Transcript Knowledge flows across European regions

Knowledge flows across
European regions
Raffaele Paci and Stefano Usai
CRENoS, University of Cagliari
forthcoming in Annals of Regional Science
EU FP7 SSH Project: Intangible Assets and Regional Economic Growth
Università di Palermo, 10 aprile 2008
Motivations /1
• Many economists have attempted to find evidence
of the existence of knowledge spillovers or flows
either embodied in R&D exchanges, in bilateral
trade, in capital goods acquisition or in foreign
direct investment….
• These indicators are mainly indirect since
technological flows, at the interregional and
international levels, are hard to be captured:
Krugman (1991) “knowledge flows are invisible
and cannot be measured and tracked”
Motivations /2
• Jaffe et al (1993): knowledge flows may leave a
“paper trail”, in the form of patent citations, which
can be measured and used to obtain information on
the geographical component of the innovation
spillover mechanism
• Citations made are a measure of previous
knowledge extracted from the cited patent and
embodied in the new invention.
• They allow us to represent this knowledge flow in
the geographical space (by using the inventors’
residence) to have an idea of the knowledge links
(network) among cited and citing regions
Aims
• Contribute to the analysis of knowledge flows and
their determinants across European regions
• Examine whether geographical distance and
spatial contiguity influence knowledge exchanges:
i.e. are knowledge spillovers locally bounded ?
• Control for the role of other types of “distances”:
production structure, economic conditions,
technological efforts, national borders
• Investigate on the changes along time
Background literature
• Seminal contributions on the USPTO database:
– Jaffe et al, 1993
– Jaffe and Trajtemberg, 1996 and 1998
• Some more recent works on EPO
– Maurseth and Verspagen, 1999 and 2002
– Lukatch and Plasmans, 2003
– Breschi and Lissoni, 2004
– Maggioni et al., 2005
– LeSage et al., 2006
CRENOS dataset
• Patents granted by EPO, 1978 – 2004
• Each patent attributed to one of 175 regions of 17
countries in Europe, the 15 members of the EU15 plus
Switzerland and Norway (based on inventors –single and
multiple- and not firms HQs).
• Citations are linked to patents from both a geographical
and an industrial point of view.
• Each patent is classified by industrial sector (3 digit)
based either on the IPC - ISIC Yale Technology
or on the Schmock-OECD concordance. Results on
sectors are preliminary and are not reported here.
• Note: citation list (both at EPO, USPTO) is completed by
the examiner during the granting procedure
Table 1. International flows of patent citations
(only patents granted by EPO to inventors resident in the 17 European countries)
1990
1998
Patents granted
19.987
18.659
Citations
77.463
72.804
% shares
% shares
24,2
24,3
17 countries
15,0
14,6
other countries
9,3
9,7
European national patent offices
45,7
33,7
USA
25,6
30,8
Japan
0,3
0,3
Rest of the world
4,1
11,0
100,0
100,0
Citations to patents granted by:
EPO
Total
Table 2. Geographical distribution of patent citations
1990
1998
11.589
10.606
% shares
% shares
28,6
21,9
- contiguous regions
9,3
9,7
- not contiguous regions
15,4
16,2
International interregional
46,7
52,1
Total
100,0
100,0
Citations by EPO to EPO (17 countries)
Geographical flows
Intraregional
National interregional
Table 3. Geographical distribution of patent citations in selected countries
(% shares on total citations made)
1990
Nation
Austria
