Evolutions of Social Networks

Invited Review
Evolution of social networks

Tim Hellmann, Mathias Staudigl
Center for Mathematical Economics (IMW), Bielefeld University, Germany

a r t i c l e i n f o
Article history:
Received 2 October 2012
Accepted 18 August 2013
Available online 30 August 2013
Keywords:
Evolution of networks
Game theory
Stochastic processes
Random graphs
a b s t r a c t
Modeling the evolution of networks is central to our understanding of large communication systems, and
more general, modern economic and social systems. The research on social and economic networks is
truly interdisciplinary and the number of proposed models is huge. In this survey we discuss a small
selection of modeling approaches, covering classical random graph models, and game-theoretic models
to analyze the evolution of social networks. Based on these two basic modeling paradigms, we introduce
co-evolutionary models of networks and play as a potential synthesis.
2013 Elsevier B.V. All rights reserved.
1. Introduction
The importance of network structure in social and economic
systems is by now very well understood. In sociology and applied
statistics the study of social ties among actors is a classical eld,
known as social network analysis (Wasserman & Faust, 1994).
More recently a large scientic community, including game theo-
rists, economists, as well as computer scientists and physicists, rec-
ognized the importance of network structure. In particular, the
dynamic evolution of networks became an important question of re-
search. Of course all these subjects put different emphasis on what
is considered to be a good model of network formation. Tradi-
tionally, economists are used to interpreting observed network
structure as equilibrium outcomes. Naturally, game theory is the
predominant tool used in this literature. Computer scientists, on
the other hand, prefer to think of network formation in terms of
dynamic algorithms. Finally, physicists tend to think of networks
as an outgrowth of complex system analysis, where the main inter-
est is to understand and characterize the statistical regularities of
large stochastic networks. Given this interdisciplinary character
of the subject, the number of publications is enormous, and it is
impossible to provide a concise survey covering the plethora of
models developed in each of the above mentioned disciplines.
For this reason, we have decided to focus in this survey on two,
in our opinion, particular promising approaches to model the evo-
lution of social and economic networks. We concentrate on dy-
namic models of network formation, using elements from
random graph theory and game theory. These two approaches ma-
tured over the years, and some recent efforts have been made to
combine them. This article summarizes a small body of these
two streams of literature; it is our aim to convince the reader that
random graph dynamics and (evolutionary) game theory have
many elements in common, and we hope that this survey provides
some ideas for future research on this young and interdisciplinary
topic. However, before jumping into the details, let us give a short
overview of topics which this survey covers, a pointer to the fur-
ther relevant literature, and an acknowledgment of the literature
which we shamefully exclude.
Section 2 starts with a short discussion of random graph mod-
els. These models are the basis for the statistical analysis of net-
works and have had a large impact on theoretical models of
network evolution. Random graph models have a long tradition
in social network analysis, and are the foundation of the recent lit-
erature on network evolution in computer science, mathematics
and physics. Following the terminology of Chung et al. (2006),
we focus on off-line models. Hence, we consider network forma-
tion models in which the number of nodes (the size of the popula-
tion) is a given parameter.
1
The most general random graph model
introduced in this survey belongs to the family of inhomogeneous
random graph models (Bollobs, Janson, & Riordan, 2007). This class
of random graphs is rather rich. It contains well-known stochastic
block-models (Karrer & Newman, 2011), variants of the important
exponential random graph (Snijders, Pattison, Robins, & Handcock,
2006), and the classical Bernoulli random graph model as special
cases. Since these models are very well documented in the literature,
0377-2217/$ - see front matter 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.ejor.2013.08.022
Corresponding author. Tel.: +49 521 106 67256.

E-mail addresses: tim.hellmann@uni-bielefeld.de (T. Hellmann), mathias.stau-
digl@uni-bielefeld.de (M. Staudigl).
1
On-line models are models in which the population is growing over time. This
important class of models contains the very popular preferential attachment models
(Barabsi & Albert, 1999), which we are not touching in this survey. Excellent
(technical) summaries of these models can be found in Newman (2003b) and, more
rigorously, in Chung et al. (2006) and Durrett (2007).
European Journal of Operational Research 234 (2014) 583596
Contents lists available at ScienceDirect
European Journal of Operational Research
j our nal homepage: www. el sevi er . com/ l ocat e/ ej or
we are satised with a short summary of their main properties. The
main purpose of this section is to introduce concepts, and to famil-
iarize the reader with our notation. Nevertheless, we consider ran-
dom graph models as the building block to synthesize random
graph models with dynamic game theoretic ideas, and therefore
we think it is useful to introduce them already at the beginning.
Section 3 presents an alternative approach of modeling net-
works, mainly developed by economic theorists. It uses game the-
oretical concepts to interpret network structures as equilibrium
phenomena of strategically acting players who create and destroy
links according to their own incentives. Our discussion is centered
around two particularly important concepts: The semi-cooperative
solution concept of pairwise-stability (Jackson & Wolinsky, 1996),
and various modications of Nash equilibrium. These concepts are
static equilibrium notions, and the natural question which comes
to ones mind is whether there are natural dynamic models sup-
porting predictions based on such equilibrium concepts. This ques-
tion leads us to consider an evolutionary approach to network
formation. These models are used as an introduction to a particular
interesting class of dynamic network formation models, called co-
evolutionary processes of networks and play in Staudigl (2010). Sec-
tion 4 is devoted to illustrate these types of models, which is our
modest attempt to synthesize the strategic approach of network
formation with a random graph approach. Section 5 summarizes
the main points contained in this article and discusses some ideas
for potential future research.
1.1. What this survey does not cover
Given the enormous number of network formation models, it
was necessary to be selective in writing this survey. Hence, there
are many important network models which we were not able to
cover. Some notable omissions are the following; we restrict the
discussion to network models where the size of the graph is xed
(off-line models). Networks with variable number of nodes
(on-line models) are of course important, but they require differ-
ent mathematical tools to be analyzed successfully. In particular,
we believe that a different game theoretic approach would be
needed to study such models.
2
Readers interested in the mechanics
of growing networks should consult the book by Dorogovtsev and
Mendes (2003), and the survey article by Albert and Barabsi
(2002). Another class of networks which we will not consider are
weighted graphs. These models are very important for applications,
and the physics community provides many interesting approaches
to model such networks (see e.g. Barthlemy, Barrat, Pastor-Satorras,
& Vespignani (2005) and Kumpula, Onnela, Saramki, Kaski, & Ker-
tsz (2007), and the references therein). There are also some
gametheoretic models on the formation of weighted networks,
see e.g. Bloch and Dutta (2009). These are just preliminary studies,
and we have the feeling that much more work on these kinds of net-
works will be needed before they should be included in a survey. Fi-
nally, we would like to point out that our main focus will be on
evolutionary models of undirected networks. This does not mean that
we think directed networks are less important. In fact, many real-
world networks, such as trafc networks or the world-wide-web,
are more naturally interpreted as directed graphs. However most
of the game-theoretic concepts emphasize bilateral externalities,
and thus admit a cleaner interpretation when links are undirected.
Nevertheless, we provide some discussion of directed networks. In
particular, the random graph models of Section 2 can be used to
model the evolution of directed, as well as undirected networks,
after straightforward modications.
1.2. Related literature
Most of the models which found no space in this article have
been surveyed elsewhere. We just provide some cross references
to the literature, and urge the reader to consult the references
therein. Recent textbooks containing in-depth discussions on dy-
namic network formation are Chung et al. (2006) and Durrett
(2007). These books focus on the mathematical aspects, and only
mention rudimentary applications. With a slight bias towards eco-
nomics applications, we recommend the beautiful books by Goyal
(2007), Vega-Redondo (2007), Jackson (2008), and Easley and
Kleinberg (2010). Additionally to textbook treatments, the reader
may want to consult one of the numerous survey articles available.
We just mention Jackson (2003), Newman (2003b), Jackson (2005),
Van den Nouweland (2005), Goyal (2005), Goldenberg, Zheng,
Fienberg, and Airoldi (2009). At the heart of our discussion is the
evolution of networks, and we use the language of stochastic pro-
cesses and/or game theory to formalize our ideas. Additionally we
try to highlight the potential connections between these two
seemingly separate modeling strategies. We hope that this inte-
grated perspective on the evolution of networks makes this survey
a good contribution to the literature and will inspire some people
to work on this fascinating topic.
2. Stochastic models of network evolution
A statistical analysis of networks is usually based on elements
of random graph theory. Random graph models are very exible
mathematical representations of interdependency relations. In-
deed, the motivation for such classical models as the Markov graph
model of Frank and Strauss (1986) was to develop a tractable rep-
resentation of complicated interdependencies in empirical data.
Random graph models in general provide a exible framework to
construct statistical ensembles of networks (i.e. probability spaces),
which are parsimonious enough to get pointed predictions, and
rich enough to be able to reproduce as many stylized facts the re-
searcher is aiming to model. In this section our discussion is cen-
tered around a rather general class of a random graph process,
which will be used as one pillar in our development of co-evolu-
tionary processes of networks and play, starting in Section 4.
2.1. Random graphs
In this survey we use the terms networks and graphs inter-
changeably. A graph is a pair G = ([N], E), where [N] := {1, 2, . . . , N}
is the set of vertices (or nodes), and E = E(G) [N] [N] is the set
of edges (or links). This notation applies to directed as well as undi-
rected networks. If G is a directed graph, then (i, j) E means that
there is a directed edge from i to j. If G is undirected, then
(i, j) E(G) if and only if (j, i) E(G). In such a symmetric setting it
will be convenient to use the shortened notation ij for the link con-
necting node j and i. We denote the collection of graphs on N ver-
tices by ([N[. Elements of this set can either be directed or
undirected networks, depending on the context.
A random graph is a probability space (([N[; 2
([N[
; P). The proba-
bility measure P : 2
([N[
[0; 1[ assigns to each graph a weight,
which should reect the likelihood that a certain graph G is drawn
from the set ([N[, when performing a statistical experiment with
distribution P. Thus, the underlying reference measure P it is
chosen by the modeler. A historically very important random
2
For instance, a basic strategic decision in a network model with variable number
of players is when and whether a single player should enter or leave the network. See
Dutta, Ghosal, and Ray (2005) for a model in this direction. In evolutionary game
theory, the standard model assumes a constant population size. However, it seems to
be likely that allowing for population growth introduces for new phenomena absent
from the stationary world. See Sandholm and Pauzner (1998) for an interesting study
in this direction. A recent model of economic network formation in a non-stationary
environment is Jackson and Rogers (2007).
584 T. Hellmann, M. Staudigl / European Journal of Operational Research 234 (2014) 583596
graph model is the ErdsRnyi graph (or Bernoulli graph).
3
The
Bernoulli random graph is built on the assumption that edges are
formed independently with constant probability p [0, 1]. Thus, let
([N[ be the set of undirected graphs constructable on N vertices.
Then the random graph model (([N[; 2
([N[
; P) is determined by the
probability measure
P(G) :=
Y
N
i=1
Y
j>i
p
A
ij
(G)
(1 p)
1A
ij
(G)
; (1)
where A
ij
(G) = 1 if and only if (i, j) E(G), and 0 otherwise.
A(G) = [A
ij
(G); 1 6 i, j 6 N] is the adjacency matrix of the graph G.
4
The advantage of the Bernoulli graph model is its simplicity. In
fact, the complete random graph is described by two parameters:
the population size (N), and the edge-success probability (p). This
model is the benchmark of random graph theory and is by now
very well understood. A detailed analysis can be found in Bollobs
(2008). It comes with little surprise that such a simple model is
rarely a good description of a real-world network. In particular,
the lack of correlation across links is a well-known decit of the Ber-
noulli graph. However, it still constitutes an important benchmark
(Newman et al., 2003a). Compared to random graphs, real world
networks are often observed to have similarly small average path
length (an effect coined small worlds phenomenon, or six degrees
of separation),
5
higher clustering (friends of friends are more likely
to be friends),
6
exhibit homophily (connections between nodes of
similar kind are more likely),
7
and often exhibit a power law degree
distribution (more nodes with very high and very low degree com-
pared to Bernoulli graphs, Newman, 2003b). For a more detailed dis-
cussion we refer the reader to Vega-Redondo (2007) and Jackson
(2008).
A different example for a commonly used random graph mea-
sure is the exponential random graph (ERG). This class of random
graph models builds on the Markov graph models of Frank and
Strauss (1986), and is also known as p
-models (Robins, Pattison,

