Jorge Britto (*)

Leonardo Costa Ribeiro (**)

Eduardo da Motta e Albuquerque (***)

Networks of international patent citations:

pattern of growth, self-organization and change(&)

Abstract

Key-words: Patent Citations; International Knowledge flows; Innovation Systems

JEL classification: O32, O34, O39

Niterói, Belo Horizonte

(25 March 2019)

(*)
Departamento de Economia da Universidade Federal Fluminense (UFF), 424 - Rua Alexandre Moura,
Bloco F, 8 - São Domingos, Niterói (RJ), CEP 24210-200, Brazil.
(**)
Cedeplar-UFMG, Gabinete 3133, Av. Antônio Carlos, 6627; Belo Horizonte (MG); CEP 31270-901,
Brazil.
(***)
Cedeplar-UFMG, Gabinete 3069, Av. Antônio Carlos, 6627; Belo Horizonte (MG); CEP 31270-901,
Brazil.
(&)
We thank the financial support from CNPq (Grants 401054/2016-0 and 307787/2018-4). The usual
disclaimer holds
 1

INTRODUCTION
Patent citations (Jaffe et al, 2002) and their networks (Breschi et al, 2004; Erdi,
2016; Valverde, 2014; Érdi et al, 2013) have been discussed in the literature. Focusing
in international or cross-border patent citations (Jaffe et al, 1999), the contribution of
this paper is the investigation of networks of those international patent citations. The
investigation of networks of cross-border patent citation unveils international
knowledge flows, an important phenomenon for discussions related to a transition "from
national to international innovation systems" (Soete et al, 2010, p. 1176).
Networks of patent citations have been investigated in the literature - examples
of this literature are Érdi et al (2013, p. 227) and Strandburg et al (2009, p. 1660). Those
investigations defined the components of their network clearly: for example, Erdi et al
(2013, p. 227) defined that in their study, "...the patent citation network is comprised of
patents (nodes) and the citations between them (links)".
In a dialogue with this literature, our paper defines differents links and nodes,
therefore a different network. The focus on international networks defines the unit of
analysis of this paper: a cross-border patent citation - a proxy for an international
knowledge flow. The identification and measurement of those cross-border patent
citations might contribute to a further understanding of international flows that tension
and connect national systems of innovation.
This unit of analysis leads to a first difference in our network: its link is a cross-
border or international patent citation. A second difference is related to our definition of
node: the node of our network an institution (patent assignee), that could be a firm, a
research institution, a government agency or even individual inventors.1
The investigation of a network formed by the combination of our link
(international patent citation) and our node (the institution that owns the patent, the
patent assignee) is the contribution of this paper.
The investigation of the nature and dynamics of this network is the objective of
this paper. Basic questions on this investigation are: Does the number of links and nodes
grow over time? Does this network spread globally? Is this network a random network?
Is it static or has it dynamic properties? If dynamic, how does it evolve over time?
Table 1 shows its growth: between 1991 and 2009 the total of patents, of patents
with citations and the total of patents with international citations grew. The number of
cited countries also grows, a hint of the growing internationalization - or global reach -
1
In a previous work (Britto et al, 2019a) we studied one institution (IBM) and its patent citation network.
 2

of this network. Using Patstat, a database was prepared with USPTO patents for
selected years between 1991 and 2009, totalling 1,022,490 patents, 786,780 patents with
international citations and 4,064,995 cross-border citations, according to Table 1.
**********
TABLE 1
**********
There are other evidences of the internationalization of patent citations in Table
1. In 1991, 67.1% of the patents cited a patent from abroad, in 2009 this percentage
grew to 79.3%. In 2009 the total of patents with international flows was greater than the
total of patents in 1994 and almost the same as in 1997. The international reach of this
network is also shown in Table 1: in 1991 there were 91 cited countries and in 2009
there were 151 countries.
Those evidences of growth in the network of international patent citations put
forward questions on the nature of its growth and other dynamic properties.
International patent citations as a channel of international flow is very difused,
but there are other channels through patents. Ribeiro et al (2016) present other different
internationalization measures using patents. Four of them (flow assignee-author, flow
GUO-assignee, flow co-author, and flow co-assignee) have been used in the literature
(see Guellec et al, 2004 and Laurens et al, 2015). Ribeiro et al (2014) investigated
another measure: patent citation of foreign ISI-indexed papers.2
International patent citations constitute a broader source of international flows
than the other five, but they might combine and overlap in the internationalization of
knowledge flows. Since those international flows through patent citations are so
generalized, they might form a network connecting different firms and other institutions
that patent. The size of this network of international citations totals 4,064,995 links. To
investigate the nature of those links and of the network that they form are the goals of
this paper.
This investigation is organized in five sections. The first section reviews the
literature on patent citations and their networks, to locate those networks within a

