The structure of European development Community detection in - - PowerPoint PPT Presentation

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

The structure of European development — Community detection in inter-industry and external trade networks

Nadia Garbellini Ariel Luis Wirkierman

Universit` a degli Studi di Pavia (nadia.garbellini@unipv.it) Universit` a Cattolica, Milano (ariwirkierman@gmail.com)

April 25, 2012

Final WIOD Conference: Causes and Consequences of Globalization Groningen (The Netherlands), April 24-26, 2012

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Motivation and Objective:

Community detection in networks: Clustering of industries and countries in inter-industry and trade matrices Cluster-based analysis and policy in EU: European Cluster Observatory Operational definition, identification and measurement of industry ‘clusters’ in EU regions (EC 2008) Method: ‘locational employment correlation coefficients’ between industry couples across regions, grouping activities nearly always geographically associated A uniform clustering template is ‘imposed’ on regional data Objective: Clustering methodology for I-O/Trade networks Data mining on national and EU27/EMU Input-Output networks: country-specific industry composition of clusters, measure (dis-)similarity between clustering structures Data mining on intra-EU external commodity-specific trade networks: patterns of persistent EU trade partners

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Motivation and Objective:

Community detection in networks: Clustering of industries and countries in inter-industry and trade matrices Cluster-based analysis and policy in EU: European Cluster Observatory Operational definition, identification and measurement of industry ‘clusters’ in EU regions (EC 2008) Method: ‘locational employment correlation coefficients’ between industry couples across regions, grouping activities nearly always geographically associated A uniform clustering template is ‘imposed’ on regional data Objective: Clustering methodology for I-O/Trade networks Data mining on national and EU27/EMU Input-Output networks: country-specific industry composition of clusters, measure (dis-)similarity between clustering structures Data mining on intra-EU external commodity-specific trade networks: patterns of persistent EU trade partners

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Motivation and Objective:

Community detection in networks: Clustering of industries and countries in inter-industry and trade matrices Cluster-based analysis and policy in EU: European Cluster Observatory Operational definition, identification and measurement of industry ‘clusters’ in EU regions (EC 2008) Method: ‘locational employment correlation coefficients’ between industry couples across regions, grouping activities nearly always geographically associated A uniform clustering template is ‘imposed’ on regional data Objective: Clustering methodology for I-O/Trade networks Data mining on national and EU27/EMU Input-Output networks: country-specific industry composition of clusters, measure (dis-)similarity between clustering structures Data mining on intra-EU external commodity-specific trade networks: patterns of persistent EU trade partners

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Motivation and Objective:

Community detection in networks: Clustering of industries and countries in inter-industry and trade matrices Cluster-based analysis and policy in EU: European Cluster Observatory Operational definition, identification and measurement of industry ‘clusters’ in EU regions (EC 2008) Method: ‘locational employment correlation coefficients’ between industry couples across regions, grouping activities nearly always geographically associated A uniform clustering template is ‘imposed’ on regional data Objective: Clustering methodology for I-O/Trade networks Data mining on national and EU27/EMU Input-Output networks: country-specific industry composition of clusters, measure (dis-)similarity between clustering structures Data mining on intra-EU external commodity-specific trade networks: patterns of persistent EU trade partners

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Motivation and Objective:

Community detection in networks: Clustering of industries and countries in inter-industry and trade matrices Cluster-based analysis and policy in EU: European Cluster Observatory Operational definition, identification and measurement of industry ‘clusters’ in EU regions (EC 2008) Method: ‘locational employment correlation coefficients’ between industry couples across regions, grouping activities nearly always geographically associated A uniform clustering template is ‘imposed’ on regional data Objective: Clustering methodology for I-O/Trade networks Data mining on national and EU27/EMU Input-Output networks: country-specific industry composition of clusters, measure (dis-)similarity between clustering structures Data mining on intra-EU external commodity-specific trade networks: patterns of persistent EU trade partners

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Motivation and Objective:

Community detection in networks: Clustering of industries and countries in inter-industry and trade matrices Cluster-based analysis and policy in EU: European Cluster Observatory Operational definition, identification and measurement of industry ‘clusters’ in EU regions (EC 2008) Method: ‘locational employment correlation coefficients’ between industry couples across regions, grouping activities nearly always geographically associated A uniform clustering template is ‘imposed’ on regional data Objective: Clustering methodology for I-O/Trade networks Data mining on national and EU27/EMU Input-Output networks: country-specific industry composition of clusters, measure (dis-)similarity between clustering structures Data mining on intra-EU external commodity-specific trade networks: patterns of persistent EU trade partners