Kalish, & Lusher, 2007). This class of models is very rich, and its
main strength are its tractability and ability to reproduce networks
with arbitrary degree distributions, and clustering. Let us give a
simple example for an exponential random graph model. For more
information we refer to the excellent paper by Snijders et al.
(2006), where a state-of-the-art survey can be found.
8
A simple
version of the exponential random graph model is given by the expo-
nential family
P
b
(G) := exp
X
k
i=1
b
i
T
i
(G) w(b)
!
; (2)
where
~
b = (b
1
; . . . ; b
k
) is a vector of real parameters, T
i
(G) are func-
tions on the space of graphs, i = 1, . . . , k for some k N, and w(b) is
the normalizing constant (the partition function). The functions T
i
are usually taken to be simple statistics of the graph, such as the de-
gree sequence of the graph, or the number of edges. Depending on
the sign of the parameter b
i
, the value of the function T
i
(G) deter-
mines the probability that the graph G is drawn from the whole
space of graphs when using the measure P
b
. Similar exponential
families for graph measures will appear in Section 4 of this survey.
As a nal word on this family of models, let us show that in fact the
ERG contains the Bernoulli graph. To wit, let ([N[ be our space of
undirected networks, and take T
1
(G) := [E(G)[ the number of edges,
and b
1
= b, b
i
= 0 for i P2. Setting p := e
b
/(1 + e
b
), one immediately
sees that
P
b
(G) =
1
exp(w(b))
exp b
X
i;j>i
A
ij
(G)
!
=
Y
N
i=1
Y
j>i
p
A
ij
(G)
(1 p)
1A
ij
(G)
:
2.2. Network formation as a stochastic process
A classical approach in social network theory is to view the evo-
lution of a network as a stochastic process (see Wasserman (1980)
for an early contribution). Snijders (2001) constructs a dynamic
ERG which can be analytically solved, and numerically simulated
via MCMC-methods. In the following we will describe a fairly gen-
eral dynamic network formation model in terms of continuous-
time Markov jump processes.
In a dynamic model of network formation we would like to cap-
ture two things: First, the network should be volatile; i.e. links are
deleted and formed over time. Second, the likelihood that a link is
formed or destroyed, should depend on the attributes of the verti-
ces in the graph. The following network formation algorithm cap-
tures both of these requirements. We are given some probability
space (X; T; P). The sample space X might be larger than the set
of graphs.
9
We call a Markov jump process c(t)
tR
a random graph
process if each c(t) is a ([N[-valued random variable, measurable
with respect to some given ltration T
t
tP0
containing the ltration
generated by the random variables {c(t)}
tP0
. We model the dynamic
evolution of the network via the following steps:
Link creation: With a constant rate k P0 the network is
allowed to expand. Let W : ([N[ R
NN
be a bounded matrix-
valued function, whose components w
ij
(G) dene the intensities
of link formation between vertex i and j. The function W will be
called the attachment mechanism of the process.
Link destruction: Let n P0 denote the constant rate of link
destruction. Let V : ([N[ R
NN
be a bounded matrix-valued
function, whose components v
ij
(G) dene the intensities of link
destruction between vertex i and j. The function V is called
the volatility mechanism of the process.
In some models of network formation the rate of link destruc-
tion n has been interpreted as environmental volatility (Ehrhardt,
Marsili, & Vega-Redondo, 2006; Marsili, Vega-Redondo, & Slanina,
2004). This is also the motivation behind our denition of a volatil-
ity mechanism. In principle the intensities of link creation and
destruction can be asymmetric, i.e. we do not require in the con-
struction that w
ij
(G) = w
ji
(G), or v
ij
(G) = v
ji
(G), respectively. Hence,
in principle the network formation process can be used to model
the formation of directed as well as undirected networks. This sim-
ple model of an evolving network, will be the building block of our
construction of a co-evolutionary model of networks and play in
Section 4.
2.2.1. Inhomogeneous random graphs
Let ([N[ be the set of all undirected graphs. Suppose that the
intensities of link creation and link destruction are given by the
functions
w
ij
(G) = (1 A
ij
(G))j
ij
; and v
ij
(G) = A
ij
(G)d
ij
: (3)
3
Erds and Rnyi (1959, 1960) introduced the slightly different model in which the
number of vertices and edges are given parameters. The Bernoulli graph model is due
to Gilbert (1959).
4
It is clear that if G is undirected, then A(G) is symmetric. For directed graphs this
symmetry may not hold.
5
A classical reference is the famous letter experiment of Milgram (1967). Other
studies include Dodds, Muhamad, and Watts (2003) and Goyal, Van Der Leij, and
Moraga-Gonzlez (2006).
6
See e.g. Watts and Strogatz (1998), Watts (1999) and Newman (2003b, 2004).
7
See McPherson, Smith-Lovin, and Cook (2001) and Golub (2012), among others.
8
Despite its huge importance, the ERG model has technical problems. See
Chatterjee and Diaconis (2011).
9
This will be necessary in order to model the co-evolution of networks and play in
Section 4.
T. Hellmann, M. Staudigl / European Journal of Operational Research 234 (2014) 583596 585
The scalars j
ij
, d
ij
are positive and symmetric, meaning that j
ij
= j
ji
and d
ij
= d
ji
for all i,j [N]. Additionally we assume that k = n = 1.
This essentially says that the processes of link creation and link
destruction run on the same time scale. Then the following general
picture emerges.
Theorem 2.1 Staudigl (2013). Consider the random graph process
c(t)
tR
with attachment and volatility mechanism W and V given
by the functions specied in Eq. (3). Then the graph process is ergodic
with unique invariant measure
P(G) =
Y
N
i=1
Y
j>i
(p
ij
)
A
ij
(G)
(1 p
ij
)
1A
ij
(G)
; (4)
where p
ij
=
j
ij
j
ij
d
ij
is the edge-success probability of vertex i and j.
The random graph measure (4) describes the probability space
of an inhomogeneous random graph. The wonderful work of Bol-
lobs et al. (2007) studies this model in quite some detail.
10
It con-
tains the ErdsRnyi model as a special case by setting j
ij
= j and
d
ij
= d. Moreover, it generalizes certain networks based on clustering
nodes according to some notion of similarity, as explained in the
next subsection.
2.2.2. Multi-type random networks
In general, networks are complex objects and therefore difcult
to analyze. However, it is often the case that vertices in a network
can be classied to belong to certain groups. Indeed, a prevalent
fact in social networks is the phenomenon of homophily, meaning
that vertices of similar characteristics are more likely to be con-
nected with each other. Therefore, Fienberg, Meyer, and Wasser-
man (1985) introduced blockmodels, where nodes are categorized
to belong to certain subgroups. Each subgroup may have its own
law of network formation. Recently, this type of networks has also
been used in economic theory (Golub, 2012), where it has been
called a multi-type random network.
Suppose that the set of vertices can be partitioned into nitely
many types k {1, . . . , m}. The number of nodes of type k is n
k
,
where 0 6 n
k
6 N. The vector ~n = (n
1
; . . . ; n
m
) denes the partition
of the population of nodes into its types. The number n
k
-
{0, 1, . . . , N} can be either deterministically given, or represent
the realization of a random variable. Assume that the intensities
of link formation and link destruction are w
ij
(G) = (1 A
ij
(G))j
rl
,
v
ij
(G) = A
ij
(G)d
rl
, whenever vertex i is a member of group r, and ver-
tex j is a member of group l for 1 6 r, l 6 m. The model then reduces
to a special case of the inhomogeneous random graph. The edge-
success probabilities between members of group r and group l
are given by p
rl
=
j
rl
j
rl
d
rl
. This grouping, of course, reduces the com-
plexity of the random graph model tremendously. Compared to the
inhomogeneous random graph, the multi-type random graph is
completely specied by the partitioning of the vertices, and the
group-specic edge-success probabilities.
2.3. Random graphs versus strategic network formation
Our proposed strategy to model dynamic network formation is
rather parsimonious. The stochastic process is entirely specied by
the attachment and the volatility mechanism. Particularly interest-
ing examples are the inhomogeneous random graph model, con-
taining the multi-type random network. All these models rely on
a special choice of the intensities of link creation and link destruc-
tion. These rates can in principle be tted from data. Thus, the sta-
tistical models are useful to describe how networks form and
evolve from an observers point of view. They are not able to
explain why networks form and evolve, without adding a comple-
mentary theory of network formation. Thus, to complete the study
of network formation, we have to go to the micro level and under-
stand the forces that drive the nodes to connect to each other. Eco-
nomic and game theoretic reasoning provides the natural language
for this task, and Section 3 summarizes this eld to some extend.
Using these game theoretic concepts, Section 4 presents some
models which combine elements from random graph theory and
game theory.
3. Game theoretic models of network evolution
In social and economic network a link is often interpreted as a
form of revealed preference: Two players are connected with each
other, because it is in their self-interest to have the link. With other
words, individuals are connected with each other because a payoff
or utility is derived from these connections. Frequently the linking
decisions of a single individual affects the utilities of other mem-
bers of the society. To understand networked systems in situations
with interdependent preferences, game theory has been devel-
oped. Using game theoretic reasoning, networks are interpreted
as equilibrium outcomes implemented by rational players. This
modeling approach is potentially useful as it allows us to rational-
ize predicted or observed network structure in terms of socioeco-
nomic behavior. A purely statistical approach to network
formation cannot provide us with such a micro-foundation of ob-
served networks. Moreover, the game theoretic approach allows
the researcher to understand the mechanisms underlying the for-
mation of the network. As a simple example, consider a situation
where a statistician observes a networked system with a center-
sponsored star architecture, such as the one depicted in Fig. 1(i).
11
Having collected these data, a game theoretic model can now
(potentially) be used to postulate preferences (i.e. utility func-
tions), rationalizing this observed interaction structure.
12
3.1. Networks and utilities
A utility-based approach to network formation is rather
straightforward. For each player, there is a utility function
u
i
: ([N[ R. An n-tuple of functions u = (u
1
, . . . , u
N
) represents
the preferences of the society over network architectures. Let us
start with some concrete examples.
3.1.1. The connections model
Research in sociology has emphasized the role of centrality in
social networks.
13
Centrality measures are based on an assessment
of the graph distance between two nodes. The following examples
uses a version of this distance measure, called decay centrality, to
construct a model of strategic network formation. Consider a given
graph G = (V, E). Let d
ij
(G) denote the geodesic distance between ver-
tex i and j, i.e. the length of the shortest path connecting i and j. If
node i cannot be connected to j, then set d
ij
(G) = . A measure of
closeness should then be decreasing in distances. Thus, let d (0, 1)
be a given parameter. The decay centrality between vertex i and j
is then dened as CL
ij
(G) := d
d
ij
(G)
. It has a natural interpretation in
terms of a probabilistic communication model. Let d (0, 1) be the
10
See also Sderberg (2002) and Park and Newman (2004) for earlier treatments of
this model.
11
In this gure, the star is presented as an undirected network. When considering
directed networks, there are several stars: we will call a star network a center-
sponsored star if all links are directed from the center to the periphery players.
Likewise, in a periphery sponsored star, links are directed from the peripheral players
to the center.
12
See Echenique, Lee, Shum, and Yenmez (2013) for a very recent paper in this
direction.
13
See, e.g. Freeman (1979) or Bonacich (1987) for an intriguing discussion of
centrality measures.
probability that communication between adjacent nodes is success-
ful. Then, if the signal transfer is independent, the probability that
the signal reaches vertex j from vertex i with d
ij
(G) = k is on the order
of d
k
. Jackson and Wolinsky (1996) introduce a model were a players
utility is increasing in closeness centrality, but there is a cost to be
paid for each maintained link. Specically, let w
ij
be the intrinsic va-
lue of a communication of player i with player j, and c
ij
> 0 be a cost
which player i has to pay when a direct connection with j is main-
tained. The utility function for player i [N] is given by
u
Co
i
(G) =
X
ji
w
ij
d
d
ij
(G)