2
For a comparison with the other four internationalization indicators used in the literature, statistics for
2010 prepared by Ribeiro et al (2016), investigating triadic patents, shows the percentages of each
international flow vis-à-vis the total of patents. The percentages are as follows: 1) international flow
Assignee in one country, Inventor in another country: 11%; 2) international flow GUO (the owner of the
group) in one country, the Assignee in another: 4%; 3) international co-authorship (inventors in different
countries): 7%; 4) Assignees in different countries: 1%. The fifth international flow was investigated in
Ribeiro et al (2014): for USPTO patents in 2009, 5.9% cited foreign ISI-indexed papers.
 3

broader process of emergence of rudiments of an international system of innovation.

The second section presents our database, and how it may be used to investigate the
specific nature of this network, with institutions as nodes and international patent
citations as links. The third section deals with the basic structure of this network,
evaluating how those links are distributed by countries and investigating the nodes of
this network - what are the relevant institutions, how they are linked through patent
citations. The fourth section investigates the properties of this network, specially its
stability, its growth and whether or not it is a self-organized system. The fifth section
compares the network of international patent citations with the networks organized
around scientific production and asks how far those networks might have gone in the
transition between national and international systems of innovation.

I- LITERATURE REVIEW: PATENT CITATIONS AND NETWORKS OF

INTERNATIONAL KNOWLEDGE FLOWS

Since the seminal analysis from Arrow (1962) on the contradictory role of
patents as source of information and as a monopoly, stocks of patents organized by
patent offices (see www.uspto.gov, as an example) may be investigated as rich source of
technological information. Results of patent searches provide free access to this stock of
knowledge codified in patent documents.
However, Arrow (1974) also puts forward a very specific cost to use this
knowledge: previous investments in knowledge and infrastructure would be necessary
to access and understand those stocks of technological information available. This
insight is related to a rich literature on absorptive capabilities necessary to use this
knowledge (Cohen and Levinthal, 1989, 1990).
Patent citations are a proxy of how the available stock of knowledge may be
used as source for new patents (Jaffe and Trajtenberg, 2002).3 Patent citations, therefore,
contributes to understanding of two sets of agents: first, those who generate and "own"
knowledge - patent owners (or patent assignees) - and, second, those who can learn and
use information of that accumulated stock of knowledge to further technological
innovation - institutions that use that knowledge and leave tracks of this use in citing
patents. The investigation on patent citations may uncover those two sides of R&D

3
For a broad review of the literature on knowledge flows through patent citations, see Britto et al (2019a,
sub-section 2.3).
 4

(Cohen and Levinthal, 1989), as firms may invest in R&D to generate new knowledge
(that may lead to a patent) and also invest in R&D to learn - to understand information
accumulated in the stock of patents (that may lead learning that will lead to a new
patent, now with a citation to previous patents). Those firms thay create technology can
learn, and firms that learn from other can create new technology. The database on
patents and patents citations may uncovered who does what in those knowledge flows.
Jaffe and Trajtenberg (1999) also pioneered the use of patent citations to
understand international flows.4 Hu and Jaffe (2003) investigated how kwowledge
diffused from US and Japan to Korea and Taiwan during their catch up processes. Those
investigations are a basis for our basic unit of analysis - cross-border patent citation as a
proxy for an international knowledge flow.
Patent citations form networks. Those networks of patent citation have been
investigated by a rich literature (Érdi et al, 2013; Érdi, 2016; Valverde, 2014; Valverde
et al, 2007). Those investigations evaluate properties of those networks, uncovering
their growth (Valverde, 2014, Figure 2), their fequency distribution over time
(Strandburg, 2009, p. 1669), their stratification (Strandburg, 2009, p. 1670), the powe-
law property of those distributions (Valverde, 2014, p. 3), characteristics of specific
networks for different products (Valverde, 2014) and use of those networks to predict
emerging technologies in specific patent classes (Érdi et al (2013).
International knowledge flows have been investigated, but for flows related to
scientific papers (Wagner et al, 2005; Wagner et al, 2015). The properties of those
networks were investigated by those pioneering studies, that found power-law
properties and self-organization. Ribeiro et al (2018) integrate this literature, discussing
the dynamic growth of networks of scientific international colaboration.
Investigations of properties of networks, specially non-random networks and
their self-organization properties were inaugurated by Barabási and Albert (1999), and
those tools were used by Wagner et al (2005) and Valverde (2014) to investigate
networks of scientific papers and patent citations.
This literature provides the basis for our investigation on networks of
international patent citations.
It is important to contextualize those networks of international patent citations in
a broader framework of international knowledge flows. Figure 1 summarizes those