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Motivation and Objective:

Community detection in networks: Clustering of industries and countries in inter-industry and trade matrices Cluster-based analysis and policy in EU: European Cluster Observatory Operational definition, identification and measurement of industry ‘clusters’ in EU regions (EC 2008) Method: ‘locational employment correlation coefficients’ between industry couples across regions, grouping activities nearly always geographically associated A uniform clustering template is ‘imposed’ on regional data Objective: Clustering methodology for I-O/Trade networks Data mining on national and EU27/EMU Input-Output networks: country-specific industry composition of clusters, measure (dis-)similarity between clustering structures Data mining on intra-EU external commodity-specific trade networks: patterns of persistent EU trade partners

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Motivation and Objective:

Community detection in networks: Clustering of industries and countries in inter-industry and trade matrices Cluster-based analysis and policy in EU: European Cluster Observatory Operational definition, identification and measurement of industry ‘clusters’ in EU regions (EC 2008) Method: ‘locational employment correlation coefficients’ between industry couples across regions, grouping activities nearly always geographically associated A uniform clustering template is ‘imposed’ on regional data Objective: Clustering methodology for I-O/Trade networks Data mining on national and EU27/EMU Input-Output networks: country-specific industry composition of clusters, measure (dis-)similarity between clustering structures Data mining on intra-EU external commodity-specific trade networks: patterns of persistent EU trade partners

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Motivation and Objective:

Community detection in networks: Clustering of industries and countries in inter-industry and trade matrices Cluster-based analysis and policy in EU: European Cluster Observatory Operational definition, identification and measurement of industry ‘clusters’ in EU regions (EC 2008) Method: ‘locational employment correlation coefficients’ between industry couples across regions, grouping activities nearly always geographically associated A uniform clustering template is ‘imposed’ on regional data Objective: Clustering methodology for I-O/Trade networks Data mining on national and EU27/EMU Input-Output networks: country-specific industry composition of clusters, measure (dis-)similarity between clustering structures Data mining on intra-EU external commodity-specific trade networks: patterns of persistent EU trade partners

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Motivation and Objective:

Community detection in networks: Clustering of industries and countries in inter-industry and trade matrices Cluster-based analysis and policy in EU: European Cluster Observatory Operational definition, identification and measurement of industry ‘clusters’ in EU regions (EC 2008) Method: ‘locational employment correlation coefficients’ between industry couples across regions, grouping activities nearly always geographically associated A uniform clustering template is ‘imposed’ on regional data Objective: Clustering methodology for I-O/Trade networks Data mining on national and EU27/EMU Input-Output networks: country-specific industry composition of clusters, measure (dis-)similarity between clustering structures Data mining on intra-EU external commodity-specific trade networks: patterns of persistent EU trade partners

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Dataset characteristics

Input-Output networks: EUROSTAT (2011) database Supply-Use Tables for EU27 and EMU aggregates for 2000 and 2007 Industry Classification: NACE Rev. 1, 2 digits Fixed product sales structure assumption: 59 × 59 industry tables for domestic output at basic prices National Input-Output Tables of 22 EU27 countries for 2005 (excluded: BG, CY, LU, LV, MT) intra-EU trade networks: EUROSTAT (2006) External Trade (Comext) Statistical regime 1: normal imports of goods for final use in the EU (excludes inward/outward processing) 29 commodity-specific 2 digit CPA trade matrices 27 countries × 27 countries trade matrices for 2007-2010

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Dataset characteristics

Input-Output networks: EUROSTAT (2011) database Supply-Use Tables for EU27 and EMU aggregates for 2000 and 2007 Industry Classification: NACE Rev. 1, 2 digits Fixed product sales structure assumption: 59 × 59 industry tables for domestic output at basic prices National Input-Output Tables of 22 EU27 countries for 2005 (excluded: BG, CY, LU, LV, MT) intra-EU trade networks: EUROSTAT (2006) External Trade (Comext) Statistical regime 1: normal imports of goods for final use in the EU (excludes inward/outward processing) 29 commodity-specific 2 digit CPA trade matrices 27 countries × 27 countries trade matrices for 2007-2010