X
jN
i
(G)
c
ij
(5)
where N
i
(G) := {j N:ij G} is the set of i
/
s neighbors. This is a ver-
sion of the connections model due to Jackson and Wolinsky (1996).
3.1.2. Local complementarities and Bonacich centrality
Bonacich (1987) introduced a parametric family of centrality
measures in order to formulate the intuitive idea that the centrality
of a single node in a network should depend on the centrality of its
direct neighbors. This self-referential denition of centrality leads
to an eigenvector-based measure, which can be derived from basic
utility-maximization ideas, as shown by Ballester, Calv-Armengol,
and Zenou (2006). Let A be the adjacency matrix of a given net-
work. The powers of this matrix give us information of the connec-
tivity structure of the network.
14
Let b > 0 be a given parameter,
chosen in such a way that the following matrix power series exists
15
:
B(b; A) =
P
n=0
A
n
b
n
= [I bA[
1
. The centrality index proposed by
Bonacich (1987) is dened as
b(a; b; A) = B(b; A)a; (6)
where a
= (a
1
, . . . , a
N
) is a given vector of individual characteristics
of the vertices.
16
Ballester et al. (2006) show that this centrality
measure is actually a Nash equilibrium of an interesting class of
non-cooperative games. There are N agents who are involved in a
team production problem. Each player chooses a non-negative quan-
tity x
i
P0, interpreted as efforts invested in the team production.
High efforts are costly, and the level of effort invested by the other
players affects the utility of player i. To capture these effects, we as-
sume that the payoff of player i from an effort prole x = (x
1
, . . . , x
N
)
is given by u
i
(x
1
; . . . ; x
N
) = a
i
x
i

1
2
x
2
i
bx
i
P
N
j=1
A
ij
x
j
. The players
choose their efforts independently, and in a utility maximizing
way. It can be shown that this game has a unique Nash equilibrium
x
given by x
= b(a, b, A). Hence, the equilibrium effort invested by

player i depends only on his centrality in the network with adja-
cency matrix A. Given the network G, and parameter vector
(a; b) R
N1
, so that (6) is well dened, we can then compute the
equilibrium payoff of player i as
u
BC
i
(G) =
1
2
b
i
(a; b; A(G))
2
(7)
This utility function expresses now preferences over a set of possi-
ble network architectures underlying the team production problem.
3.1.3. R&D collaborations between rms
The above example made the interesting point that equilibrium
structure of non-cooperative games can be closely linked to the
underlying interaction structure of the agents. Another example
where this structure is clearly visible is the model due to Goyal
and Joshi (2003). In this model the nodes in the graph are rms
which produce a homogeneous product and compete in quantities
on a single market. Additionally rms can agree to form bilateral
R&D collaborations which lower their marginal cost of producing
the output. Let us assume that marginal costs of production are
given by w
i
(G) = c
0
c
1
g
i
(G), with parameters c
0
; c
1
R
such that
c
1
<
c
0
N1
and g
i
(G) := [N
i
[ denoting the rm is degree. The output of
rm i is denoted by x
i
, and the market is completely described by a
commonly known inverse demand function P(X) = max[0, a X],
where X P0 is the total output of the industry, and a P1 is the
size of the market. The prot function of rm i is then given by
u
i
(x
1
; . . . ; x
N
) = x
i
P(x
i

P
ji
x
j
) w
i
(G)

. For a given network of
R& D collaborations, it can be shown that the quantity choice game
has a unique Nash equilibrium in which each rm produces
x
+
i
(G) = a X
+
w
i
, with X
+
=
Na
N1