4
For a broad review of the literature on international knowledge flows through cross-border patent
citations, see Britto et al (2019b, section 1).
 5

flows, graphically illustrating case studies described in the literature (see Britto et al,
2013) - each flow drawn in Figure 1 is supported by a specific study. There are flows
within transnational corporations (both tacit and codified knowledge), there are flows
connecting different universities, there are flows between firms and universities and
flows between different firms - patent citations are one of the sources of those flows.

**********
FIGURE 1
**********
This tentative theoretical framework (Britto et al, 2013) informs an important
choice for our analysis of the network of international patent citations: institutions (that
could be firms or universities, inter allia) as patent assignees. The network that this
paper investigates is built upon decisions and actions taken by those key agents, and
they shape the formation and evolution of those networks.
At this stage, those international flows are growing in importance (see Ribeiro et
al, 2014, 2018; Britto et al, 2019a, 2019b) and they connect different national
innovation systems, putting forward new issues and oppening new opportunities.
Between the challenges facing those more internationally connected national systems of
innovation (Silva, 2014), there is a growing tension between the forces of
internationalization and national frontiers of each innovation system. An investigation
of networks of international patent citations might help to understand this source of
connection and tension among national systems of innovation (Soete, 2010, p. 1176).

II- DATA AND METHODOLOGY

- Patstat and USPTO.
- description (size, years, problems)
- why patstat: citation of citation of citation
- Nodes: institutions (assignees as location of the patent: why?)
- Links: patent citation
- Data prepared to allow network analysis.

III- A NETWORK OF FIRMS, INSTITUTIONS AND COUNTRIES CONNECTED

BY PATENT CITATIONS
 6

This section describes the basic statistics of the network of patent citations. First,
it describes its nodes - patent assignees - and the links (international patent citations)
that each of them have. Second, those data are rearranged to see how countries
aggregate those links.

III.1- FIRMS AND INSTITUTIONS AS NODES

In 2009 there were 148,051 different patent assignees - different institutions. The
distribution of links (international patent citations) accross those nodes is very
concentrated: the first 500 nodes have 451,234 links - almost the same total as the
remaining 147,551 nodes.
Table 2, data for 2009, shows the 30 leading nodes of this network, according to
the total of links (cross-border patent citations) that each node has. Table 2 shows only
one type of institution - firms. Given the role of transnational corporations in those
flows, according to the framework presented in Figure 1, this ranking is not surprising.
**********
TABLE 2
**********
IBM leads this ranking, with 16,560 cross-border patent citations, to patents
from 2,100 different institutions from 47 different countries.5 Over time this ranking
changes: in 1991 IBM was in the third position, behind Hitachi and Siemens. In 1991
cross-border patent citations connected IBM with 477 institutions from 22 different
countries. In a previous paper, Britto et al (2019a) describe changes in IBM patent
citations, both quantitatively and qualitatively, showing how this transnational
corporation increased the number of citations and moved to new technological sectors.6