Belgium
Switzerland
Germany
Spain
Finland
France
Italy
Netherlands
Sweden
United Kingdom
no of regions
9
3
7
40
15
6
22
20
4
8
12
intraregional
flows
18,3
31,3
29,2
27,0
22,2
19,8
30,7
27,4
43,7
17,9
28,8
1998
international
flows
72,1
56,1
53,8
36,8
72,1
70,3
51,8
58,8
46,7
76,6
48,7
intraregional
flows
21,0
36,4
27,6
17,5
19,3
19,0
30,6
30,6
24,3
14,9
16,4
international
flows
70,1
56,3
54,3
44,9
76,9
71,5
48,7
54,3
66,7
77,9
65,3
Map 1. Distribution of patents citations by region of origin
1990
Gini = 0.72
1998
Gini = 0.68
Table 5. Concentration indices across 175 regions
Concentration
indices
Citations made
Citations received
1990
1998
1990
1998
CR 5
0.26
0.24
0.26
0.25
CR 10
0.41
0.37
0.41
0.38
CR 20
0.60
0.53
0.61
0.56
Gini
0.72
0.68
0.73
0.70
Herfindhal
0.025
0.021
0.026
0.023
Table 6. Sectoral composition of citations
1990
Sectors
1998
Shares (%)
origin
Citations
within sector (%)
Shares (%)
origin
Citations
within sector (%)
Food, beverages
Tobacco
Textiles
Wearing apparel
Leather and footwear
Wood products, except furniture
Paper
Printing and publishing
Coke and refined petroleum products
Chemicals and chemical products
Rubber and plastic
Non metallic mineral products
Basic metals
Fabricated metal products
Machinery
Office, computing
Electrical machinery
Radio, tv, communication equipment
Medical, precision instruments
Motor vehicles, trailers
Other transport equipment
Furniture and other
Recycling
1.0
0.0
1.0
0.2
0.3
0.9
0.8
0.3
2.1
27.3
2.2
2.1
0.6
6.3
20.0
1.5
8.6
6.2
7.7
3.9
3.1
3.6
0.3
47.4
39.1
9.7
2.8
51.3
4.0
12.4
6.8
12.3
72.6
12.4
14.1
12.8
21.0
37.4
7.7
27.2
30.8
22.4
19.0
6.6
11.4
2.6
0.8
0.2
1.2
0.3
0.3
0.8
1.0
0.3
1.3
18.4
2.6
2.2
0.9
8.0
25.7
1.7
8.8
4.9
6.8
5.3
3.4
4.7
0.4
32.3
51.2
9.3
3.7
48.1
3.9
14.7
7.0
10.6
66.5
14.8
13.5
15.0
20.9
40.4
7.4
28.2
26.5
18.4
17.3
6.9
14.2
2.5
Total / average
100.0
37.9
100.0
33.1
Table 7. Geographical distribution of patent citations in selected sectors
Nation
Towards
Towards
Within the
Within the
contiguous
other
region
nation
regions
countries
Total
Average
distance of
citations, Km
1990
Footwear
Machinery
Computing, office
Total citations
32.7
28.3
31.7
28.6
9.5
8.8
7.0
9.3
3.4
15.3
15.2
15.4
54.4
47.7
46.1
46.7
100.0
100.0
100.0
100.0
339
542
549
516
1998
Footwear
Machinery
Computing, office
Total citations
47.5
23.8
22.2
21.9
9.2
9.1
8.1
9.7
10.8
16.6
16.7
16.2
32.5
50.5
52.9
52.1
100.0
100.0
100.0
100.0
290
573
625
570
The basic model
The main hypothesis to test is if knowledge linkages
are localized in space and therefore if geographical
distance and spatial contiguity influence knowledge
flows across regions.
Moreover, the use of a set of national and regional
dummies allows to control for other potential
influences coming from institutional and cultural
differences specific either to the country or to the
local area.
Dependent variable: patent citations (C)
Knowledge flows are proxied by the number of citations
between each couple of the 175 European regions
considered.
175x175 matrix where the generic element Cij is the number of
citations originated from patents granted by EPO to inventors
resident in the citing region i and directed to patents granted
by EPO to inventors resident in the cited region j.
Explanatory variables
Geographical distance (GD)
175x175 matrix, the generic element GDij represents the
distance in hundreds of kilometers between the centroids of the
citing region i and the cited region j.
Hypothesis:
a higher distance has a negative impact on the strength of
knowledge spillovers.