1
N1
P
j
w
j
. The resulting equilib-
rium prot of rm i is
u
R&D
i
(G) :=
(a c
0
) Nc
1
g
i
(G) c
1
X
ji
g
j
(G)
N 1
0
B
B
@
1
C
C
A
2
(8)
Fig. 1. Some important network architectures.
14
Indeed, A
~
1 is a N 1 matrix whose entries are just the degree of the individual
nodes. The vector A
2
~
1 counts the number of walks of length 2 starting from the
individual nodes, and so on.
15
The necessary condition for this to be the case is that 0 < b < k
1
(A)
1
, where k
1
(A)
is the eigenvalue of A having largest modulus.
16
The original Bonacich centrality index is dened as c[I bA[
1
A
~
1. Formula (6) is a
true generalization by setting a = cA
~
1.
Again we see that the equilibrium industry performance can be as-
sessed by knowing the collaboration structure of the rms.
3.2. Pairwise stability and renements
One of the most important equilibriumnotations in the strategic
approach to network formation is the concept of pairwise stability.
This concept considers the differential impact an added link has on
the utility of the involved parties. Thus, let us dene player is utility
differential of a set of links l # E(G) deleted from the network G as
mu
i
(G; l) := u
i
(G) u
i
(G l): (9)
where G l denotes the network obtained by deleting the set of
links l # E(G) from network G. Similarly, G l denotes the network
obtained by adding the set of links l # E(G
c
)E(G) to G. In the fol-
lowing denition G
c
denotes the complete undirected network.
Denition 3.1 Jackson and Wolinsky (1996). A network G ([N[
is pairwise stable (PS) if
(i) for all ij E(G):mu
i
(G, ij) P0 and mu
j
(G, ij) P0, and
(ii) for all ij E(G
c
)E(G):mu
i
(G ij,ij) > 0 = mu
j
(G ij, ij) < 0.
Thus, a network is pairwise stable, if (i) no player has an incen-
tive to delete one of her links and (ii) there does not exists a pair of
players who want to form a mutually benecial link. Pairwise sta-
bility can be reformulated in terms of a simple test performed on
the set of all possible links. Dene
ADD(G) := ij E(G
c
) E(G)[mu
i
(G ij; ij)
> 0 & mu
j
(G ij; ij) P0; (10)
the set of links that can be added to G, and similarly
DEL(G) := ij E(G)[k i; j : mu
k
(G; ij) < 0 ; (11)
the set of links that can be deleted from G. Then G is a pairwise sta-
ble network if and only if ADD (G) = DEL (G) = O.
One potential limitation of Pairwise stability is that it only con-
siders single link changes in each updating step. A natural rene-
ment is thus to consider the formation of mutually benecial
links as well as the deletion of more than one link in one period.
Such an extended stability concepts has been used in the literature
as well, running under the name of Pairwise Nash Stability.
17
Let E
i
(G) := {ij E(G)[j [N]} denote the set of player is links in G,
and E
i
(G) := E(G)E
i
(G) denote the set of links in G in which player
i is not involved. A network is pairwise Nash stable (PNS) if
(i) u
i
(G) Pu
i
(G l
i
) "l
i
E
i
(G), and
(ii) mu
i
(G ij, ij) > 0 = mu
j
(G ij, ij) < 0 "ij E(G
c
)E(G).
Thus, a network that is pairwise Nash stable is also robust
against the deletion of a set of links by any player, while pairwise
stability only considers one link at a time.
Beyond these commonly used stability notions, several rene-
ments have been introduced in the literature. Some concepts,
which we do not discuss in this survey, are the following:
v strong and weak stability (Dutta & Mutuswami, 1997),
v bilateral stability (Goyal & Vega-Redondo, 2007),
v pairwise stability with transfers (Bloch & Jackson, 2007),
v strict pairwise stability (Chakrabarti & Gilles, 2007),
v unilateral stability (Buskens & van de Rijt, 2008)
v cognitive stability (Gallo, 2012), and
v (strict) Nash stability (supported by (strict) Nash equilibrium in
a link formation game due to Myerson (1991)).
3.3. Existence of stable networks
The concept of pairwise stability is one of the most important
equilibrium concepts used in modern network theory. In this sec-
tion we address the question of existence of PS-networks using
two different techniques. The rst technique uses the potential
function method (Monderer & Shapley, 1996; Nisan, Roughgarden,
Tardos, & Vazirani, 2007). The second technique emphasizes the
incentive structure directly. We start with the important concept
of an improving path.
Denition 3.2. An improving path from network G to network G
/
is
a sequence of networks (G
1
, . . . , G
K
) such that G
1
= G, G
K
= G
/
, and
for all 1 6 k < K, it holds that either
(i) G
k+1
= G
k
ij and ij DEL (G
k
), or
(ii) G
k+1
= G
k
ij and ij ADD (G
k
).
Hence, in an improving path each graph G
k+1
is obtained by
either deleting a single link, contained in DEL (G
k
), from the prede-
cessor G
k
, or adding a single link, contained in ADD (G
k
), to the pre-
decessor G
k
. The algorithm is running as long as there exists a
player who wants to delete a link, or there exists a pair of players
who want to form a link connecting them. It is clear that if this
algorithm converges, then the limit network must be pairwise sta-
ble. Therefore, the set of improving paths emanating from a PS net-
work is empty. Based on this concept, Jackson and Watts (2001)
dene a cycle C as an improving path (G
1
, . . . , G
K
) such that
G
1
= G
K
. They call a cycle C closed, if for all networks G C there
does not exists an improving path leading out of it. Jackson and
Watts (2001) show that for any prole of utility functions u, there
exists at least one pairwise stable network or a closed cycle of net-
works. The idea to prove existence of a PS network is now to ex-
clude the existence of cycles. A well-known sufcient condition
ruling out cyclic behavior in non-cooperative games is the exis-
tence of a potential function (Monderer & Shapley, 1996). In a sim-
ilar way, Jackson and Watts (2001) rule out the existence of
improving cycles by imposing the existence of a function w which
could be called a potential function as well.
Proposition 3.3 Jackson and Watts (2001). If there exists a function
w : ([N[ R such that ij ADD (G) if and only if w(G ij) > w(G) and
ij DEL (G) if and only if w(G ij) > w(G) for all G ([N[, then there
exist no cycles. If u exhibits no indifference then there exist no cycles
only if such a function exists.
The no-indifference-condition ensures that for two networks
G, G
/
, separated by one link, the two players affected by that link
have a strict preference ordering between the two networks. The
function w acts as an ordinal potential function.
18
Ordinal potentials
must respect the incentive structure of the underlying utility func-
tions, and thus not every game will have an ordinal potential. As
an example, the connections model of Section 3.1.1 has an ordinal
potential only for certain parameter values (see Jackson & Watts,
2001).
As already indicated, the existence of an ordinal potential
function is stronger than necessary to show existence of pairwise
stable networks. Moreover, it is not easy to check whether a utility
17
Pairwise Nash stability was rst discussed in Jackson and Wolinsky (1996). It has
been used (without formal denition) in Goyal and Joshi (2003), Belleamme and
Bloch (2004) and has been dened in, e.g. Gilles (2005), Bloch and Jackson (2006), and
Calv-Armengol and Ilkili (2009).
18
Existence of such a function is implied by the existence of an ordinal network
potential. See also Chakrabarti and Gilles (2007) and Tardos et al. (2007).
function delivers such a function.
19
Externality conditions on
marginal utilities are, however, easy to check. Hellmann (2013) uses
such externality conditions directly to prove an existence result of
PS-networks. Say a prole of utility function exhibits ordinal convex-
ity in own links if
mu
i
(G ij; ij) P0 = mu
i
(G l
i
ij; ij) P0
for all G ([N[; l
i
#E
i
(G
c
) E
i
(G) and for all ij E
i
(G
c
)E
i
(G l
i
).
20
In
words, if in G the link ij is protable for player i, then it remains
protable in the graph G
/
= G l
i
which is obtained from G by adding
a set of links l
i
. Say a utility function exhibits ordinal strategic
complements (Goyal & Joshi, 2006; Hellmann, 2013) if, for all
G ([N[, for all l
i
# E
i
(G
c
)E
i
(G), and for all ij E
i
(G
c
)E
i
(G) one has
mu
i
(G ij; ij) P0 = mu
i
(G l
i
ij; ij) P0:
In an environment in which players utility functions exhibit
ordinal convexity in own links and the ordinal strategic comple-
ments property, Hellmann (2013) reports the following result.
Proposition 3.4 Hellmann (2013). If a prole of utility functions
u = (u
1
, . . . , u
n
) satises ordinal convexity in own links and the ordinal
strategic complements property, then:
(1) There does not exist a closed improving cycle.
(2) There exists a PNS (and hence also a PS) network.
Compared to Proposition 3.3 cyclic behavior is not ruled out
completely by assuming ordinal convexity and the ordinal strate-
gic complements property. Only the existence of closed cycles is
excluded.
Example 3.5. An example, where the conditions of Proposition 3.4
are satised, is given by u
BC
dened in Section 3.1.2, which may also
include linking costs. Applying Proposition 3.4, there exists a PNS
network. Moreover, because all players have the same utility
function, one can easily conclude that if the empty network is not
PNS then the complete network is uniquely PNS and if the complete
network is not PNS then the empty network is uniquely PNS.
Remark 3.6. When imposing negative externalities (i.e. ordinal
concavity and ordinal strategic substitutes), existence of a PS net-
work cannot be guaranteed, but several uniqueness conditions
are satised. Hellmann (2013) shows that no pairwise stable net-
work can contain or can be contained in another pairwise stable
network. Furthermore conditions for existence of a globally unique
stable network are given (for details see Hellmann, 2013).
3.4. Stable networks in the connections model
To derive structural properties of stable networks beyond exis-
tence and uniqueness, more assumptions have to be imposed on
the players utility functions. A particular nice environment is
described by the connections model (Jackson & Wolinsky, 1996).
Let us focus on the special case, known as the symmetric connections
model
u
Co
i
(G) =
X
ji
d
d
ij
(G)
cg
i
(G): (12)
For this utility function Jackson and Wolinsky (1996) partially char-
acterize the set of pairwise stable networks.
21
Proposition 3.7 Jackson and Wolinsky (1996). Suppose the utility
function is given by (12). Any pairwise stable network has at most one
non-singleton component. For c < d d
2
the complete network is
uniquely pairwise stable, for d d
2
< c < d the star is among the stable
networks and for c > d the empty network is among the stable
networks and no stable network can contain a player with only one
link.
The star network, as visualized in Fig. 1(i), is the network where
one player (the center) has links to all other players (the periphery)
and those players do not have any other links. Beside the star net-
work, many more networks can be pairwise stable.
For the general utility function (5), Jackson and Wolinsky
(1996) do not report any stability results. Johnson and Gilles
(2000) study a version of the connections models, allowing for het-
erogeneity in costs, but homogeneity in benets. In their setting
players are ordered by name from 1, . . . , N such that the cost of a
link between i to j is linearly increasing in the number of players
l
ij
:= [{k [N]:i < k 6 j}[. Thus, in Eq. (5) c
ij
is replaced by l
ij
c. For
high costs c similar features of PS networks as in Proposition 3.7
can be observed. If instead costs are low (c < d d
2
), then there al-
ways exist k networks where players form links to all players
j [N] such that l
ij
6 k and no other links are formed. Thus, in these
k networks links remain local implying that clustering is quite
high, similar to the Small-World architectures displayed in
Fig. 1(iii) (Watts, 1999; Watts & Strogatz, 1998).
A different way of introducing cost heterogeneity is the island
model by Jackson and Rogers (2005). There it is assumed that the
player set [N] is partitioned into islands of equal size J such that
cost of linking is small c
ij
= c if i and j are from the same island,
but are high c < C = c
ij
if i and j belong to different islands. More-
over, benets are only obtained for small distances, i.e. w
ij
= 1 if
d
ij
(G) 6 D and w
ij
= 0 else. The interesting case occurs when
c < d d
2
and C < d + (J 1)d
2
. Then, in any PS network the islands
are (internally) completely connected and there is a path between
any two islands. Moreover, stable networks have low diameter
(bounded above by D + 1) and high clustering coefcients. Thus,
this model matches some key statistics empirically observed in
real word networks.
Persitz (2010) allows for heterogeneous benets w
ij
in the con-
nections model. In his model the players are characterized by
either one of two types t
i
{a, b} such that benets depend on
the types: w
ij
= w
1
if t
i
= t
j
= a, w
ij
= w
2
if t
i
= a, t
j
= b, and w
ij
= w
3
if
t
i
= t
j
= b such that w
1
> w
2
> w
3
. Thus all players prefer the connec-
tion to a type-a player over the connection to a type-b player.
While the trivial networks (complete respectively empty) are sta-
ble for very low respectively high costs, stable networks for med-
ium cost levels have the core periphery architecture where all
a players (the core) are completely connected (if (d d
2
)w
1
> c)
and the b players (the periphery) have connections to the core.
Whether b-players connect to each other or how many links to
a players they have, depends on the exact cost structure. Hence,
stable networks in Persitz (2010) have low diameter and high clus-
tering coefcients and therefore share some statistical features of
stable networks in Jackson and Rogers (2005).
In an incomplete information setting similar to McBride (2006)
(see Section 3.5), Gallo (2012) also shows that stable networks
19
For recent efforts to characterize the class of potential games see Sandholm
(2010b) and Candogan, Menache, Ozdaglar, and Parrilo (2011).
20
For the various denitions of convexity of a network utility function the reader is
referred to Goyal and Joshi (2006), Bloch and Jackson (2007), Calv-Armengol and
Ilkili (2009), Hellmann (2013). It is shown in Hellmann (2013) that all these
denitions are equivalent. The weaker ordinal version is dened in Hellmann (2013).
21
Although the issue of efcient networks and the tension between stable and
efcient networks displays a recurrent theme in the literature, it is beyond the scope
of this survey to review the results on efciency. The reader may be referred e.g. to
Jackson (2005); Buechel and Hellmann (2012) and textbooks by Goyal (2007) and
Jackson (2008).
exhibit low diameter (smaller than 5) and high clustering coef-
cient.
22
In fact, Gallo (2012) studies a slightly more general utility
function than eq. (5), called distance-based utility function. This
class of utility functions were rst introduced in Bloch and Jackson
(2007) and also studied in Mhlmeier, Rusinowska, and Tanimura
(2013). Thus, by generalizing the utility function from relatively sim-
ple settings, like the homogeneous connections model, to more
sophisticated models, stability predictions may reect more realistic
characteristics like low diameter and high clustering.
3.5. One-sided network formation
Most of the concepts introduced so far were formulated for
undirected networks. There are many applications where directed
networks seem to be more appropriate. Examples of directed net-
works include citation networks, hyperlinks connecting web-
pages, ow of knowledge, or more general any communication
network. When network formation can be done without mutual
consent, it is natural to speak about one-sided link formation. In a
very inuential paper, Bala and Goyal (2000) introduce a game the-
oretic model of one-sided network formation. They distinguish be-
tween two basic cases:
One-Way ow: Costs and benets are only incurred by the ini-
tiator of a link.
Two-Way ow: Costs are carried by the initiator of a link, while
benets are shared by both parties.
A pure strategy for player i is a list s
i
S
i
= {0, 1}
N{i}
, with the
interpretation that player i wants a link with j if and only if
s
ij
= 1. Since no consent is required to link to a given player, strat-
egies directly translate into a directed network G ([N[ such that
(i, j) E(G) if and only if s
ij
= 1.
23
In the one-way ow model, each player i receives benets from
all players who are accessible to i. Dene the set C
i
(G) := {j [N]
[$j
1
. . . j
k
[N]: j
1
= j, j
k
= i, (j
l
, j
l+1
) E(G) "1 6 l 6 k 1}, and call
l
i
(G) := [C
i
(G)[. In the one-way ow model of Bala and Goyal
(2000) the utility of player i [N] is given by
u
OW
i
(G) := /(l
i
(G); g
i
(G)); (13)
where g
i
(G) := [j [N[[(i; j) E(G)[ is the outdegree of player i, and
/(, ) is strictly increasing in the rst argument and strictly decreas-
ing in the second argument.
In the two-way ow model both linked players receive benets,
making it necessary to dene the closure of a directed network G.
This is the network G, dened by the property that (i; j) E(G) if
and only if (i, j) E(G) or (j, i) E(G). The set of players who are
accessible to i in the two-way ow model, C
i
(G), is then simply
the component to which player i belongs. Utility in the two-way
ow model is given by
u
TW
i
(G) := /(l
i
(G); g
i
(G)); (14)
such that above assumptions on / holds.
Since in one-sided network formation consent to link is not re-
quired, the concept of Nash equilibrium is more appealing than in
the undirected case. Formally, a Nash network is a network G such
that no single player has an incentive to create, respectively delete,
a subset of his links, given the links announced by all other players.
Given this denition, some of the ndings of Bala and Goyal (2000)
are summarized by the following proposition.
Proposition 3.8 Bala and Goyal (2000). Nash networks are either
minimally connected or empty in both the one-way ow model (13)
and the two-way ow model (14). Among those, only the empty
network and the wheel network are strict Nash networks for the
one-way ow model, and only the empty network and the
center sponsored star are strict Nash networks for the two-way ow
model.
24
The Bala and Goyal (2000) model has been extended along
various directions. In the framework of the one-way ow model
(13), Galeotti (2006) allows for heterogeneous benets and for
heterogeneous cost of connecting. Galeotti, Goyal, and Kam-
phorst (2006) allow for heterogeneity in the two-way ow model
(14). Both papers show that the basic equilibrium architectures
obtained in Bala and Goyal (2000) are quite robust. Even with
heterogeneous benets and cost, Nash networks are either min-
imal or empty, and also strict Nash networks share the same ba-
sic architectures as the homogeneous case, but can consist of
more than one non-singleton component such that the different
components consists of the basic equilibrium structures.
25
Intro-
ducing incomplete information, so that players can only observe
those parts of the network within some distance x, McBride
(2006) shows that for low enough values of x cycles may occur
(i.e. equilibrium networks are not necessarily minimal anymore)
and Nash networks may be disconnected in the two-way ow
model.
The fact that Nash networks are always minimal in the com-
plete information setting is driven by the assumption that players
benet from all accessible players equally, no matter how long
each path is. Bala and Goyal (2000) therefore introduce decay for
both the one-way ow and two-way ow model in a similar fash-
ion as the connections model (see Eq. (12)). For the latter case they
dene
u
TWd
i
(G) = 1
X
jC
i
(G)i
d
d
ij
(G)
g
i
c; (15)
with c being some positive constant. As a result, Bala and Goyal
(2000) get very similar architectures as Jackson and Wolinsky
(1996) for the connections model. Strict Nash networks are for
small costs (c < d d
2
) the complete network, for medium costs
(d d
2
< c < d) the star among others, and for high costs c > d
the empty network among others. Hojman and Szeidl (2008)
extend this approach and assume payoff structure u
HS
i
(G) =
f a
1
d
1
. . . a
D
d
D
( ) g
i
c where d
k
:=[{j [N]:d
ij
= k}[ such that f is
increasing and concave. Under some additional assumptions, they
show that the periphery sponsored star and an extended star are
unique Nash networks.
26
Another extension is due to Feri and
Melndez-Jimnez (2013) where co-evolution of networks and play
(also cf. Section 4.2) in the setup of the two-way ow model with
decay (15) is studied. In this paper, the decay is link specic and
endogenous since it depends on the actions taken by the (involved)
players in a coordination game.
22
Gallo (2012) renes the concepts of pairwise stability to cognitive stability to
account for the incomplete information setting.
23
Recall that ([N[ is used as general representation of the set of possible networks;
The direction of links is then clear from the context.
24
The notion of a minimal connected network depends on whether the network is
directed or undirected. A centersponsored star and a wheel network are depicted in
Fig. 1 and (ii). See Bala and Goyal (2000) for precise denitions.
25
In Galeotti et al. (2006), this holds when costs are player specic. However, when
costs are allowed to be heterogeneous with respect to links (called partner specic
costs), then different structures may emerge as strict Nash equilibria.
26
The periphery sponsored star is such that there exists a player i [N] (the center)
such that E(G) = {(j, i)[j [N], j i}, i.e. all links are directed from the peripheral
players to the center. In an extended star there is also a central player who does not
support any links. All peripheral players maintain a single link and from every
peripheral player there is a directed path to the center.
4. Evolutionary and co-evolutionary processes
4.1. Evolution
In the context of the connections model (cf. Sections 3.1.1 and
3.4), a natural adjustment dynamics is described by the improving
paths (Denition 3.2). Watts (2001) investigates the long-run
properties of such a dynamic explicitly. For intermediate costs such
that the star network is pairwise stable in the static model (i.e.
ensuring that the star is always an absorbing state of the dynamic
process), the likelihood that the process starting from the empty
network converges to the star goes to 0 as the number of players
N grows. The reason is that the myopic best-response dynamics
is path dependent. Once two distinct pairs of players form a link
(which is benecial when starting from the empty network), the
improving path will never lead to the star network.
To avoid path dependency Jackson and Watts (2002a) introduce
perturbations in the decisions of players in the spirit of evolution-
ary game theory (Kandori, Mailath, & Rob, 1993; Young, 1993).
These perturbations are interpreted as mistakes the players may
make when evaluating the protability of link. Formally the timing
of the process is as follows:
1. At each point in time t N a network G
t
([N[ is given and one
link ij E(G
c
) is selected according to a probability distribution
with full support on E(G
c
);
2. The link ij is added if ij ADD (G
t
); it is deleted if ij DEL (G
t
),
and it is not changed otherwise;
3. With probability > 0 the decision is reversed.
For every 0 < < 1 this process denes an irreducible Markov
chain on the set of networks ([N[. It therefore has a unique invari-
ant distribution l
, which describes the probability with which a