5
Unfortunately there are remanining different identifications for IBM (IBM Corporation and IBM Corp.),
besides typos and other minor mistakes. This means that IBM would have more citations than those
shown in Table 2 - the total would be 17,941 citations. Since it was not feasible to correct all database,
and assuming that those mistakes would be distributed among all institutions, we decided to use those
data mentioning its problems. There would be no change in the ranking. and the hierarchical nature of this
network would be more unequal than shown in section IV. Other differences derive from decisions of the
firms - IBM seems to put all patents under the ownership of IBM in the US, while Novartis distributes its
ownership through its different divisions and subsidiaries.
6
For a comparison with other networks of international knowledge flows, IBM is at 1,230th position in
2015 in the ranking of institutions with international co-autorships (Ribeiro et al, 2018). The first firm in
that ranking is Novartis, which is in the 545th position (with more than 5,000 connections) - Novartis is in
the 175th position in the ranking presented in Table 2.
 7

The first university in the ranking presented in Table 2 is the University of

California - 62nd position,7 with 1,198 citations from 34 countries. MIT follows in the
73rd position, University of Princeton in the 216th position and Stanford University is
in the 226th position.
The first government agency in Table 2 ranking is the "United States of America
as represented by the Secretary of the Navy", in the 91st position, with 888 international
citations, from 37 different countries.
Individual inventors (identified by no assignee name, or by an assignee name
equal to the inventor name) are present around the 502th position, with 20 citations.
Table 3 presents another view of this network, focusing the links in themselves:
a ranking of the leading links, with the two institutions connect by cross-border patent
citations. Table 3 also presents only firms as nodes connected in those 20 leading
positions. As an evidence that the two sides of R&D (Cohen and Levinthal, 1989) are
deeply correlated, four firms are in Table 3 both as citing and cited assignee.
**********
TABLE 3
**********
The links described in Table 3 suggest that there is also a huge concentration in
the distribution of those links. For instance, IBM (as a citing institution) in four links
presented in Table 3 concentrates almost 20% of all its links in four firms.
Table 4, a ranking of the 21 leading cited patent assignees also display only
firms in those positions.
**********
TABLE 4
**********
Among those 21 firms, 10 are also in leading positions as citing patent assignees
- a hint on the two faces of R&D: firms that use other firms stock of knowledge also

7
University of California would be the first research institution in the ranking of institutions with
international co-autorships (Ribeiro et al, 2018), if we put together all campi, as is the case with its
patents. In the WebOfScience, scientific papers have addresses of the specific campus of the University of
California, spread through its nine campi - UC Berkeley is the 56th with 32,536 international co-
authorships, UC Irvine in the 125th position with 24,554 international co-authorships. If we add the co-
authorships of all campi, University of California total would be 176,699 international co-authorships
(almost three times the total of Oxford University, the first institution in that rankinng). Oxford University
leads the ranking of international co-authorships and it has only two links in our database (a quick search
at the USPTO database shows only 2 patents granted in the years of our database). Cambrige University,
second at the ranking of international co-authorships has only 17 links in our database (patent assignee:
Cambridge University Technical Services Ltd). MIT is 25th in international co-autorships.
 8

provide knwoledge to other firms. This phenomenon might be also another evidence on
Rosenberg suggestion that firms would invest their money in basic R&D as an entry
ticket to flows of knowledge (Rosenberg, 1990).

III.2- AGGREGATING BY COUNTRIES

Table 5 organizes the data by countries, aggregating the total of cross-border
patent citations according to the location of the patent assignee. The developed
countries’ national systems of innovation lead the ranking – US, Japan, Germany and
Canada are the four countries that more intensively access the international stock of
knowledge available through patents.
**********
TABLE 5
**********
Table 5 shows how recently successful catch up countries (South Korea and
Taiwan) ranks well in this regard – 6th and 8th positions, and how China is improving her
position (in 2009 in the 20th, an improvement compared to the 35th in 1991).
In the ranking shown in Table 5, the first country at the periphery – besides
China - is at the 29th position: South Africa. Russia (37th), Argentina (38th), Brazil (39th)
and Mexico (40th) follow.
Table 6 shows the pairs of countries, highlighting how those flows are
concentrated among developed countries – there are no countries at the periphery in the
30 leading pairs. Recently successful catch up countries (South Korea and Taiwan) are
in this list, both citing US patents (respectively pairs number 12 and 13). However, both
rank better as source of knowledge (countries of cited patents, South Korea in positions
6 and 7, Taiwan in position 10).
**********
TABLE 6
**********
Table 6 also show how the two sides of R&D are present, as countries in the
ranking are listed as citing patents and as with patents cited. The US, for example, are
12 times in pairs as a country of a citing assignee and 13 times as a country of a cited
assignee.
 9

IV- THE NETWORK: ITS GROWTH, LONG TERM DYNAMICS AND

PROPERTIES
Those data on nodes (institutions) and links (cross-border patent citations)
introduce an analysis of the network of international patent citations. Barabási and
Albert (1999) introduce tools for analysis of non-random networks and their specific
properties. Valverde (2014) and Érdi (2016) present analysis of networks of patent
citations.