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Dataset characteristics

Input-Output networks: EUROSTAT (2011) database Supply-Use Tables for EU27 and EMU aggregates for 2000 and 2007 Industry Classification: NACE Rev. 1, 2 digits Fixed product sales structure assumption: 59 × 59 industry tables for domestic output at basic prices National Input-Output Tables of 22 EU27 countries for 2005 (excluded: BG, CY, LU, LV, MT) intra-EU trade networks: EUROSTAT (2006) External Trade (Comext) Statistical regime 1: normal imports of goods for final use in the EU (excludes inward/outward processing) 29 commodity-specific 2 digit CPA trade matrices 27 countries × 27 countries trade matrices for 2007-2010

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Dataset characteristics

Input-Output networks: EUROSTAT (2011) database Supply-Use Tables for EU27 and EMU aggregates for 2000 and 2007 Industry Classification: NACE Rev. 1, 2 digits Fixed product sales structure assumption: 59 × 59 industry tables for domestic output at basic prices National Input-Output Tables of 22 EU27 countries for 2005 (excluded: BG, CY, LU, LV, MT) intra-EU trade networks: EUROSTAT (2006) External Trade (Comext) Statistical regime 1: normal imports of goods for final use in the EU (excludes inward/outward processing) 29 commodity-specific 2 digit CPA trade matrices 27 countries × 27 countries trade matrices for 2007-2010

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Dataset characteristics

Input-Output networks: EUROSTAT (2011) database Supply-Use Tables for EU27 and EMU aggregates for 2000 and 2007 Industry Classification: NACE Rev. 1, 2 digits Fixed product sales structure assumption: 59 × 59 industry tables for domestic output at basic prices National Input-Output Tables of 22 EU27 countries for 2005 (excluded: BG, CY, LU, LV, MT) intra-EU trade networks: EUROSTAT (2006) External Trade (Comext) Statistical regime 1: normal imports of goods for final use in the EU (excludes inward/outward processing) 29 commodity-specific 2 digit CPA trade matrices 27 countries × 27 countries trade matrices for 2007-2010

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Dataset characteristics

Input-Output networks: EUROSTAT (2011) database Supply-Use Tables for EU27 and EMU aggregates for 2000 and 2007 Industry Classification: NACE Rev. 1, 2 digits Fixed product sales structure assumption: 59 × 59 industry tables for domestic output at basic prices National Input-Output Tables of 22 EU27 countries for 2005 (excluded: BG, CY, LU, LV, MT) intra-EU trade networks: EUROSTAT (2006) External Trade (Comext) Statistical regime 1: normal imports of goods for final use in the EU (excludes inward/outward processing) 29 commodity-specific 2 digit CPA trade matrices 27 countries × 27 countries trade matrices for 2007-2010

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Dataset characteristics

Input-Output networks: EUROSTAT (2011) database Supply-Use Tables for EU27 and EMU aggregates for 2000 and 2007 Industry Classification: NACE Rev. 1, 2 digits Fixed product sales structure assumption: 59 × 59 industry tables for domestic output at basic prices National Input-Output Tables of 22 EU27 countries for 2005 (excluded: BG, CY, LU, LV, MT) intra-EU trade networks: EUROSTAT (2006) External Trade (Comext) Statistical regime 1: normal imports of goods for final use in the EU (excludes inward/outward processing) 29 commodity-specific 2 digit CPA trade matrices 27 countries × 27 countries trade matrices for 2007-2010

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Dataset characteristics

Input-Output networks: EUROSTAT (2011) database Supply-Use Tables for EU27 and EMU aggregates for 2000 and 2007 Industry Classification: NACE Rev. 1, 2 digits Fixed product sales structure assumption: 59 × 59 industry tables for domestic output at basic prices National Input-Output Tables of 22 EU27 countries for 2005 (excluded: BG, CY, LU, LV, MT) intra-EU trade networks: EUROSTAT (2006) External Trade (Comext) Statistical regime 1: normal imports of goods for final use in the EU (excludes inward/outward processing) 29 commodity-specific 2 digit CPA trade matrices 27 countries × 27 countries trade matrices for 2007-2010

