of Networks Strategic Formation Lecture 10
of NetworksStrategic Formation
Lecture 10
We knowPeople make choices thinking about what others are doing
We knowPeople make choices thinking about what others are doing
our social relationships have an important effect on
how we behave
We knowPeople make choices thinking about what others are doing
our social relationships have an important effect on
how we behave
Our networks are valuable, they are the capital we use
to achieve our goals
Network FormationIn many social settings, our relationships are not given
Network FormationIn many social settings, our relationships are not given
We decide with whom to link
Network FormationIn many social settings, our relationships are not given
We decide with whom to link
In this class we will study how networks are formed by
strategic decision-makers
Aim Lecture 10To Understand
The basic of network strategic formation
Simple cases of how to model network formation
The relation between individual incentives & social welfare from a network perspective (emergence)
+
+
+
Some basic intuition
Some basic intuition
Costs & Benefits
Actors choose links
Contrasts incentives & Social efficiency
Some basic intuition
Costs & Benefits
Actors choose links
Contrasts incentives & Social efficiency
Nodes as actors making choices
Some basic intuition
Costs & Benefits
There are costs and benefits associated with each network
Actors choose links
Contrasts incentives & Social efficiency
Nodes as actors making choices
Some basic intuition
Costs & Benefits
There are costs and benefits associated with each network
Actors choose links
Contrasts incentives & Social efficiency
Nodes as actors making choices
We can also call them agents, players, people, countries, etc...
Some basic intuition
Costs & Benefits
There are costs and benefits associated with each network
Actors choose links
Contrasts incentives & Social efficiency
Nodes as actors making choices
We can also call them agents, players, people, countries, etc...
We will contrast the incentives individuals have to form relationships with what’s best for society
How to model choicesIncentives to form and severe links
How to model choicesIncentives to form and severe links
Is consensus needed? (undirected/directed)
What happens on the network?
Are links adjustable in intensity?
Can they coordinate changes in the network?
How to model choicesIncentives to form and severe links
Is consensus needed? (undirected/directed)Does somebody has to say yes to me to relate with her?
What happens on the network?
Are links adjustable in intensity?
Can they coordinate changes in the network?
How to model choicesIncentives to form and severe links
Is consensus needed? (undirected/directed)Does somebody has to say yes to me to relate with her?
What happens on the network?
Are links adjustable in intensity?
Can they coordinate changes in the network?
Citations networks vs. friendships or alliances
How to model choicesIncentives to form and severe links
Is consensus needed? (undirected/directed)Does somebody has to say yes to me to relate with her?
What happens on the network?
Are links adjustable in intensity?
Can they coordinate changes in the network?
Citations networks vs. friendships or alliances
I’ll form a new alliance with you but only if we can form an alliance with someone else at the same time
How to model choicesIncentives to form and severe links
Is consensus needed? (undirected/directed)Does somebody has to say yes to me to relate with her?
What happens on the network?
Are links adjustable in intensity?
Can they coordinate changes in the network?
Citations networks vs. friendships or alliances
I’ll from a new alliance with you but only if we can form an alliance with someone else at the same time
What’s generating value in the network? where? Costs & benefits?
How to model choicesIncentives to form and severe links
Is consensus needed? (undirected/directed)Does somebody has to say yes to me to relate with her?
What happens on the network?
Are links adjustable in intensity?
Can they coordinate changes in the network?
Citations networks vs. friendships or alliances
I’ll from a new alliance with you but only if we can form an alliance with someone else at the same time
What’s generating value in the network? where? Costs & benefits?
Only whether a relationship exists or not? How intense is it?
Questions that ariseWe will address some of them
Which networks are likely to form?
Are some networks more stable than others?
We will address some of them
Which networks are likely to form?Are they the right ones from society’s perspective?
Are some networks more stable than others?
Questions that arise
We will address some of them
Which networks are likely to form?Are they the right ones from society’s perspective?
Are some networks more stable than others?
if not, how different are they?
Questions that arise
We will address some of them
Which networks are likely to form?Are they the right ones from society’s perspective?
Are some networks more stable than others?
if not, how different are they?
How likely is it that changes can occur in certain networks?New people coming in? Creating new relationships?
Questions that arise
Representationof strategic network formation
(Jackson & Wollinsky, 1996)
Representationof strategic network formation
(Jackson & Wollinsky, 1996)
Payoffs:
The network g generates some payoffs to a given individual i
Representationof strategic network formation
Payoff to i if the network is gui(g)
(Jackson & Wollinsky, 1996)
Payoffs:
The network g generates some payoffs to a given individual i
Representationof strategic network formation
Payoff to i if the network is gui(g)
(Jackson & Wollinsky, 1996)
Payoffs:
The network g generates some payoffs to a given individual i
For simplicity:
We are going to look at undirected network formation
Representationof strategic network formation
Payoff to i if the network is gui(g)
(Jackson & Wollinsky, 1996)
Payoffs:
The network g generates some payoffs to a given individual i
For simplicity:
We are going to look at undirected network formation
We could look at directed, weighted, etc.