IV.1- NODES AND LINKS - SIZE AND GROWTH OF THE NETWORK

Table 7 summarizes data regarding size, growth and basic features of this
network. Between 1991 and 2009, the total of nodes grew 2.58 times and the total of
links grew 4.73 times. The growth of those two components of this network is higher
than the growth shown in Table 1 for patents and patents with international citations,
respectively 1.58 and 1.88 times. Table 1 also shows that the spread of this network
through different countries increased 1.57 times. Those comparisons show that the
network in itself has a pattern of growth more intense than its basic components –
patents and countries involved.
**********
TABLE 7
**********
As the growth of links (international patent citations) is greater than the growth
of nodes (institutions, patent assignees), the level of connectiveness of each node also
increases, as shown in Table 7: in 1991 there are 3.67 links per node, and in 2009 in
reaches 6.72 links per node.
The nature of the distribution of those links per node is initially evaluated in the
previous section, that highlighted how the leading institutions (mainly firms)
concentrated a huge number of links. Those links also are very strong between leading
pairs (connections between two nodes of the network), also indicating a uneven
distribution of those connections – hints of a hierarchical organization of this network.
The exponents shown in Table 7 further suggest a hierarchical organization of this
network.

IV.2- PROPERTIES OF THE NETWORK: SCALE-FREE AND SELF-

ORGANIZATION
 10

The pattern of long term growth of this network is presented in Graph 1, that
investigates that nature of this frequency distribution of links per node, from 1991 to
2009.8
Graph 1 shows that the network grows preserving its basic structure. The
network displays a power law distribution of connections, in all those years – few nodes
(hubs) have many links, many nodes have few links. The identification of this power
law distribution highlights the scale-free nature of this network, an indication of the
self-organization of this network.
**********
GRAPH 1
**********
The exponents of those power law distributions are shown in Table 7 – relatively
constant over time, around 2.05. This stability suggests the preservation of the
hierarchical distribution over time, a feature of this self-organized system that grows,
expands – as shown in Tables 1 and 7 –, with more patents, more patents with
international citations, more countries cited in patents, more nodes and more links,
while preserving its basic structure.9
The structure of this network of international patent citations is slightly different
from the structure described by Valverde (2007, p. 3; 2014, p. 3): his network is “neither
exponential nor a simple power law” (2007, p. 3). Probably this difference lies in our
definition of node – the patent assignee, an institution, predominantly a firm, as shown
in section II. Valverde’s node is a patent – patent may have many or few citations, but
an institution may accumulate much more patents and their citations, reaching larger
order of magnitude (?).
The network of international patent citations organized by institutions (mainly
firms, with an important role for transnational corporations) has properties of scale-free
networks, following a simple power law distribution – without deviations -, a
characterization that is enough to highlight its self-organization – an important finding
to open further research on the meaning of this international network.

8
To a cleaner expression of those data, the exponents were calculated for all years presented in Table 7,
but the Graph shows the curves only for three selected years – 1991, 2000 and 2009.
9
Comparing this network with the network of scientific co-authorships, whose exponents are around 1.75
(Ribeiro et al, 2018, Table 5). For the sub-network that involved only firms as first author and their
scientific co-authorships (Ribeiro et al, 2018, Table 7), the exponents are closer to those of the network of
international patent citations. Those comparisons indicate that the network of international patent citations
is more hierarchical than the network (and sub-network) of international scientific collaboration.
 11

Self-organization stresses the stability of this network, its resilience, therefore,

its role as a structural feature of contemporary economy.

