Hypotehesised to generate benefits through knowledge spillovers,
economic creativity, hard to imitate comparative advantages, resilience
and diversity…
In this paper...
We follow the example of others who have measured complexity at the sub-
national level. We use industrial (not exports) data.1
Questions
1. What is the EC of UK local authority districts?
2. What is the link between EC and local economic outcomes?
3. What is the link between EC and local networking?
Q3 sets to further understand the mechanisms linking EC to economic outcomes:
are they relational or structural?
It’s also an opportunity to demo some interesting web data2 :-) his
Paper spins out from a data pilot for Arloesiadur, an innovation analytics project
for Welsh Government
1Cimini et al (2014), Antonelli et al (2016); 2 Mateos-Garcia & Bakhshi (2016)
Data sources & pipeline...also complicated
IDBR, APS, IO tables
Sets of similar industries
Specialisation matrix
APS, ASHE, UK Census, Electoral Commission
Economic / Political outcomes & controls
EC Measures (Q1)
Link EC-Outcomes (Q2)
Meetup.com
Activity segments
Activity, diversity, crossover
Link EC-Networking (Q3)
Stage 1: Similarity and complexity
IDBR, APS, IO tables
Sets of similar industries
Specialisation matrix
EC Measures (Q1)
IDBR: 2015 SIC4 business counts by LAD
APS: 2014 Distribution of SOCs over SICs
IO Tables: 2014 2-digit SIC
Look for sets of similar industries based on co-location, labour force
composition and trade using cluster segmentation algorithm by
Delgado et al.1 Goals are to reduce noise and capture diversity.
Measure industry clustering and apply HH method of reflections
algorithm to generate local measures of complexity and measures of
industry complexity
1 Delgado et al, 2016.
Similarity analysis
We have considered 162 combinations of variables and clustering algorithms
Best result (based on within-group similarities vs between group similarities and
robustness) = 64 industry segments
Not just a replication of SIC structure. 72% clusters contain SIC4s from different
SIC divisions.
Complexity analysis (places)
EC calculated using HH method of reflections: it recursively weights the industrial
profile of an area (diversity of industries it specialises on) by the ubiquity of those
industries.
Southern LADs have, on average, higher EC. We see big variations inside regions too.
Complexity analysis (industries)
The method of reflections
also identifies ‘rare’
industries that tend to be
part of highly diversified
economies
Knowledge intensive and creative sectors are a stronger feature of complex local economies.
Primary sectors, some low-tech manufactures and low-knowledge sectors appear in less complex economies
Stage 2: Economic aspects of complexityQ2: What is the link between EC and local economic outcomes?
The EC variable correlates with other area level measures of economic performance/knowledge intensity. The results don’t seem to be driven by urbanisation.
Multivariate analysis
Our multivariate analysis suggests a strong and
significant link between the EC index and annual earnings
even after controlling for level of education, presence of highly skilled workers and
region.
Stage 3: Relational aspects of EC
Web platform for the organisation of events. 28.3 million members
and 261,347 groups in 179 countries. We extract information about
3800 technology and business groups, and 379K members.
We use topic modelling to identify ‘segments’ of activity and classify
groups into them.
We use member co-participation in groups to measure networking
between segments. We calculate this in each LAD and normalise by
UK networking propensity.
Meetup.com
Activity segments
Activity, diversity, crossover
What are the mechanisms that link EC and improved economic outcomes? If high
EC places are more interconnected, this would lend weight to a urbanisation
economies hypothesis. We use web data to explore this idea.
What does this look like?
Location: we can map them
Topics: We can determine a
group’s segment
Members: We can measure levels of
activity, and determine crossover
between communities
Group segmentation
We use Latent Dirichlet Allocation (a text mining algorithm) to extract 60 topics
(segments) from our keywords, and allocate each group to its dominant segment.
This graph shows relationships between segments, based on member co-participation in
groups with different segments (normalised)
Local networking
In general, higher HC areas tend to have a stronger presence of meetup segments, and specially those which are rarer. There is a positive (although not very strong) correlation between complexity and measures of networking, specially segments (number of segments present in a LAD) and out_mix
(tendency to network across segments).
Multivariate analysis
Preliminary analysis suggests a link between EC and:
● The number of meetup segments in a LAD,
● Meetup groups as a share of all businesses, and
● Levels of networking inside and (more strongly) between meetup segments (a proxy for crossover between industries)
Going back to our questions...
1. What is the EC of UK local authority districts?
● London and the South tend to have more EC areas, but there are also EC
areas in other parts of the UK.
● Differences don’t seem to be purely driven by urbanisation
● High knowledge intensive services & creative sectors are a stronger
feature of EC areas.
2. What is the link between EC and local economic outcomes?
● Initial analysis suggests a link between EC and better economic outcomes.
3. What is the link between EC and local networking?
● Initial analysis suggests a link between high EC and:
○ Diversity in tech and business communities
○ Levels of networking (especially between communities)
Discussion
● The EC indicator contains interesting
information about a local economy
and its performance
● EC seems to have a relational
dimension: more complexity linked to
more opportunities for networking and
spillovers, could enable more
complexity.
● Our experimental web data helps us
start unpicking a mechanism linking
EC and local economic performance.
Policy implications
● Operations: Economic complexity could
be an interesting metric for local
economic developers. A bit complex
though… how can we start using it in
policy?
● Local industrial strategy: Complexity is
likely to be highly path-dependent. What
are the right policies to diversify less
complex economies?
● National industrial strategy: Economic
complexity appears to be a feature of
larger, urban local economies. What about
locations with other characteristics?
Limitations and next steps
● Data limitations
○ LADs are an administrative boundary: Use other functional
geographies
○ Industry similarity draws on business counts: Access employment
data
○ Industry cluster composition is noisy: Further manual QA.
○ Current impact measure is crude. Consider others.
● Analytical limitations
○ Multivariate analysis is cross-sectional. Make longitudinal.
○ MVA missing the spatial dimension: Add spatial weights.
○ Simplistic, linear modelling framework: Consider
interdependencies between variables, and nonlinearities.