given network can be observed. From which network the process
starts does not inuence this probability. The networks G ([N[
such that lim
?0
l
(G) > 0 are called stochastically stable. These are

the networks observed most of the time, when the noise limit goes
to 0. Naturally, only networks contained in recurrent classes of the
Markov process with zero noise are candidates for stochastic sta-
bility. Hence, the concept of stochastic stability is an evolutionary
selection mechanism among the stable networks (and closed
cycles).
Jackson and Watts (2002a) show that stochastically stable net-
works can be detected by counting mistakes involved in the tran-
sition from one recurrent class to another.
27
However, the
construction of these trees for every recurrence class, and the calcu-
lation of minimal number of mistakes can be quite complex. There-
fore, Tercieux and Vannetelbosch (2006) present conditions for
pairwise stable networks (or sets of networks) to be stochastically
stable that are easy to check, called p-pairwise stability. A network
G is said to be p-pairwise stable, if altering a fraction p of the
N(N 1) possible links in network G implies that any improving path
starting from the altered network leads back to G (in particular pair-
wise stable networks are at least 0-pairwise stable). It is shown that
1/2-pairwise stable networks (if existing) are uniquely stochastically
stable.
Dawid and Hellmann (2012) study the evolution of R&D collab-
orations in the model setting of Goyal and Joshi (2003) (see
Section 3.1.3). It has been shown by Goyal and Joshi (2003) that
any PNS network is of dominant group architecture; i.e. there
is one completely connected group and all other players are iso-
lated (the same is true for PS networks, see Dawid & Hellmann,
2012). The sizes of the dominant group are sensitive to the cost
of link formation, but there is no unique prediction with respect
to the networks which will be observed. When introducing the
evolutionary process above it can be shown that the size of the
dominant group in stochastically stable networks is generically un-
ique and monotonically decreasing in cost of link formation. In the
context of the evolution of R&D collaborations, these unique pre-
dictions give way to interesting comparative static analysis with
respect to welfare, industry prots, and consumer surplus.
In the context of directed networks Bala and Goyal (2000) pres-
ent a myopic best response process in discrete time for the one-
way ow model (13) and two-way ow model (14). They assume
that at any time step t N each player i [N] chooses a myopic
best response to last periods network with probability 0 < p
i
< 1
while with probability 1 p
i
exhibits inertia.
28
For both models
the myopic best response process always converges to the set of
strict Nash networks (see Proposition 3.8).
Feri (2007) adds random perturbations to this dynamics and re-
stricts himself to the two-way ow model with decay (see (15)).
Feri (2007) obtains an almost complete characterization of the sto-
chastically stable networks. Trivially, for low respectively large
costs, only the complete respectively empty network are stochasti-
cally stable. For intermediate costs, and large networks, all stars
where each link is supported by only one player are among the sto-
chastically stable networks while if costs are slightly higher, no
other network is stochastically stable. These results stand in con-
trast to Watts (2001), who shows that for the connections model
in bilateral network formation the likelihood of emergence of the
star decreases in the number of players N for the pure best re-
sponse dynamics.
4.2. Co-evolution of networks and play
Having discussed models of pure network formation, we nowex-
tendour model to allowthe players to control (at least partially) two
dimensions of the model. These twodimensions consist of the links a
player supports ina graph, andanactionshemaychoose, whichgen-
erates a certain payoff according to an underlying game in normal
form. Already the basic examples of Section 3.1 can be used to moti-
vate an analysis of such co-evolutionary processes. For instance, in
the local complementarity game of Section 3.1.2, the effort choice
of a player depends on its position in the networks. As new links
are added or deleted, this will have an effect on the effort invested
by a player, and vice versa. We now discuss a mathematical model
in which such joint dynamics can be analyzed.
4.2.1. Games with local interaction structure
Recall that a normal form game is a tuple [N], (u
i
)
i[N]
, (A
i
)
i[N]
),
where [N] = {1, 2, . . . , N} is the set of players, and u
i
:
i[N[
A
i
R is
the utility function of player i, assigning a utility index u
i
(a
1
, . . . , a
N
)
to each action prole a := (a
1
, . . . , a
N
)
i[N]
A
i
. In this general
denition the interaction structure of the players is hidden in the
utilityfunction. Inorder tomakethis dependencystructuremoreex-
plicit, it is useful to separate these effects and dene a preference
relation directly on the product set of action proles and interaction
structures (i.e. networks). Todoso, let us redene the strategic inter-
actionas agamewithlocal interactionstructure (Morris, 1997), which
is a game in normal form, but the preference relation is directly
dened over a product set A
i
= S
i
([N[. Such models have a quite
long tradition (early contributions are Blume, 1993; Ellison, 1993).
27
The arguments used in Jackson and Watts (2002a) are based on the classical
Freidlin and Wentzell (1998) tree constructions to understand the long-run behavior
of dynamical systems subject to random perturbations. These methods have found
numerous application in game theory. See Young (1998), Sandholm (2010a), and
Staudigl (2012).
28
Note that compared to the best-response dynamics with two-sided link formation
in Watts (2001) and Jackson and Watts (2002a), the revision process is now
formulated on the individual player level, rather than on the set of edges.
4.2.2. Adding network dynamics
Starting from a given local interaction game, we now construct
an explicit dynamic model, which includes a dynamic model of
network formation as well as a dynamic model of action revision.
The models we will discuss are based on the following hypothesis:
(i) Agents are myopic in their linking and action decision (cf.
Section 4.1).
(ii) The dynamic process is stationary.
Hypothesis (i) is a huge simplication and essentially boils
down to the hypothesis that players base their decisions today
only on the current information they have. They do not attempt
to forecast the impact of their behavior on the future evolution
of the state and its consequence on their own payoff. Such an
assumption is standard practice in evolutionary game theory (Wei-
bull, 1995). Together with hypothesis (ii), it allows us to model a
co-evolutionary process as a time-homogeneous Markov processes
living on the product set A = S ([N[, where S =
i[N]
S
i
is the
set of action proles. Hence, a co-evolutionary model with noise
is a family of continuous-time Markov jump processes
X
(t)
tR
n o
(0;[
, dened on the nite state space A. The stochas-
tic process X
is indexed by the parameter > 0. This parameter

often has the interpretation of a behavioral noise parameter
(cf. Section 4.1).
A realization {X
(t) = x} denes an action prole r(x) =