ModelingIncentives & Equilibrium
ModelingIncentives & Equilibrium
is modeled as a game where
The network:
ModelingIncentives & Equilibrium
is modeled as a game where
The network:
Each agent announces who they wish to link with
ModelingIncentives & Equilibrium
is modeled as a game where
The network:
Each agent announces who they wish to link with
& a link forms if and only if both agents name each other
ModelingIncentives & Equilibrium
is modeled as a game where
The network:
What is a Nash equilibrium in this game?
Each agent announces who they wish to link with
& a link forms if and only if both agents name each other
Equilibrium NetworksNash equilibrium
Equilibrium NetworksNash equilibrium
A situation where nobody can gain by changing his/her actions (announcements)
Equilibrium NetworksNash equilibrium
Given the actions (announcements) of other individuals
A situation where nobody can gain by changing his/her actions (announcements)
for undirected networksProblem of Nash
mutual agreement to form a link
Problem of Nashfor undirected networks
mutual agreement to form a link
If they are separated, each gets a value of 0 and if together 1.Example (i & j):
for undirected networks
Example (i & j):If they are separated, each gets a value of 0 and if together 1.
They play a game of simultaneous announcement
mutual agreement to form a link
Problem of Nash
for undirected networks
If they are separated, each gets a value of 0 and if together 1.
They play a game of simultaneous announcement
If i thinks j won’t announce, she might as well not announce
Example (i & j):
mutual agreement to form a link
Problem of Nash
for undirected networks
i
j
0
0
Case 1: Neither announces is a Nash equilibrium
If they are separated, each gets a value of 0 and if together 1.
They play a game of simultaneous announcement
Example (i & j):
If i thinks j won’t announce, she might as well not announce
mutual agreement to form a link
Problem of Nash
for undirected networks
ii
jj
1
1
0
0
Case 1: Neither announces is a Nash equilibrium
If they are separated, each gets a value of 0 and if together 1.
They play a game of simultaneous announcement
Case 2: Both announce is a Nash equilibrium
Example (i & j):
If i thinks j won’t announce, she might as well not announce
mutual agreement to form a link
Problem of Nash
Pairwise StabilityModeling Incentives
A network (N,g) is pairwise stable if
ii
jj
1
1
0
0
Pairwise StabilityModeling Incentives
A network (N,g) is pairwise stable if
ii
jj
1
1
0
0
No actor gains from severing a link (one person can do it)
No two actor gain from forming a link (at least one is better off)
Pairwise StabilityModeling Incentives
A network (N,g) is pairwise stable if
ii
jj
1
1
0
0
Relationship must be beneficial to be maintained
No actor gains from severing a link (one person can do it)
No two actor gain from forming a link (at least one is better off)
Pairwise StabilityModeling Incentives
A network (N,g) is pairwise stable if
ii
jj
1
1
0
0
Relationship must be beneficial to be maintained
Beneficial relationship are pursued when available
No actor gains from severing a link (one person can do it)
No two actor gain from forming a link (at least one is better off)
Some Notation
A network (N,g) is pairwise stable if
Pairwise Stability
Some Notation
A network (N,g) is pairwise stable if
No agent gains from severing a link (one person can do it)
ui(g) ≥ ui(g-ij) for i and ij in g
Pairwise Stability
Some Notation
A network (N,g) is pairwise stable if
No agent gains from severing a link (one person can do it)
ui(g) ≥ ui(g-ij) for i and ij in g
No two agents gain from forming a link (at least one is better off)
ui(g+ij) ≥ ui(g) implies uj(g+ij) < uj(g) for ij not in g
Pairwise Stability
Some Notation
A network (N,g) is pairwise stable if
No agent gains from severing a link (one person can do it)
ui(g) ≥ ui(g-ij) for i and ij in g
ui(g+ij) ≥ ui(g) implies uj(g+ij) < uj(g) for ij not in g
It’s still a very weak concept
Pairwise Stability
No two agents gain from forming a link (at least one is better off)
Some Notation
A network (N,g) is pairwise stable if
No agent gains from severing a link (one person can do it)
ui(g) ≥ ui(g-ij) for i and ij in g
ui(g+ij) ≥ ui(g) implies uj(g+ij) < uj(g) for ij not in g
It’s still a very weak concept
Only looks at pairs of individuals
Pairwise Stability
No two agents gain from forming a link (at least one is better off)
Some Notation
A network (N,g) is pairwise stable if
No agent gains from severing a link (one person can do it)
ui(g) ≥ ui(g-ij) for i and ij in g
ui(g+ij) ≥ ui(g) implies uj(g+ij) < uj(g) for ij not in g
It’s still a very weak concept
Only looks at pairs of individualsOne link at a time
Pairwise Stability
No two agents gain from forming a link (at least one is better off)
Some NotationPairwise Stability
A network (N,g) is pairwise stable if
No agent gains from severing a link (one person can do it)
ui(g) ≥ ui(g-ij) for i and ij in g
ui(g+ij) ≥ ui(g) implies uj(g+ij) < uj(g) for ij not in g
It’s still a very weak concept
Only looks at pairs of individualsOne link at a time
(It’s a minimum)
No two agents gain from forming a link (at least one is better off)
Back to our example
ii
jj
1
1
0
0
Back to our example
ii
jj
1
1
0
0
Relationship must be beneficial to be maintained
Beneficial relationship are pursued when available
Back to our example
ii
jj
1
1
0
0
Relationship must be beneficial to be maintained
Beneficial relationship are pursued when available
Which is the pairwise stable situation (network)?