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Dataset characteristics

Input-Output networks: EUROSTAT (2011) database Supply-Use Tables for EU27 and EMU aggregates for 2000 and 2007 Industry Classification: NACE Rev. 1, 2 digits Fixed product sales structure assumption: 59 × 59 industry tables for domestic output at basic prices National Input-Output Tables of 22 EU27 countries for 2005 (excluded: BG, CY, LU, LV, MT) intra-EU trade networks: EUROSTAT (2006) External Trade (Comext) Statistical regime 1: normal imports of goods for final use in the EU (excludes inward/outward processing) 29 commodity-specific 2 digit CPA trade matrices 27 countries × 27 countries trade matrices for 2007-2010

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Dataset characteristics

Input-Output networks: EUROSTAT (2011) database Supply-Use Tables for EU27 and EMU aggregates for 2000 and 2007 Industry Classification: NACE Rev. 1, 2 digits Fixed product sales structure assumption: 59 × 59 industry tables for domestic output at basic prices National Input-Output Tables of 22 EU27 countries for 2005 (excluded: BG, CY, LU, LV, MT) intra-EU trade networks: EUROSTAT (2006) External Trade (Comext) Statistical regime 1: normal imports of goods for final use in the EU (excludes inward/outward processing) 29 commodity-specific 2 digit CPA trade matrices 27 countries × 27 countries trade matrices for 2007-2010

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Dataset characteristics

Input-Output networks: EUROSTAT (2011) database Supply-Use Tables for EU27 and EMU aggregates for 2000 and 2007 Industry Classification: NACE Rev. 1, 2 digits Fixed product sales structure assumption: 59 × 59 industry tables for domestic output at basic prices National Input-Output Tables of 22 EU27 countries for 2005 (excluded: BG, CY, LU, LV, MT) intra-EU trade networks: EUROSTAT (2006) External Trade (Comext) Statistical regime 1: normal imports of goods for final use in the EU (excludes inward/outward processing) 29 commodity-specific 2 digit CPA trade matrices 27 countries × 27 countries trade matrices for 2007-2010

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

The Method — Spectral Bisection and Similarity Measure

Network: n nodes, m edges, stored in a square (n × n) matrix W. In each community: mc nodes; mc(mc − 1) edges ec Spectral Bisection (SB) — due to Leicht & Newman (2008)

1

Divisive clustering method based on spectral decomposition of W

2

Communities are defined as indivisible subgraphs

3

Significant edges: edges above bi-proportional averages

4

Nodes mutually directly and indirectly connected are linked together

5

Returns a membership vector (n × 1) m Similarity Measure

1

Comparison of clustering structure of K homogeneous networks

2

Memberships stored in a membership matrix M = [mk], k ∈ [1, K]

3

Directed similarity between k and −k: weighted (by mc to total nodes) average of the fraction of ec,k also included in e−k

4

Undirected similarities: average of directed ones

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

The Method — Spectral Bisection and Similarity Measure

Network: n nodes, m edges, stored in a square (n × n) matrix W. In each community: mc nodes; mc(mc − 1) edges ec Spectral Bisection (SB) — due to Leicht & Newman (2008)

1

Divisive clustering method based on spectral decomposition of W

2

Communities are defined as indivisible subgraphs

3

Significant edges: edges above bi-proportional averages

4

Nodes mutually directly and indirectly connected are linked together

5

Returns a membership vector (n × 1) m Similarity Measure

1

Comparison of clustering structure of K homogeneous networks

2

Memberships stored in a membership matrix M = [mk], k ∈ [1, K]

3

Directed similarity between k and −k: weighted (by mc to total nodes) average of the fraction of ec,k also included in e−k

4

Undirected similarities: average of directed ones

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

The Method — Spectral Bisection and Similarity Measure

Network: n nodes, m edges, stored in a square (n × n) matrix W. In each community: mc nodes; mc(mc − 1) edges ec Spectral Bisection (SB) — due to Leicht & Newman (2008)

1

Divisive clustering method based on spectral decomposition of W

2

Communities are defined as indivisible subgraphs

3

Significant edges: edges above bi-proportional averages

4

Nodes mutually directly and indirectly connected are linked together

5

Returns a membership vector (n × 1) m Similarity Measure

1

Comparison of clustering structure of K homogeneous networks

2

Memberships stored in a membership matrix M = [mk], k ∈ [1, K]