(r
i
(x))
i[N]
S, and a network c(x) ([N[. In concrete examples it
is often more convenient to encode the network via its adjacency
matrix. With a slight abuse of notation we denote the adjacency
matrix of the network c(x) by A(x).
The elementary updating steps of the Markov process are the
following:
Action adjustments: With constant rate 1 the action prole
r(x) is allowed to change. Conditional on this event, player
i [N] is chosen uniformly at random to update his current
action. If i gets a revision opportunity, he draws an action from
a distribution b
i
([x) D(S
i
). The mapping b
i
: A D(S
i
) is
called the choice function of player i.
Link creation: With constant rate k P0 the process allows the
network to expand. As in Section 2 we model this process by a
matrix-valued function W
: A R
NN
, whose elements are
bounded non-negative functions w
ij
(x), interpreted as the inten-
sity of link creation given the current state x A. For each x A
and > 0 the matrix W
(x) is symmetric, and called the attach-

ment mechanism of the process.
Link destruction: With constant rate n P0 a link becomes
destroyed. The intensity with which the link between player i
and j is destroyed is modeled by matrix-valued function
V
: A R
NN
, whose elements are bounded non-negative
functions v
ij
(x). For each x A and > 0 the matrix V
(x) is sym-
metric, and called the volatility mechanism of the process.
Co-evolutionary processes are general enough to cover most
evolutionary models which have been studied in the literature. In
particular, evolutionary learning models on xed interaction struc-
tures are obtained by specializing the setting to k = n = 0. Second, it
is instructive to see that X
is indeed an extension of the random

graph process of Section 2. The added feature is that the attach-
ment and the volatility mechanism are now functions dened on
the domain A.
4.2.3. A micro-founded model for inhomogeneous random graphs
To give a characterization of the class of networks generated by
co-evolutionary processes, we would like to rely on our knowledge
about general random graph models. In particular we would like to
use the general characterization Theorem 2.1. For a given prole
s S, let us dene the s-section of the state space A as the set
A
s
:= s ([N[. An s-conditional random graph process is a con-
tinuous-time Markov process {c
(t)}
tP0
taking values in the set
A
s
, with attachment mechanisms W
[
As
and V
[
As
. On an s-section
the intensities can only vary with the network. Our goal is to derive
a probability law dened over the set of possible networks, condi-
tional on the event that the players play according to the action
prole s, which we denote by P(s).
Assumption 4.1. The co-evolutionary process X
satises the
following assumptions:
(i) k, n > 0;
(ii) If A
ij
(x) = 0 and > 0, then w
ij
(x) > 0. If A
ij
(x) = 1 or i = j, then
w
ij
(x) = 0 for all ;
(iii) For all pairs of players i,j [N] and states x A we have
w
ij
(x) = j
ij
(r(x))(1 A
ij
(x)), and v
ij
(x) = d
ij
(r(x))A
ij
(x),
where j
ij
and d
ij
are positive functions.
The set of assumptions is not minimal in order to be able to ap-
ply Theorem 2.1 to characterize the conditional random graph
measure. Item (ii) is more restrictive than actually needed. The
only requirement we need is that the generator of the conditional
random graph process satises the necessary irreducibility
assumptions in order to guarantee the existence of a unique invari-
ant measure. However, item (iii) is necessary.
Item(iii) of Assumption 4.1 is arguably the most restrictive one. It
requires that the intensities of link creation and destruction are func-
tions only of the given action prole. One can imagine examples
where this assumption makes sense, but it is clear that many exam-
ples will not t this description. Nevertheless we can give the follow-
ing characterizationof the conditional randomgraphmeasure, which
is obtained froma straightforward application of Theorem 2.1.
Proposition 4.2. Consider a co-evolutionary process X
as dened
above whose attachment and volatility mechanism satises Assumption
4.1. Conditional on an action prole s S, the conditioned random
graph process {c
(t)}
tP0
has a unique invariant graph measure agreeing
with the probability measure of an inhomogeneous random graph
P
(G[s) =
Y
N
i=1
Y
j>i
(p
ij
(s))
A
ij
(G)
(1 p
ij
(s))
1A
ij
(G)
;
where p
ij
(s) :=
j
ij
(s)
j
ij
(s)d
ij
(s)
is the edge-success probability of vertex i and j.
4.3. Examples of co-evolutionary processes
Let us now present some simple examples of co-evolutionary
processes of networks and play. The rst example is a version of
a model due to Jackson and Watts (2002b). The second example
is a version of Staudigl (2011, 2013).
4.3.1. A co-evolutionary model based on pairwise stability
Jackson and Watts (2002b) combine the network formation
model of Jackson and Wolinsky (1996) (see Section 3.2) with the
dynamic network formation model due to Watts (2001), (see
Section 4.1).
29
The local interaction game is a symmetric 2 2
29
Jackson and Watts (2002b) describe a dynamic evolving in discrete periods. The
continuous-time version of their model given here differs in no substantial details,
because the stochastic law of motion is the one of a Markov jump process. The
discrete time models can be seen as embedded jump chains of the continuous-time
model. See, e.g. Stroock (2005) for a general reference, and Staudigl (2010) for an
explicit construction.
coordination game with action set S
1
= S
2
= {1, 2}. For each link
formed by a player, a xed marginal cost / > 0 has to be paid. The
utility function of player i is given by
u
i
(s; G) =
X
ji
p(s
i
; s
j
)A
ij
(G) /g
i
(G);
where p(1; s) =
a ifs = 1;
b ifs = 2
and p(2; s) =
c ifs = 1;
d ifs = 2:
The tuple
(a, b, c, d) are the payoff parameters of the underlying game,
satisfying the ordering a > c, b > d, and min{c,d} > 0.
30
Note that,
for a xed prole of actions, the marginal utility of the link ij for
player i is given exactly by mu
i
(G(x) ij,ij) = p(r
i
(x), r
j
(x)) /. The
co-evolutionary process is set up as follows:
Action adjustment: With rate 1 each player receives the oppor-
tunity to revise his action. Conditional on this event he selects
action s with probability
b
i
(s[x) =
1 ifs = arg max
s
/
1;2
u
i
((s
/
; r
i
(x)); G(x));
1 if r
i
(x) = s and r
i
(x) arg max
s
/
1;2
u
i
((s
/
; r
i
(x)); G(x));
otherwise:
8
>
>
>
<
>
>
>
:
This choice function says that a player abandons his currently used
action with relatively high probability 1 , if there exists a strictly
better action. Otherwise he sticks to his action and switches only
with the relatively small probability .
31
Link creation: With rate
k > 0 a link becomes created. Jackson and Watts (2002b) model
network formation in the avor of pairwise stability as dis-
cussed in Section 3. Let
ADD(x) = ij E(G
c
) E(G(x))[p(r
i
(x); r
j
(x))
> /; and p(r
j
(x); r
i
(x)) P/
be the set of links that are mutually protable. Similarly we dene
DEL(x) = ij E(G(x))[p(r
i
(x); r
j
(x)) < /; or p(r
j
(x); r
i
(x))
< /:
Jackson and Watts (2002b) assume that a previously non-existing
link becomes active with probability 1 if both players mutually
agree. With the small probability all links have a chance to be
formed. The rate that a currently non-existing link ij will be
added is
\(i; j) R E(G(x)) : w
ij
(x) :=
1
[ADD(x)[

[E(G
c
G(x))[
if ij ADD(x);
[E(G
c
G(x))[
otherwise:
(
Link destruction: With rate n > 0 links become destroyed. Con-
ditional on this event, pick one edge ij E(G(x)) uniformly at
random and allow the incident players to re-evaluate the bene-
ts arising from this connection. If ij is a link where at least one
player is better off after its deletion it is assumed that with large
probability 1 it will be destroyed. With the small probability
every active link can be destroyed. This leads to the following
version of the volatility mechanism:
\(i; j) E(G(x)) : v
i;j
(x) =
1
[DEL(x)[

[E(G(x))[
if ij DEL(x);
[E(G(x))[
otherwise:
(
Clearly this model is a version of a co-evolutionary process. The
attachment and volatility mechanism depends however in a non-
trivial way on the current network and the action chosen by the
players, so that we cannot apply Proposition 4.2 to this model.
However, this model can still be quite well understood in the ex-
treme case where ?0. Taking this limit, Jackson and Watts
(2002b) show that the complete network will be observed most
of the time (i.e. it is a stochastically stable state), and the players
coordinate on a single strongly symmetric action prole. The inter-
esting open question is which of the two symmetric equilibria will
be played. Jackson and Watts (2002b) provide an exhaustive char-
acterization of the long-run equilibria in this model.
32
Proposition 4.3 (Preposition 1, Jackson and Watts (2002b)). Sup-
pose that N is sufciently large. The unique stochastically stable states
of the co-evolutionary model with constant linking costs / > 0 consist
of the complete network and the following pattern of play:
(i) If a / > 0 and d / > 0 then either all players play action 1 or
action 2.
(ii) If b / > 0 and c / > 0 then all players play action 1.
(iii) If b / < 0 and/or d / < 0, and a / > 0 and d / > 0, then
either all play 1 or all players play 2.
Related to the Jackson and Watts (2002b) model is the coordi-
nation game scenario investigated by Goyal and Vega-Redondo
(2005). Their paper is based on the two-way ow model of Bala
and Goyal (2000), but adds the additional layer of complexity by
allowing agents to derive utility from a coordination game as in
Jackson and Watts (2002b). In the spirit of the two-way ow mod-
el, the unit of a one-step transition is not the edge, but the single
individual. Specically, Goyal and Vega-Redondo (2005) envision
a dynamic co-evolutionary process in which at each time step a
single individual receives a revision opportunity (independent of
the other players) in which she can simultaneously change her
action played in the coordination and the set of links she would like
to support in the network realized in the next period. Once a
network is realized, the agents play the coordination game p with
all their neighbors. Thus, if G is the closure of the network G
(cf. Section 3.5), player is total instantaneous payoff is
P
ji
p(a
i
; a
j
)A
ij
(G). Each initiated link has some costs / > 0, which
have to be paid only by the initiator. Hence, the utility function
in the Goyal and Vega-Redondo (2005) model is
u
i
(s; G) =
X
ji
p(a
i
; a
j
)A
ij
(G) /g
i
(G):
Using this payoff specication, the main result of Goyal and
Vega-Redondo (2005) is the following.
Proposition 4.4 (Theorem 3.1, Goyal and Vega-Redondo
(2005)). Suppose the stage game p is a coordination game with data
a + b < c + d, and a > d. There exists a value

/ (b; d) such that for
/ <

/ the unique stochastically stable conguration is the complete
network in which all agents play action 2. If / (
/; a) all players play

