A Solution Concept for Network Formation
Games
June 10, 2022
A Solution Concept for Network Formation Games
Introduction
Network formation games typically present a multiplicity of
Nash equilibria
Some of them are such that players fail to coordinate on
forming mutually beneficial links
Pairwise stability (Jackson and Wolinsky, 1996) capture
this coordination requirement, but does not rule out
dominated equilibria
This work proposes an equilibrium refinement for this class
of games which naturally involves pairwise stability while
guaranteeing admissibility
A Solution Concept for Network Formation Games
Environment
Environment
A set of agents
Links connecting agents
(Net) Payoffs
A Solution Concept for Network Formation Games
Example
A Solution Concept for Network Formation Games
A Network Formation Game
How do people decide to build connections?
Network Formation Games study the individual strategic
incentives agents face when choosing who they wished to
be linked with
Simultaneous Link-Announcement Game (Myerson,
1991)
1 Undirected network: a link is formed if and only if both
parties wish to be linked
2 Static game : all links are announced simultaneously
A Solution Concept for Network Formation Games
Applications
Models of network formation using the link-announcement
game studied here have been used in many different
settings
Friendship ( Jackson and Wolinsky, 1996)
Coauthoring (Jackson and Wolinsky, 1996)
Free-Trade Networks (Furusawa and Konishi, 2002)
Market-Sharing Agreements (Belleflamme and Bloch, 2002)
Labor Markets ( Calvo-Armengol and Jackson, 2001)
Financial Networks Models (Acemoglu, 2015)
A Solution Concept for Network Formation Games
Equilibrium Analysis
Is Nash Equilibrium Enough?
Nash Equilibrium: no player can improve her equilibrium
payoff by changing the set of links she announces. This
rules out profitable unilateral deviations
The double-consent requirement generates a multiplicity of
Nash equilibria; in some of them, couples of players do not
announce a link even if it would make them better off
For example, the empty network is always a NE even if
player would be better off together.
A Solution Concept for Network Formation Games
Equilibrium Analysis
Example
The empty network is a (bad) Nash Equilibrium!
A Solution Concept for Network Formation Games
Pairwise-Nash Stability
Jackson and Wolinsky (1996) introduce the notion of
Pairwise-Nash Equilibrium
A network is Pairwise-Nash when the following two
conditions hold together:
1 The network is the outcome of a Nash-Equilibrium play: no
player wants to cut one of her links
2 Couple of players always coordinate on mutually beneficiaI
links: if a link is not formed and one of the two players
benefitted from it, it means that the link was detrimental for
the other player
A Solution Concept for Network Formation Games
Pairwise Nash Stability
Example
The empty network is not Pairwise Stable: couple of
players have mutually profitable deviations
A Solution Concept for Network Formation Games
Drawbacks of Pairwise Stability
The point of Pairwise Nash stability is to provide a notion of
equilibrium that ensures all bad links are excluded and all
good links are included
Unfortunately, often Pairwise Nash networks can be the
outcome of suboptimal plays
In particular, we may obtain unreasonable outcomes
supported by weakly dominated Nash Equilibria
A Solution Concept for Network Formation Games
Drawbacks of Pairwise Stability
Example
The empty network is Pairwise Stable, but it is a dominated
equilibrium
A Solution Concept for Network Formation Games
B-Perfect Equilibrium
To guarantee the selection of equilibria in undominated
strategies, we propose the solution concept of B-Perfect
Equilibrium (De Sinopoli and Meroni, 2018)
We introduce a different notion of mixed strategy. Normally,
a mixed strategy is a randomization over pure strategies.
Here we consider a vector of randomized choices for each
single issue
Since announcing a link or not is a dichotomous choice, it
is more natural to think about the stochastic version of a
strategy in a different way
We call b-strategy a n − 1 dimensional vector with entries
bij ∈ [0, 1], where each entry represents the probability that
a player will announce a link with another player.
A Solution Concept for Network Formation Games
B-Perfect Equilibrium
Definition
We now define a refinement of perfect equilibrium, which
requires the profiles in the converging sequence to be
completely-mixed b-strategy profiles (i.e. b-strategies with
entries strictly between 0 and 1).
Definition (b-perfect equilibrium). Let b(s) be the
b-strategy profile corresponding to the pure strategy profile
s. The pure strategy profile s is a b-perfect equilibrium of
the network formation game if there exists a sequence of
completely-mixed b-strategy profiles converging to b(s)
such that s is best reply to every element of the sequence.
A Solution Concept for Network Formation Games
B-perfect Equilibrium
Properties
B-perfect equilibrium is a refinement of trembling-hand
perfect equilibrium.
To see this, notice that every completely mixed b-strategy
profile has a corresponding completely mixed strategy
profile (the converse is not true). Therefore, every b-perfect
equilibrium is trembling-hand perfect.
Since every trembling-hand perfect equilibrium is
undominated, we are ensured that in a b-perfect
equilibrium players will never play dominated strategies.
A Solution Concept for Network Formation Games
B-Perfect Equilibrium
Example
The empty network is Pairwise Stable and dominated, but
it is not b-perfect
A Solution Concept for Network Formation Games
B-Perfect Equilibrium and Pairwise Stability
In general, the sets of pairwise-stable and b-perfect
networks may not coincide.
However, a mild non-indifference condition on the payoff
function ensures that a b-perfect network is pairwise stable.
Whenever players are not indifferent after a new link with
another player is added (local link-responsiveness), then
b-perfection implies pairwise-stability.
Thus, checking this non-indifference condition is enough to
ensure that a b-perfect equilibrium captures the
coordination requirement of pairwise stability, while at the
same time ruling out dominated strategies.
A Solution Concept for Network Formation Games