IV.3- DYNAMICS OF THE NETWORK: MATRICES OF PATENT

CITATIONS AND THEIR LONG TERM CHANGES
This network grows, self-organizes, reproduces its basic and hierarchical
structure. Furthermore, it is necessary to investigate its long term evolution and possible
changes in its technological specialization.
To investigate this long term evolution, Figure 2 shows, for selected years,
global matrices of technological interaction between cross-border flows among citing
and cited patents. Those matrices, processing the original USPTO classes through an
algorithm suggested by the Observatoire des Sciences et des Techniques (OST, 2006), that
aggregates those classes into 30 different technological subdomains (see Appendix
Table A1 for this list).10
An intertemporal comparison between those matrices might show how this self-
organized system moves over time. Those matrices evaluates our links – cross-border
patent citation – through the investigation of each matrix cell, that contains the
technological class of the citing patent (x-axis) and the technological class of the cited
patent (y-axis). In other words, how the new knowledge (citing patent) is using the
stock of knowledge (cited patents). The intensity of this use might be measure by third
axis (z-axis), that shows the number of citations (cross-border patent citations) for that
cell – the height of the cell is shown in each matrix.
Over time, the number of cells with international citations has grown – the
matrix fulfillment grows -, the height of cells grows (see the order of magnitude in the
z-axis growing between 1991 and 2009), and the position of the peaks change.
Comparing the three matrices in Figure 2, there are movements that differentiate those
three snapshots of the network of international patent citations through the lens of
technological classes.
**********
FIGURE 2
**********

10
Ribeiro et al (2010) present a matrix for interaction between science and technology, a starting point for
the preparation of Figure 2. Britto et al (2019b) show matrices of international patent citations, with the
same methodology used in this paper, but prepared from a different database (USPTO, not Patstat).
 12

There are three main changes in this network over time.11

First, the growth in the network and in the number of links is reflected in Figure
2 through the height of z-axis: the leading peak (cell with more citations) was around
xxx citations in 1991, xxx citations in 2000 and xxx citations in 2009.
Second, there are movements in the ranking of peaks: the first five peaks in 1991
are xxx1, xxx2, xxx3, xxx4 and xxx5, in 2000 are xxx1, xxx2, xxx3, xxx4 and xxx5 and
in 2009 xxx1, xxx2, xxx3, xxx4 and xxx5. In general terms, the movement towards
digital technologies and health-related technologies is captured by those matrices.
Third, over time peaks outside the diagonal become more relevant, as xxx. xxx.
and xxx. This means xxxx
This introductory analysis of those matrices shown in Figure 2 contributes to
uncover in what directions this self-organizing system is moving, how it changes in the
long term – probably a consequence of structural changes related to technological
revolutions and the emergence of new GPTs processing during those years.

V- COMPARING NETWORKS OF PATENT CITATIONS, SCIENTIFIC

COLABORATION AND PATENT CITATIONS OF PAPERS
Xxx database and main conclusions – network is scale-free; evolution preserves
scale-free properties; the network changes its technological structure over time.
Xxx conclusions of this analysis – for the network of international patent
citations: What are their implications for our analysis of that emergence of a GIS? Those
implications will be discussed in relation to growth, hierarchy, opportunity, challenges,
robustness (a step towards a global innovation system) and for the generation and
transfer of technology.
- Growth
- Hierarchy
- Opportunity
- Challenges
- Robustness
- Generation and transfer of technology
Xxx Concluding this analysis, now we have two different networks – with
evidences and data – organized around different flows of international knowledge,
organized around different institutions, but sharing a common feature: self-organization
and resilience. Those findings leads us to two sets of questions:
1) Differences in the structure of networks: Ribeiro et al (2018): links between
co-authors (although the node is the institution) and here the links are between assignees
(there are more inventors than assignees. Size, pattern of growth, hierarchy etc

11
For a broader list of different indicators such as fulfilment indexes, diagonalization indexes,
concentration of leading cells, etc, that can be used in the analyses of those matrices, see Britto et al
(2019b).
 13

2) Overlapping between those networks. Firms internationally co-author papers,

universities have international patent citations, firms cite foreign scientific papers. But
this overlapping and those intersections may be shaping innovation systems – as other
complex systems, innovation systems are composed of different structures, different
layers – and those constitutive structures (network of international scientific
collaboration, network of international patent citations) are self-organized systems that
are connected by diverse institutions and flows – that might be part of a broader process
of emergence of an international system of innovation. This paper can only point to this
subject, suggesting further research on how those self-organized systems connect,
interact and might be aggregated in a broader complex system.