1 and the network is complete for N sufciently large.
Contrasted with the Jackson and Watts (2002b) model we see
that one-sided link formation leads to sharper predictions as the
noise in the players decision function is small and the society is
sufciently large.
30
This means that the action proles (1,1) and (2,2) are (strict) Nash equilibria of
the bilateral game.
31
This choice function uses a tie-breaking rule. Namely, if both actions are optimal
for player i at x, then the choice function says that i continuous playing his current
action.
32
In the proposition we require the number of players to be sufciently large. This
ensures that certain special cases are ruled out. See Jackson and Watts (2002b) for a
complete statement of the Proposition.
4.3.2. A volatility model
In this example we consider a family of games in which the util-
ity function u
i
consists of two components. The rst component is a
common payoff term, interpreted as the externality a single players
actions exert on the other players. The second component is an idi-
osyncratic payoff. As such it depends only on the players own
choice. We assume that all players have the same action set, de-
noted by S. Making a slight change in notation, we write S
N
for
the set of action proles. A typical element of this set is an N-tuple
~s := (s
i
)
i[N[
. The common payoff term from bilateral social interac-
tion is modeled by a reward function p : S S R. In many appli-
cations it is conceivable that the common payoff terms displays a
strong symmetry property of the form
p(s; s
/
) = p(s
/
; s) \s; s
/
S;
which we assume henceforth to hold. The idiosyncratic payoff
component is modeled via a function s
i
: S R, where s
i
is an
element of the set of functions H = h
1
; . . . ; h
m
[h
l
: S R. We
interpret this idiosyncratic component as the type of the player.
Thus, a type of a player is a function h
r
: S R, assigning to each
action s S an idiosyncratic utility h
r
(s). We assume that players
learn their types before the co-evolutionary process starts. This
assumption allows us to interpret the types of the players as param-
eters of the co-evolutionary process.
33
Given an action prole ~s S
N
and a prole of types s H
N
, the
(ex-post) payoff of player i is assumed to be
u
i
(s; G; s
i
) =
X
ji
A
ij
(G)p(s
i
; s
j
) s
i
(s
i
):
The co-evolutionary process is specied by the following data.
Action adjustment: Agents use the logit-choice function to
choose actions. This choice function is dened as
(\s S) : b
i
(s[x; s
i
) =
exp u
i
((s; r
i
(x)); G(x); s
i
)= [ [
P
s
/
S
exp u
i
((s
/
; r
i
(x)); G(x); s
i
)= [ [
:
The rate of the transition x = (~s; G) x
/
= ((s; r
i
(x)); G(x)) is thus
b
i
(s[x; s
i
).
Link creation: The attachment mechanism is given by a collec-
tion of functions j
(s; s
/
)
(s;s
/
)SS
, dened as
j
(s; s
/
) =
2
N
exp(p(s; s
/
)=) \s; s
/
S:
The rate of a transition x = (~s; G) x
/
= (~s; G ij) is then given by
k(1 A
ij
(x))j
(r
i
(x), r
j
(x)).
Link destruction: The volatility mechanism is
v
ij
(x) = d
s
i
;s
j
;
where d
k;l
= d
l;k
is the volatility rate of a link between a player of
type k and a player of type l. The rate of the transition
x = (~s; G) x
/
= (~s; G(x) ij) is then given by nA
ij
(x)d
s
i
;s
j
.
As shown in Staudigl (2013) this model can be completely ana-
lyzed using elementary arguments. The (strong) assumptions mak-
ing the model tractable are the strong symmetry of the reward
function p and the particular specication of the volatility and
attachment mechanism. Working with these assumptions, the
long-run properties of the stochastic dynamics can be completely
understood.
Theorem 4.5 (Staudigl (2011, 2013)). The unique invariant distri-
bution of the co-evolutionary process X
;s
N
(t)
tP0
is the Gibbs
measure
l
;s
N
(x) =
exp(
1
H
N
(x; s))
P
x
/
A
exp(
1
H
N
(x
/
; s))
=
l
;s
0;N
(x) exp
1
q(x; s)

P
x
/
A
l
;s
0;N
(x
/
) exp(
1
q(x
/
; s))
; (16)
where, for all x = (s; G) A,
H
N
(x; s) := q(x; s) log l
;s
0;N
(x);
l
;s
0;N
(x) :=
Y
N
i=1
Y
j>i
2
Nd
s
i
;s
j
!
A
ij
(G)
;
q(x; s) :=
1
2
X
ji
A
ij
(G)p(r
i
(x); r
j
(x))
X
i
s
i
(s
i
):
We can use the measure l
;s
N
to derive a random graph measure
over the~s-section A~s
. The random graph process {c
,s
(t)}
tP0
dened
on ([N[ for a xed prole of actions~s S
N
can be formally identied
with a birthdeath process with birth rates of the link ij given by 2
exp (p(s
i
, s
j
)/), and death-rates d
s
i
;s
j
. Calling the rate ratio
u
k;l
(s; s
/
) =
2 exp(p(s; s
/
)=)
d
k;l
;
for s, s
/
S and 1 6 k, l 6 m, Staudigl (2013) goes on in proving the
following result.
Proposition 4.5. Consider the random graph process described
above. This process is ergodic with unique invariant graph measure
P
;s
N
(G[~s) =
Y
N
i=1
Y
j>i
p
ij
(~s; s)
A
ij
(G)
1 p
ij
(~s; s)