Back to our example
ii
jj
1
1
0
0
Relationship must be beneficial to be maintained
Beneficial relationship are pursued when available
Which is the pairwise stable situation (network)?
Back to our example
ii
jj
1
1
0
0
Relationship must be beneficial to be maintained
Beneficial relationship are pursued when available
Which is the pairwise stable situation (network)?
Although both are Nash equilibria
RememberLave & March
RememberLave & March
ModelsAre a simplified version of the world
RememberLave & March
ModelsAre a simplified version of the world
Models are created by speculating about the process that could have produced
the observed outcomes
Collaboration NetworksPairwise Stability
0 0
0 0
Think of situations in which we collaborate with others (i.e.,
friendships, projects, inter-firm R&D)
The more collaborations you have the better for you
&
ButThe more collaborations your
collaborators have the worse for you
Negative Externalities
Collaboration NetworksPairwise Stability
0 0
0 0
0 3
0 3
i f 2 actors are related, each gets 3 out of it.
Collaboration NetworksPairwise Stability
0 0
0 0
0 3
0 3
Will this situation tend to change or is it likely to be stable?
i f 2 actors are related, each gets 3 out of it.
Collaboration NetworksPairwise Stability
0 0
0 0
0 3
0 3
3 3
3 3
The two other actors can be better if they relate
Collaboration NetworksPairwise Stability
0 0
0 0
0 3
0 3
3 3
3 3
3.25 3.25
2 2
NEGATIVE EXTERNALITIES
But now, the new connection benefits directly the linked actors & affects negatively & indirectly the existing neighbors
Collaboration NetworksPairwise Stability
0 0
0 0
0 3
0 3
3 3
3 3
3.25 3.25
2 2
Think about the attention you get from your parents if you are alone or if you have to
share it with your brothers & sisters
NEGATIVE EXTERNALITIES
Collaboration NetworksPairwise Stability
0 0
0 0
0 3
0 3
3 3
3 3
3.25 3.25
2 2
2.5 2.5
2.5 2.5
Everyone has incentives to create a new connection
without intending to harm their neighbors
Think about Schelling’s Segregation
Collaboration NetworksPairwise Stability
0 0
0 0
0 3
0 3
3 3
3 3
3.25 3.25
2 2
2.5 2.5
2.5 2.5
2.78 2
2 2.78
Collaboration NetworksPairwise Stability
0 0
0 0
0 3
0 3
3 3
3 3
3.25 3.25
2 2
2.5 2.5
2.5 2.5
2.78 2
2 2.78
2.33 2.33
2.33 2.33
Until the networks is
complete
Collaboration NetworksPairwise Stability
0 0
0 0
0 3
0 3
3 3
3 3
3.25 3.25
2 2
2.5 2.5
2.5 2.5
2.78 2
2 2.78
2.33 2.33
2.33 2.33
PairwiseStable
Stability & EfficiencyThink about this two networks
Stability & EfficiencyThink about this two networks
3 3
3 3
Each could have gotten 3 & society 12
Stability & EfficiencyThink about this two networks
3 3
3 3
Each got 2.33 & society 9.32
2.33 2.33
2.33 2.33
PairwiseStable
Stability & EfficiencyThink about this two networks
3 3
3 3
Each got 2.33 & society 9.32
2.33 2.33
2.33 2.33
PairwiseStableEfficient
Not stable to individual incentives
EfficiencyA network is efficient if the total
sum of all the utilities from all the players is the maximal one can
achieve
The situation in which society as a whole is better offBut still some can be in very bad shape
Notation:A network g is efficient if:
g maximizes ∑ui(g’)
Emergence
3 3
3 3
Although individuals do not intend to harm their neighbors
2.33 2.33
2.33 2.33
Efficiency
Pursuing personal incentives leads to social inefficiency
Their actions have implications for other people they are not
necessarily taking into account
From our friends’ friends
Positive Externalities
Other Effects
In different settings
I get indirect benefits the more friends my friends have
Think about popularity, transmission of valuable information, access to jobs
We will cover some of this in other lectures
Some Notation
Positive Externalities
uk(g+ij)≥uk(g) if ij not in g, for every k≠i,j
Everyone is equal or better with the new link
Negative Externalities
uk(g+ij)≤uk(g) if ij not in g, for every k≠i,j
At least someone is worse with the new link
Checklist
Strategic network formation assumes actors are rational decision makers
Externalities from relationships influence the networks that are likely to be formed
Social efficiency can be affected by individual incentives
+
+
+
Questions?