3

Directed similarity between k and −k: weighted (by mc to total nodes) average of the fraction of ec,k also included in e−k

4

Undirected similarities: average of directed ones

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

The Method — Spectral Bisection and Similarity Measure

Network: n nodes, m edges, stored in a square (n × n) matrix W. In each community: mc nodes; mc(mc − 1) edges ec Spectral Bisection (SB) — due to Leicht & Newman (2008)

1

Divisive clustering method based on spectral decomposition of W

2

Communities are defined as indivisible subgraphs

3

Significant edges: edges above bi-proportional averages

4

Nodes mutually directly and indirectly connected are linked together

5

Returns a membership vector (n × 1) m Similarity Measure

1

Comparison of clustering structure of K homogeneous networks

2

Memberships stored in a membership matrix M = [mk], k ∈ [1, K]

3

Directed similarity between k and −k: weighted (by mc to total nodes) average of the fraction of ec,k also included in e−k

4

Undirected similarities: average of directed ones

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

The Method — Spectral Bisection and Similarity Measure

Network: n nodes, m edges, stored in a square (n × n) matrix W. In each community: mc nodes; mc(mc − 1) edges ec Spectral Bisection (SB) — due to Leicht & Newman (2008)

1

Divisive clustering method based on spectral decomposition of W

2

Communities are defined as indivisible subgraphs

3

Significant edges: edges above bi-proportional averages

4

Nodes mutually directly and indirectly connected are linked together

5

Returns a membership vector (n × 1) m Similarity Measure

1

Comparison of clustering structure of K homogeneous networks

2

Memberships stored in a membership matrix M = [mk], k ∈ [1, K]

3

Directed similarity between k and −k: weighted (by mc to total nodes) average of the fraction of ec,k also included in e−k

4

Undirected similarities: average of directed ones

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

The Method — Spectral Bisection and Similarity Measure

Network: n nodes, m edges, stored in a square (n × n) matrix W. In each community: mc nodes; mc(mc − 1) edges ec Spectral Bisection (SB) — due to Leicht & Newman (2008)

1

Divisive clustering method based on spectral decomposition of W

2

Communities are defined as indivisible subgraphs

3

Significant edges: edges above bi-proportional averages

4

Nodes mutually directly and indirectly connected are linked together

5

Returns a membership vector (n × 1) m Similarity Measure

1

Comparison of clustering structure of K homogeneous networks

2

Memberships stored in a membership matrix M = [mk], k ∈ [1, K]

3

Directed similarity between k and −k: weighted (by mc to total nodes) average of the fraction of ec,k also included in e−k

4

Undirected similarities: average of directed ones

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

The Method — Spectral Bisection and Similarity Measure

Network: n nodes, m edges, stored in a square (n × n) matrix W. In each community: mc nodes; mc(mc − 1) edges ec Spectral Bisection (SB) — due to Leicht & Newman (2008)

1

Divisive clustering method based on spectral decomposition of W

2

Communities are defined as indivisible subgraphs

3

Significant edges: edges above bi-proportional averages

4

Nodes mutually directly and indirectly connected are linked together

5

Returns a membership vector (n × 1) m Similarity Measure

1

Comparison of clustering structure of K homogeneous networks

2

Memberships stored in a membership matrix M = [mk], k ∈ [1, K]

3

Directed similarity between k and −k: weighted (by mc to total nodes) average of the fraction of ec,k also included in e−k

4

Undirected similarities: average of directed ones

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

The Method — Spectral Bisection and Similarity Measure

Network: n nodes, m edges, stored in a square (n × n) matrix W. In each community: mc nodes; mc(mc − 1) edges ec Spectral Bisection (SB) — due to Leicht & Newman (2008)

1

Divisive clustering method based on spectral decomposition of W

2

Communities are defined as indivisible subgraphs

3

Significant edges: edges above bi-proportional averages

4

Nodes mutually directly and indirectly connected are linked together

5

Returns a membership vector (n × 1) m Similarity Measure

1

Comparison of clustering structure of K homogeneous networks

2

Memberships stored in a membership matrix M = [mk], k ∈ [1, K]

3

Directed similarity between k and −k: weighted (by mc to total nodes) average of the fraction of ec,k also included in e−k