Dummy Contiguity (DC)
175x175 dummy matrix, the generic element takes value
one where citing and receiving regions share a border (even
in different countries) and 0 otherwise
Hypothesis:
knowledge flows are facilitated by physical proximity
between regions which share a common border,
irrespective of distance in kilometers (already included in
GD). Positive impact
Dummy Nation (DN)
175x175 dummy matrix, the generic element is equal to 1 if
the citing region i and the cited one j belong to the same
nation, or equal to 0 elsewhere
Hypothesis:
knowledge flows take place more frequently among regions
located in the same nation: exchanges are facilitated by
language, cultural, institutional homogeneity. Positive impact.
Dummy Region (DR)
A set of 175 fixed effects for each region i is included to allow for
idiosyncratic aspects not appropriately measured by the other
explanatory variables, it implies that the model is estimated with
the Least Squares Dummy Variable (LSDV) method
The estimated basic model
Cij = b1 GDij + b2 DCij + b3 DNij + giDRi + eij
Estimation method: LSDV
Two periods: 1990 and 1998
To reduce the problem of firms self-citations, we exclude the
observations within the same region (exclude i = j )
(Maurseth and Verspagen, 2002)
Total number of observations: 30450
Table 8. Determinants of knowledge flows at the aggregate level
Geographical Distance (GDij)
Dummy Contiguity (DCij)
Dummy Nation (DNij)
Dummies 175 Regions (DRi)
Adj. R2
1990
1998
-1.252
-1.349
(0.105)***
(0.089)***
0.835
0.735
(0.048)***
(0.040)***
0.290
0.269
(0.026)***
(0.022)***
yes
yes
0.14
0.16
Extensions to the basic model:
robustness checks
•
•
•
•
•
•
Estimation methods: Poisson
Intra regional citations (include i = j )
Structural distance (SD)
Economic distance (ED)
Technological effort (TE)
General specification
Robustness /1 Estimation methods: Poisson
Poisson estimation may be helpful to ensure a control for the
presence of zeros in the knowledge exchange matrix,
i.e. pair of regions without citation flows.
Reg 1
Reg 2
Estimation year
1990
1998
Number of observations
30450
30450
Estimation method
Geographical Distance (GDij)
Dummy Contiguity (DCij)
Dummy Nation (DNij)
Poisson with f.e. Poisson with f.e.
-0,953
-1,008
(0.021)***
(0.021)***
0,536
0,449
(0.034)***
(0.034)***
0,304
0,197
(0.027)***
(0.026)***
Robustness /2
Intra regional citations
Consider citations originated and received by the same regions.
Problem: it may also include some intra firm citations.
Include also a dummy “within region” (DW) which controls for i = j
Reg 3
Reg 4
Estimation year
1990
1998
Number of observations
30625
30625
random eff.
random eff.
-0,852
-1,013
(0.313)***
(0.221)***
0,304
0,747
(0.081)***
(0.103)***
0,853
0,287
(0.147)***
(0.056)***
20,4
14,0
(0.287)***
(0.201)***
0,18
0,19
Estimation method
Geographical Distance (GDij)
Dummy Contiguity (DCij)
Dummy Nation (DNij)
Dummy Within region (DWij)
Adj. R2
Robustness /3
Structural distance (SD)
Hypothesis: knowledge flows occur with greater intensity
between regions with comparable production structure since
exchanges are easier within similar sectors.
K
175x175 matrix:
the generic element SDij is :
SDij  1  Pij  1 
f
k 1
K
( f
k 1
ik
f jk
K
2
ik
f
k 1
2
jk
Pij measures the similarity between i and j (correlation index)
fik represents region i share in sector k with respect to the total
(measured in terms of patents)
The index ranges between:
0 (identical sectoral structure between the two regions)
1 (the production structures are orthogonal)
Robustness /3
Structural distance (SD)
Reg 5
Reg 6
Estimation year
1990
1998
Number of observations
30450
30450
fixed eff.
fixed eff.