(1A
ij
(G))
;
where the edge-success probabilities are dened as
p
ij
(~s; s) =
u
k;l
(s; s
/
)
u
k;l
(s; s
/
) N
if s
i
= s; s
j
= s
/
; s
i
= h
k
; s
j
= h
l
(17)
for all i, j [N].
The proof of this Proposition is a simple application of Proposi-
tion 4.2. The interpretation of this result is as follows. For a xed
action prole ~s and known type prole, the stationary probability
distribution over graphs will be an inhomogeneous random graph
(cf. Section 2.2.1), with edge-success probabilities dened as in
eq. (17).
34
5. Summary and suggestions for future research
In this survey we have tried to give a short overview on a partic-
ular class of models designedfor the modeling of the dynamic evolu-
tion of networks. Random graph models are combined with game
theorytoformalizetheco-evolutionof networks andplay. This mod-
eling approachreects the interdisciplinary character of researchon
dynamic networks (both theoretical and applied). The models
touched in this survey can and should be considered as only preli-
minary steps towards a more general theory. In the following lines
we give some suggestions for future research and their challenges.
33
Models with similar payoff structure are frequently encountered in economic
models with local interactions. See e.g. Brock and Durlauf (2001), and the references
therein.
34
Staudigl (2013) goes on in characterizing the supports of the invariant measure
l
;s
N
as e ?0 (the so-called stochastically stable states mentioned in Section 4.1), and
also derives the large deviation rate function of this measure in the limit of large
player sets, i.e. where N ?.
5.1. Weighted networks
As mentioned in Section 1, we have avoided any discussion of
weighted networks in this survey. The simple reason for this is that
we feel that there is not a satisfactory theory out there in the mo-
ment. Much more efforts should be devoted to this very important
class of networks, which are so prevalent in applications.
5.2. Mechanism design of networks
In many practical situations a system controller would like to
ensure that a certain network structure will be achieved via decen-
tralized decision making of the players. In such a model a control-
ler collects a map of the vertices preferences before evolution
starts, and given these data, a tax/subsidy scheme is designed so
that a certain network architecture will be the outcome (at least
with high probability).
35
It is an interesting question how such a
model extends to a dynamic evolutionary setting. We envision a
Bayesian interaction model as in Section 4.3.2 as a reasonable mod-
eling approach for this endeavor. Related to this question, we think
that it is important to invest more effort to understand the connec-
tion between structural features of the players utility functions and
the structure of stable networks (see Sections 3.3, 3.4 and 3.5), but
there is room for improvement. General characterizations of stable
networks are so far missing (e.g. necessary conditions for existence
and uniqueness or general conditions for emergence of particular
networks). An important contribution to foster our understanding
of incentive-driven network formation models is to start with a
class of network games (potential games would be a reasonable
starting point) and investigate the question how equilibrium net-
work topologies change when basic properties of the utility function
change. This is not only an important theoretical question, but may
also be important for statistical identication.
Acknowledgments
We would like to thank the editor and four anonymous referees
for their very constructive comments on earlier versions of this
survey. Mathias Staudigl acknowledges nancial from the Vienna
Science and Technology Fund (WWTF) under project fund MA
09-017.
References
Albert, R., & Barabsi, A.-L. (2002). Statistical mechanics of complex networks.
Reviews of Modern Physics, 74, 4797.
Bala, V., & Goyal, S. (2000). Non-cooperative model of network formation.
Econometrica, 68(5), 11811229.
Ballester, C., Calv-Armengol, A., & Zenou, Y. (2006). Whos who in networks.
Wanted: The key player. Econometrica, 74(5), 14031417.
Barabsi, A., & Albert, R. (1999). Emergence of scaling in random networks. Science,
286, 509512.
Barthlemy, M., Barrat, A., Pastor-Satorras, R., & Vespignani, A. (2005).
Characterization and modeling of weighted networks. Physica A: Statistical
Mechanics and its Applications, 346(12), 3443.
Belleamme, P., & Bloch, F. (2004). Market sharing agreements and collusive
networks. International Economic Review, 45(2), 387411.
Bloch, F., & Dutta, B. (2009). Communication networks with endogenous link
strength. Games and Economic Behavior, 66(1), 3956.
Bloch, F., & Jackson, M. (2006). Denitions of equilibrium in network formation
games. International Journal of Game Theory, 34(3), 305318.
Bloch, F., & Jackson, M. O. (2007). The formation of networks with transfers among
players. Journal of Economic Theory, 133(1), 83110.
Blume, L. (1993). Statistical mechanics of strategic interaction. Games and Economic
Behavior, 5, 387424.
Bollobs, B. (2008). Random graphs (2nd ed.). Cambridge University Press.
Bollobs, B., Janson, S., & Riordan, O. (2007). The phase transition in inhomogeneous
random graphs. Random Structures and Algorithms, 31(1), 3122.
Bonacich, P. (1987). Power and centrality: A family of measures. American Journal of
Sociology, 92(5), 11701182.
Brock, W. A., & Durlauf, S. N. (2001). Discrete choice with social interactions. The
Review of Economic Studies, 68(2), 235260.
Buechel, B., & Hellmann, T. (2012). Under-connected and over-connected networks:
The role of externalities in strategic network formation. Review of Economic
Design, 16, 7187.
Buskens, V., & van de Rijt, A. (2008). Dynamics of networks if everyone strives for
structural holes. American Journal of Sociology, 114, 371407.
Calv-Armengol, A., & Ilkili, R. (2009). Pairwise-stability and Nash equilibria in
network formation. International Journal of Game Theory, 38(1), 5179.
Candogan, O., Menache, I., Ozdaglar, A., & Parrilo, P. A. (2011). Flows and
decompositions of games: Harmonic and potential games. Mathematics of
Operations Research, 36(3), 474503.
Chakrabarti, S., & Gilles, R. (2007). Network potentials. Review of Economic Design,
11(1), 1352.
Chatterjee, S. & Diaconis, P. (2011). Estimating and understanding exponential random
graphs. arXiv:1102.2650v3 [math.PR].
Chung, F. & Lu, L. (2006). Complex graphs and networks. In CBMS regional conference
series in mathematics.
Dawid, H. & Hellmann, T. (2012). The evolution of R&D networks. Working Papers
467. Bielefeld University, Institute of Mathematical Economics.
Dodds, P., Muhamad, R., & Watts, D. (2003). An experimental study of search in
global social networks. Science, 301(5634), 827829.
Dorogovtsev, S., & Mendes, J. (2003). Evolution of networks-from biological nets to the
internet and WWW. Oxford University Press.
Durrett, R. (2007) Random graph dynamics. Cambridge Series in Statistical and
Probabilistic Mathematics.
Dutta, B., Ghosal, S., & Ray, D. (2005). Farsighted network formation. Journal of
Economic Theory, 122(2), 143164.
Dutta, B., & Mutuswami, S. (1997). Stable networks. Journal of Economic Theory,
76(2), 322344.
Easley, D., & Kleinberg, J. (2010). Networks, crowds and markets reasoning about a
higly connected world. Cambridge, UK: Cambridge Univ. Press.
Echenique, F., Lee, S., Shum, M., & Yenmez, M. B. (2013). The revealed preference
theory of stable and extremal stable matchings. Econometrica, 81(1),
153171.
Ehrhardt, G., Marsili, M., & Vega-Redondo, F. (2006). Diffusion and growth in an
evolving network. International Journal of Game Theory, 34, 383397.
Ellison, G. (1993). Learning, local interaction, and coordination. Econometrica, 61(5),
10471071.
Erds, P., & Rnyi, A. (1959). On random graphs i. Publicationes Mathematicae, 6,
290297.
Erds, P., & Rnyi, A. (1960). On the evolution of random graphs. Publications of the
Mathematical Institute of the Hungarian Academy of Sciences, 5, 1761.
Feri, F. (2007). Stochastic stability in networks with decay. Journal of Economic
Theory, 135(1), 442457.
Feri, F., & Melndez-Jimnez, M. A. (2013). Coordination in evolving networks with
endogenous decay. Journal of Evolutionary Economics.
Fienberg, S. E., Meyer, M. M., & Wasserman, S. S. (1985). Statistical analysis of
multiple sociometric relations. Journal of the American Statistical Association,
80(389), 5167.
Frank, O., & Strauss, D. (1986). Markov graphs. Journal of the American Statistical
Association, 81(395), 832842.
Freeman, L. (1979). Centrality in social networks: Conceptual clarication. Social
Networks, 1(3), 215239.
Freidlin, M., & Wentzell, A. (1998). Random perturbations of dynamical systems (2nd
ed.). New York: Springer Verlag.
Galeotti, A. (2006). One-way ow networks: The role of heterogeneity. Economic
Theory, 29(1), 163179.
Galeotti, A., Goyal, S., & Kamphorst, J. (2006). Network formation with
heterogeneous players. Games and Economic Behavior, 54, 353372.
Gallo, E. (2012). Small world networks with segregation patterns and brokers.
Review of Network Economics, 11(3).
Gilbert, E. N. (1959). Random graphs. The Annals of Mathematical Statistics, 30(4),
11411144.
Gilles, R. & Sarangi, S. (2005). Stable networks and convex payoffs. Discussion Paper
2005-84, Tilburg University, Center for Economic Research.
Goldenberg, A., Zheng, A. X., Fienberg, S. E., & Airoldi, E. M. (2009). A survey of
statistical network models. Foundations and Trends in Machine Learning, 2(2),
129233.
Golub, B. & Jackson, M. (2012). How homophily affects the speed of learning and
best-response dynamics. The Quarterly Journal of Economics, 127(3).
Goyal, S. (2005). Learning in networks. In G. Dema nge & M. Wooders (Eds.), Group
formation in economics: Networks, clubs and coalitions (pp. 122167). Cambridge
University Press.
Goyal, S. (2007). Connections. Princeton: Princeton Univ. Press.
Goyal, S., & Joshi, S. (2003). Networks of collaboration in oligopoly. Games and
Economic Behavior, 43(1), 5785.
Goyal, S., & Joshi, S. (2006). Unequal connections. International Journal of Game
Theory, 34(3), 319349.
35
For motivation, reconsider the local complementarity model of Section 3.1.2.
Assuming Nash equilibrium play in the game where players choose efforts, we can
now use this derived preference relation over graphs (7) to investigate the
equilibrium payoffs as functions of the interaction structure. This leads naturally to
questions of mechanism design and network optimization, both of which are very
important questions in economics and operations research.
Goyal, S., Van Der Leij, M. J., & Moraga-Gonzlez, J. L. (2006). Economics: An
emerging small world. Journal of Political Economy, 114(2), 403412.
Goyal, S., & Vega-Redondo, F. (2005). Network formation and social coordination.
Games and Economic Behavior, 50, 178207.
Goyal, S., & Vega-Redondo, F. (2007). Structural holes in social networks. Journal of
Hellmann, T. (2013). On the existence and uniqueness of pairwise stable networks.
International Journal of Game Theory, 42, 211237.
Hojman, D. A., & Szeidl, A. (2008). Core and periphery in networks. Journal of
Jackson, M. O. (2003). The stability and efciency of economic and social networks.
In M. Sertel & S. Koray (Eds.), Advances in economic design. Heidelberg: Springer-
Verlag.
Jackson, M. O. (2005). A survey of models of network formation: stability and
efciency. In G. Demange & M. Wooders (Eds.), Group formation in economics:
Networks, clubs and coalitions (pp. 1149). Cambridge University Press.
Jackson, M. O. (2008). Social and economic networks. Princeton University Press.
Jackson, M. O., & Rogers, B. W. (2005). The economics of small worlds. Journal of the
European Economic Association, 3(23), 617627.
Jackson, M. O., & Rogers, B. W. (2007). Meeting strangers and friends of friends: How
random are social networks? American Economic Review, 97(3), 890915.
Jackson, M. O., & Watts, A. (2001). The existence of pairwise stable networks. Seoul
Journal of Economics, 14(3), 299321.
Jackson, M. O., & Watts, A. (2002a). The evolution of social and economic networks.
Journal of Economic Theory, 106, 265295.
Jackson, M. O., & Watts, A. (2002b). On the formation of interaction networks in
social coordination games. Games and Economic Behavior, 41, 265291.
Jackson, M. O., & Wolinsky, A. (1996). A strategic model of social and economic
networks. Journal of Economic Theory, 71, 4474.
Johnson, C., & Gilles, R. P. (2000). Spatial social networks. Review of Economic Design,
5(3), 273299.
Kandori, M., Mailath, G., & Rob, R. (1993). Learning, mutation, and long run
equilibria in games. Econometrica, 61, 2956.
Karrer, B., & Newman, M. E. J. (2011). Stochastic blockmodels and community
structure in networks. Physical Review E, 83, 016107.
Kumpula, J. M., Onnela, J.-P., Saramki, J., Kaski, K., & Kertsz, J. (2007). Emergence of
communities in weighted networks. Physical Review Letters, 99, 228701.
Marsili, M., Vega-Redondo, F., & Slanina, F. (2004). The rise and fall of the networked
society: A formal model. PNAS, 101, 14391442.
McBride, M. (2006). Imperfect monitoring in communication networks. Journal of
McPherson, M., Smith-Lovin, L., & Cook, J. M. (2001). Birds of a feather: Homophily
in social networks. Annual Review of Sociology, 27(1), 415444.
Milgram, S. (1967). The small world problem. Psychology Today, 2, 6067.
Mhlmeier, P., Rusinowska, A., & Tanimura, E. (2013). A degree-distance-based
connections model with negative and positive externalities. Technical report. IMW,
Bielefeld University.
Monderer, D., & Shapley, L. (1996). Potential games. Games and Economic Behavior,
14, 124143.
Morris, S. (1997). Interaction games: A unied analysis of incomplete information, local
interaction and random matching. <http://ssrn.com/abstract=290880 or http://
dx.doi.org/10.2139/ssrn.290880>.
Myerson, R. B. (1991). Game theory: Analysis of conict. Harvard University Press.
Newman, M. (2003a). Hanbook of graphs and networks, Random graph models as
models of networks. Wiley-VCH (chapter 2, pp. 3565).
Newman, M. (2003b). The structure and function of complex networks. SIAM
Review, 45(2), 167256.
Newman, M. E. J. (2004). Coauthorship networks and patterns of scientic
collaboration. Proceedings of the National Academy of Science (USA), 101,
52005205.
Nisan, N., Roughgarden, T., Tardos, va, & Vazirani, V. V. (2007). Algorithmic game
theory. Cambridge University Press.
Park, J., & Newman, M. E. J. (2004). The statistical mechanics of networks. Physical
Review E, 70. 066117-1066117-15.
Persitz, D. (2010). Power and core-periphery networks. Technical report.
Robins, G., Pattison, P., Kalish, Y., & Lusher, D. (2007). An introduction to
exponential random graph (p
*
) models for social networks. Social Networks,
29(2), 173191.
Sandholm, W. (2010a). Evolutionary dynamics and population games. The MIT Press.
Sandholm, W. H. (2010b). Decompositions and potentials for normal form games.
Games and Economic Behavior, 70(2), 446456.
Sandholm, W. H., & Pauzner, A. (1998). Evolution, population growth, and history
dependence. Games and Economic Behavior, 22(1), 84120.
Snijders, T. A. B. (2001). The statistical evaluation of social network dynamics.
Sociological Methodology, 31, 361395.
Snijders, T. A. B., Pattison, P. E., Robins, G. L., & Handcock, M. S. (2006). New
specications for exponential random graph models. Sociological Methodology,
36, 99153.
Sderberg, B. (2002). General formalism for inhomogeneous random graphs.
Physical Review E, 66. 066121-1066121-6.
Staudigl, M. (2010). Co-evolutionary dynamics of networks and play. Ph.D. thesis.
University of Vienna.
Staudigl, M. (2011). Potential games in volatile environments. Games and Economic
Behavior, 72, 271287.
Staudigl, M. (2012). Stochastic stability in asymmetric binary choice coordination
games. Games and Economic Behavior, 75, 372401.
Staudigl, M. (2013). Co-evolutionary dynamics and bayesian interaction games.
International Journal of Game Theory, 42(1), 179210.
Stroock, D. (2005). An introduction to markov processes. Graduate texts in mathematics
230. Berlin, Heidelberg, New York: Springer.
Tardos, E., & Wexler, T. (2007). Algorithmic game theory, Network formation games
and the potential function method. Cambridge University Press (Chapter 19, pp.
487513).
Tercieux, O., & Vannetelbosch, V. (2006). A characterization of stochastically stable
networks. International Journal of Game Theory, 34(3), 351369.
Van den Nouweland, A. (2005). Models of network formation in cooperative games.
In G. Demange & M. Wooders (Eds.), Group formation in economics; networks,
clubs, and coalitions (pp. 5888). Cambridge University Press.
Vega-Redondo, F. (2007). Complex social networks. Econometric society monograph
series. Cambridge University Press.
Wasserman, S. (1980). Analyzing social networks as stochastic processes. Journal of
the American Statistical Association, 75(370), 280293.
Wasserman, S., & Faust, K. (1994). Social network analysis. Cambridge Univ. Press.
Watts, D. J. (1999). Small worlds. Princeton, NJ: Princeton Univ. Press.
Watts, A. (2001). A dynamic model of network formation. Games and Economic
Behavior, 34(2), 331341.
Watts, D. J., & Strogatz, S. H. (1998). Collective dynamics of small-world networks.
Nature, 393(6684), 440442.
Weibull, J. (1995). Evolutionary game theory. Massachusetts: The MIT Press
Cambridge.
Young, H. (1993). The evolution of conventions. Econometrica, 61(1), 5784.
Young, H. (1998). Individual strategy and social structure An evolutionary theory of
institutions. Princeton University Press.

Evolutions of Social Networks

Enviado por

Dados do documento

Descrição original:

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

Evolutions of Social Networks

Enviado por

Direitos autorais:

Formatos disponíveis

Invited Review

Evolution of social networks

Center for Mathematical Economics (IMW), Bielefeld University, Germany

Corresponding author. Tel.: +49 521 106 67256.

-models (Robins, Pattison,

= b(a, b, A). Hence, the equilibrium effort invested by

, which describes the probability with which a

(G) > 0 are called stochastically stable. These are

is indexed by the parameter > 0. This parameter

(t) = x} denes an action prole r(x) =

(x) is symmetric, and called the attach-

is indeed an extension of the random

/; a) all players play

Você também pode gostar