4

Undirected similarities: average of directed ones

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

The Method — Spectral Bisection and Similarity Measure

Network: n nodes, m edges, stored in a square (n × n) matrix W. In each community: mc nodes; mc(mc − 1) edges ec Spectral Bisection (SB) — due to Leicht & Newman (2008)

1

Divisive clustering method based on spectral decomposition of W

2

Communities are defined as indivisible subgraphs

3

Significant edges: edges above bi-proportional averages

4

Nodes mutually directly and indirectly connected are linked together

5

Returns a membership vector (n × 1) m Similarity Measure

1

Comparison of clustering structure of K homogeneous networks

2

Memberships stored in a membership matrix M = [mk], k ∈ [1, K]

3

Directed similarity between k and −k: weighted (by mc to total nodes) average of the fraction of ec,k also included in e−k

4

Undirected similarities: average of directed ones

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

The Method — Spectral Bisection and Similarity Measure

Network: n nodes, m edges, stored in a square (n × n) matrix W. In each community: mc nodes; mc(mc − 1) edges ec Spectral Bisection (SB) — due to Leicht & Newman (2008)

1

Divisive clustering method based on spectral decomposition of W

2

Communities are defined as indivisible subgraphs

3

Significant edges: edges above bi-proportional averages

4

Nodes mutually directly and indirectly connected are linked together

5

Returns a membership vector (n × 1) m Similarity Measure

1

Comparison of clustering structure of K homogeneous networks

2

Memberships stored in a membership matrix M = [mk], k ∈ [1, K]

3

Directed similarity between k and −k: weighted (by mc to total nodes) average of the fraction of ec,k also included in e−k

4

Undirected similarities: average of directed ones

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

The Method — Spectral Bisection and Similarity Measure

Network: n nodes, m edges, stored in a square (n × n) matrix W. In each community: mc nodes; mc(mc − 1) edges ec Spectral Bisection (SB) — due to Leicht & Newman (2008)

1

Divisive clustering method based on spectral decomposition of W

2

Communities are defined as indivisible subgraphs

3

Significant edges: edges above bi-proportional averages

4

Nodes mutually directly and indirectly connected are linked together

5

Returns a membership vector (n × 1) m Similarity Measure

1

Comparison of clustering structure of K homogeneous networks

2

Memberships stored in a membership matrix M = [mk], k ∈ [1, K]

3

Directed similarity between k and −k: weighted (by mc to total nodes) average of the fraction of ec,k also included in e−k

4

Undirected similarities: average of directed ones

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

The Method — Spectral Bisection and Similarity Measure

Network: n nodes, m edges, stored in a square (n × n) matrix W. In each community: mc nodes; mc(mc − 1) edges ec Spectral Bisection (SB) — due to Leicht & Newman (2008)

1

Divisive clustering method based on spectral decomposition of W

2

Communities are defined as indivisible subgraphs

3

Significant edges: edges above bi-proportional averages

4

Nodes mutually directly and indirectly connected are linked together

5

Returns a membership vector (n × 1) m Similarity Measure

1

Comparison of clustering structure of K homogeneous networks

2

Memberships stored in a membership matrix M = [mk], k ∈ [1, K]

3

Directed similarity between k and −k: weighted (by mc to total nodes) average of the fraction of ec,k also included in e−k

4

Undirected similarities: average of directed ones

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

The Method — Spectral Bisection and Similarity Measure

Network: n nodes, m edges, stored in a square (n × n) matrix W. In each community: mc nodes; mc(mc − 1) edges ec Spectral Bisection (SB) — due to Leicht & Newman (2008)

1

Divisive clustering method based on spectral decomposition of W

2

Communities are defined as indivisible subgraphs

3

Significant edges: edges above bi-proportional averages

4

Nodes mutually directly and indirectly connected are linked together

5

Returns a membership vector (n × 1) m Similarity Measure

1

Comparison of clustering structure of K homogeneous networks

2

Memberships stored in a membership matrix M = [mk], k ∈ [1, K]

3

Directed similarity between k and −k: weighted (by mc to total nodes) average of the fraction of ec,k also included in e−k

4

Undirected similarities: average of directed ones

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

The Method — Spectral Bisection and Similarity Measure

Network: n nodes, m edges, stored in a square (n × n) matrix W. In each community: mc nodes; mc(mc − 1) edges ec Spectral Bisection (SB) — due to Leicht & Newman (2008)