Estimation method
Geographical Distance (GDij)
Dummy Contiguity (DCij)
Dummy Nation (DNij)
Structural Distance (SDij)
Adj. R2
-1,048
-1,170
(0.107)***
(0.091)***
0,833
0,733
(0.048)***
(0.040)***
0,300
0,278
(0.026)***
(0.022)***
-0,445
-0,389
(0.046)***
(0.039)***
0,14
0,17
Robustness /4 Economic distance (ED)
Hypothesis: more knowledge exchanges happen among
regions which are closer in terms of economic conditions.
175x175 matrix: the generic element EDij is computed as the
absolute difference in GDP over population between the origin
and the destination region:
EDij =  (GDP / POP)i – (GDP / POP)j 
Robustness /4 Economic distance (ED)
Reg 7
Reg 8
Estimation year
1990
1998
Number of observations
30450
30450
fixed eff.
fixed eff.
Estimation method
Geographical Distance (GDij)
Dummy Contiguity (DCij)
Dummy Nation (DNij)
Economic Distance (EDij)
Adj. R2
-1,176
-1,257
(0.106)***
(0.090)***
0,827
0,729
(0.048)***
(0.041)***
0,286
0,266
(0.026)***
(0.022)***
-0,006
-0,005
(0.001)***
(0.000)***
0,14
0,16
Robustness /5 Technological effort (TE)
Hypothesis: more knowledge exchanges happen among
regions which are characterised by a larger amount of
resources allocated to technological activity
Include two vectors calculated as the shares of R&D
expenditure over GDP both in the origin region i and in the
destination region j:
TDi = (R&D / GDP)i
TDj = (R&D / GDP)j
Robustness /5 Technological effort (TE)
Reg 9
Reg 10
Estimation year
1990
1998
Number of observations
30450
30450
random eff.
random eff.
Estimation method
Geographical Distance (GDij)
Dummy Contiguity (DCij)
Dummy Nation (DNij)
Technological effort origin (TEi)
Technological effort destination (TEj)
Adj. R2
-0,554
-0,885
(0.104)***
(0.087)***
0,845
0,740
(0.047)***
(0.039)***
0,354
0,308
(0.026)***
(0.021)***
0,153
0,214
(0.018)***
(0.025)***
0,160
0,222
(0.004)***
(0.006)***
0,17
0,20
The general specification
Cij = b1 GDij + b2 DCij + b3 DNij + b4 SDij + b5 EDij + b6 TEi + b7 TEj + gi DRi + eij
Estimation year
Number of observations
Estimation method
Reg 11
1990
30450
random eff.
Reg 12
1998
30450
random eff.
Geographical Distance (GDij)
-0,344
(0.107)***
-0,709
(0.087)***
Dummy Contiguity (DCij)
0,832
(0.047)***
0,734
(0.039)***
Dummy Nation (DNij)
0,353
(0.026)***
0,309
(0.021)***
Structural Distance (SDij)
-0,218
(0.045)***
-0,205
(0.038)***
Economic Distance (EDij)
-0,009
(0.001)***
-0,005
(0.000)***
Technological effort origin (TEi)
0,153
(0.018)***
0,211
(0.025)***
Technological effort destination (TEj)
0,161
(0.004)***
0,219
(0.006)***
0,17
0,20
Adj. R2
Summary results
• Knowledge flows are bounded in space and characterized by
a spatial declining effect (due to spatial transaction costs in
knowledge exchange).
• Flows between neighboring regions are higher.
• Flows are more likely when the two regions belong to the
same country; national borders constitute an obstacle to
knowledge leakages, the national systems of innovation still
play a role with respect to a unified European system.
• The diffusion of technological spillovers is improved when the
origin and destination regions are similar in terms of
production/technological structure and economic conditions
and allocate more resources to innovative activities.
• The importance of such effects is changing along time
Future work
• Integrate the sample of citations with data provided by
OECD
• Analysis of spatial dependence with spatial
econometric techniques
• Focus on the industrial dimension (traditional vs hightech)
• Deeper analysis of technological networks among
firms and inventors in specific industries