1

Divisive clustering method based on spectral decomposition of W

2

Communities are defined as indivisible subgraphs

3

Significant edges: edges above bi-proportional averages

4

Nodes mutually directly and indirectly connected are linked together

5

Returns a membership vector (n × 1) m Similarity Measure

1

Comparison of clustering structure of K homogeneous networks

2

Memberships stored in a membership matrix M = [mk], k ∈ [1, K]

3

Directed similarity between k and −k: weighted (by mc to total nodes) average of the fraction of ec,k also included in e−k

4

Undirected similarities: average of directed ones

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

The Method — Spectral Bisection and Similarity Measure

Network: n nodes, m edges, stored in a square (n × n) matrix W. In each community: mc nodes; mc(mc − 1) edges ec Spectral Bisection (SB) — due to Leicht & Newman (2008)

1

Divisive clustering method based on spectral decomposition of W

2

Communities are defined as indivisible subgraphs

3

Significant edges: edges above bi-proportional averages

4

Nodes mutually directly and indirectly connected are linked together

5

Returns a membership vector (n × 1) m Similarity Measure

1

Comparison of clustering structure of K homogeneous networks

2

Memberships stored in a membership matrix M = [mk], k ∈ [1, K]

3

Directed similarity between k and −k: weighted (by mc to total nodes) average of the fraction of ec,k also included in e−k

4

Undirected similarities: average of directed ones

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Empirical Methodology Roadmap

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Empirical Methodology Roadmap

Figure: Community Detection in EMU/EU27 IO networks of interindustry flows

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Empirical Methodology Roadmap

Figure: Community Detection in EMU/EU27 IO networks of interindustry flows Figure: From I-O matrices to Similarity in Countries’ Clustering Structures

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Empirical Methodology Roadmap

Figure: Community Detection in EMU/EU27 IO networks of interindustry flows Figure: From I-O matrices to Similarity in Countries’ Clustering Structures Figure: From Trade matrices to Similarity in CPA Clustering Structures

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Community Detection in EU/EMU-IO Networks

Figure: Clustering structure of the EU27 IO networks (2000 and 2007)

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Community Detection in EU/EMU-IO Networks

Figure: Clustering structure of the EU27 IO networks (2000 and 2007)

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Key results — Evolution from 2000 to 2007

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Key results — Evolution from 2000 to 2007

Transport-Trade specific cluster, same composition in both country aggregates

Figure: Cluster Transport-Trade (EU27, 2007)

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Key results — Evolution from 2000 to 2007

Transport-Trade specific cluster, same composition in both country aggregates Pharma-Hi Tech cluster (spec. machinery, chemicals, R&D) in EU27 vs Heavy Machinery cluster

Figure: Cluster Transport-Trade (EU27, 2007) Figure: Cluster Pharma-Hi Tech (EU27, 2007)

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Key results — Evolution from 2000 to 2007

Transport-Trade specific cluster, same composition in both country aggregates Pharma-Hi Tech cluster (spec. machinery, chemicals, R&D) in EU27 vs Heavy Machinery cluster Services cluster. EU27: increasingly autonomous and

• interconnected. EMU: sharp

separation of the MC:Finance. Tertiarisation of European economies

Figure: Cluster Transport-Trade (EU27, 2007) Figure: Cluster Pharma-Hi Tech (EU27, 2007)

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Intermediate consumption and labour redistribution

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Intermediate consumption and labour redistribution

Cluster-based inter-industry proportions:

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Intermediate consumption and labour redistribution

Cluster-based inter-industry proportions: ⊲ Intra-cluster productive consumption: 49-53% of the value of total intermediate uses

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Intermediate consumption and labour redistribution

Cluster-based inter-industry proportions: ⊲ Intra-cluster productive consumption: 49-53% of the value of total intermediate uses Subsystem labour: Employment of an industry producing a final good plus employment from all its supporting industries

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Intermediate consumption and labour redistribution

Cluster-based inter-industry proportions: ⊲ Intra-cluster productive consumption: 49-53% of the value of total intermediate uses Subsystem labour: Employment of an industry producing a final good plus employment from all its supporting industries Labour redistribution: From industries to subsystems

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Intermediate consumption and labour redistribution

Cluster-based inter-industry proportions: ⊲ Intra-cluster productive consumption: 49-53% of the value of total intermediate uses Subsystem labour: Employment of an industry producing a final good plus employment from all its supporting industries Labour redistribution: From industries to subsystems New decomposition: flows between intra-/extra-cluster industries and intra-/extra-cluster subsystems

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Intermediate consumption and labour redistribution

Cluster-based inter-industry proportions: ⊲ Intra-cluster productive consumption: 49-53% of the value of total intermediate uses Subsystem labour: Employment of an industry producing a final good plus employment from all its supporting industries Labour redistribution: From industries to subsystems New decomposition: flows between intra-/extra-cluster industries and intra-/extra-cluster subsystems ⊲ Around 20-24% of total employment is redistributed between intra-cluster industries and intra-cluster subsystems

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Intermediate consumption and labour redistribution

Cluster-based inter-industry proportions: ⊲ Intra-cluster productive consumption: 49-53% of the value of total intermediate uses Subsystem labour: Employment of an industry producing a final good plus employment from all its supporting industries Labour redistribution: From industries to subsystems New decomposition: flows between intra-/extra-cluster industries and intra-/extra-cluster subsystems ⊲ Around 20-24% of total employment is redistributed between intra-cluster industries and intra-cluster subsystems ⊲ Around 23-24% of total employment is redistributed between intra-cluster and extra-cluster industries/subsystems

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Intermediate consumption and labour redistribution

Cluster-based inter-industry proportions: ⊲ Intra-cluster productive consumption: 49-53% of the value of total intermediate uses Subsystem labour: Employment of an industry producing a final good plus employment from all its supporting industries Labour redistribution: From industries to subsystems New decomposition: flows between intra-/extra-cluster industries and intra-/extra-cluster subsystems ⊲ Around 20-24% of total employment is redistributed between intra-cluster industries and intra-cluster subsystems ⊲ Around 23-24% of total employment is redistributed between intra-cluster and extra-cluster industries/subsystems ⊲ Around 30-33% of the labour from (to) intra-cluster industries (subsystems) remains in (comes from) intra-cluster subsystems (industries)

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Key results

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Key results

Figure: Country clusters (EU22, 2005)

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Key results

DE-RO-HU cluster: does not include a Mediterranean country; by removing Germany, Portugal (and Poland) enter

Figure: Country clusters (EU22, 2005)

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Key results

DE-RO-HU cluster: does not include a Mediterranean country; by removing Germany, Portugal (and Poland) enter CZ-SK still have similar clustering structure

Figure: Country clusters (EU22, 2005)

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Key results

DE-RO-HU cluster: does not include a Mediterranean country; by removing Germany, Portugal (and Poland) enter CZ-SK still have similar clustering structure Typologies corresponding to clusters of countries, playing symmetrical roles in different geographical areas

Figure: Country clusters (EU22, 2005)

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Key results

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Key results

Four commodity groups with similar trade pattern, all including well defined ‘commodity-chain’-subgroups; no correspondence with commodity groups found from IO tables

Figure: Trade-based commodity clusters (2007)

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Key results

Four commodity groups with similar trade pattern, all including well defined ‘commodity-chain’-subgroups; no correspondence with commodity groups found from IO tables Seven country groups persistently in the same cluster for at least 70% of the traded

• commodities. 24 EU27
• countries. Only FR-IT-MT do

not have persistent trade partners

Figure: Trade-based commodity clusters (2007) Figure: Persistently trading countries (2007)

Introduction Methodology Community Detection Labour and Subsystems From IO to Countries From Trade to Commodities

Bibliography

EC (2008). The Concept of Clusters and Cluster Policies and their Role for Competitiveness and Innovation: Main Statistical Results and Lessons Learned. Technical report, European Commission,

• Luxembourg. Commission Staff Working Document SEC (2008)

2637. EUROSTAT (2006). Statistics on the trading of goods — User guide. Technical report, Eurostat, Luxembourg. ——— (2011). Creating consolidated and aggregated EU27 Supply, Use and Input-Output Tables, adding environmental extensions (air emissions), and conducting Leontief-type modelling to approximate carbon and other ‘footprints’ of EU27 consumption for 2000 to 2006. Technical report, Eurostat, Luxembourg. Leicht, E. & Newman, M. (2008). Community Structure in Directed

• Networks. Phys. Rev. Lett., 100.