Static Games of Incomplete

Information

So far, in the static games of complete information we assumed common knowledge

which helped us to use solution concepts like IESDS, rationalizability and most impor-
tantly Nash Equilibrium.

We can describe many situation where players have some idea about their opponent’s
characteristics but dont know for sure what their characteristics are. Like, in a Counot
Oligipoly Model, my cost may not be known to my opponent firms in the market.

In a simultaneous game in which the players do not know what actions his opponents
are taking, but instead knows what the set of actions can be. A player forms conjecture
about the behavior of his opponents in order to choose his best response, this idea was
called player’s belief over the actions that his opponents will choose. The beliefs are
assumed to be correct. John Harsayani extended this idea of belief on the set of actions
o the opponent to beliefs over the characteristics of the opponent. The idea of belief
over the characteristics of other players is captured by defining types of the players.

Games where a player can have different types, where the types are privately known are
called games with incomplete information.

An incomplete information game is a game where a player knows something that the
other party does not know. The payoff function of at least one player is not known.

Some motivating examples:

1
2

Sealed Bid Auctions: here the players know their on valuation for the good being sold
but does not know any other bidder’s valuation. Bids are submitted in sealed envelopes
which makes this a simultaneous move game.

Another example, consider a counot duopoly where Firm 1 marginal cost of production
is c and Firm 2’s cost of production is ch with probability θ and cl with probability 1−θ.
Firm 1 does not know the actual cost of Firm 2.

Normal Representation of a Static Bayesian Game

Let us look at what a static information game consists of and what is known/unknown
to the players:

• Players: there is a set of players N ≡ 1, 2, ..., n with n ≥ 2.

• A set of states Ω, and a common prior µ on Ω. (this is common knowledge).

• Actions sets: A1 , A2 , ..., An

• A strategy of a player is now a prescription that tells each type of a player what
he should do if this is the type Nature chose for the game.

• Type Spaces/Sets of types: T1 , T2 , ..., Tn (Nature chooses the type of each player
from his type space). We assume that the type space is finite.

• Type set : T = T1 × T2 × T3 × ... × Tn

A profile of types for n players is given by any element t ∈ T where t =
(t1 , t2 , t3 , ..., tn ) and ti ∈ Ti ∀i ∈ N .

• Nature chooses t according to a probability function p(.) given by p : T → [0, 1].

• The private signal (about own type) is revealed to each player and then each player
p(t−i ,ti ) p(t−i ,ti )
has a conditional distribution pi (t−i |ti ) = p(ti )
= Σt−i ∈T−i p(t−i ,ti )

• Payoff functions : u1 , u2 , ..., un , where the utility function for each player is given
by ui : A×T → R. Note that the payoff is now type-dependent. The utility is not
 3

dependent on t−i , that is, one’s payoffs doe not depend on the types of the other
players. This is a setup of private values, since each players’s payoff depends
only on his private information.

An incomplete information game (Bayesian Game) has the following components: G ≡

⟨N, {Ai }i∈N , {Ti }i∈N , {ui }i∈N , {pi }i∈N ⟩

A Bayesian Game proceeds in the following steps:

1. Nature choose a profile of players types t = (t1 , t2 , t3 , ..., tn ) ∈ T according to a

prior probability distribution p(.).

2. Each player i learns hi own type ti , which is a private information and then used
his prior pi to form posterior beliefs over the other types of the players.

3. Players simultaneously (static games) choose actions ai ∈ Ai , i ∈ N .

4. Given the players’s actions a = (a1 , a2 , a3 , ..., an ) the payoff ui (a, ti ) are realized
for each player i ∈ N .

Common Prior and Consistent belief

The beliefs of players in a game with incomplete information are said to be consis-
tent if they are derived from the same probability distribution (the common prior) by
conditioning on each player’s private information.

In other words, if the beliefs are consistent, the only source of differences in beliefs is
difference in information.

Bayesian Equilibrium: A Nash equilibrium of a Bayesian game is a list of behavior

and beliefs such that each player is doing his best to maximize his payoff, according to
his beliefs about the behavior of the other players.

Strategy : In the static Bayesian game G, a strategy for player i is a function si (ti ),
where for each type ti ∈ Ti , si (ti ) specifies the action from the feasible set Ai , that ti
would choose if drawn by nature. si : Ti → Ai . Si is the (pure) strategy space of player
4

i. A mixed strategy is a probability distribution over a player’s pure strategy. If all types
choose same action then such a strategy is called pooling strategy. If each type chooses
a different action, its called separating strategy.

This can be thought of as each player chooses his type-contingent strategy before he
learned his type and then played according to that strategy.

Bayesian Nash Equilibrium: In a static game G strategies s∗ = (s∗1 , ..., s∗n ) are a pure
strategy BNE if for each player i and for each of i’s type ti ∈ Ti , s∗i (ti ), s∗i (ti ) solves:

maxai ∈Ai Σt−i ∈T−i [ui ((s∗1 (t1 ), ..., s∗i−1 (ti−1 ), ai , s∗i+1 (ti+1 ), s∗n (tn )))]pi (t−i |ti )

That is no player wants to change her strategy even if the change involves only one
action by one type.

All the players agree on the way in which players types differ from each other and in
the way in which the nature chooses among these profiles of types.

Joint distribution on the types is common knowledge. Nature chooses t according to a

probability function p(.). p : T → [0, 1]. p is common knowledge.

Conditional probabilities differ based on private information.

An example with two players and two types with common prior:

G B
1 1
G 4 4
1 1
B 3 6

Figure 0.1: Joint Distribution: Two Types & Two Players

P r(t1 =t2 =G) p(G,G) 1/4 1

Now, p1 (G|G) = P r(t1 =G)
= p(G,G)+p(G,B)
= 1/4+1/4
= 2

1 2 1
Similarly, p1 (B|G) = 2
, p1 (G|B) = 3
and p1 (B|B) = 3
. We can find conditional
probabilities for player 2 in similar manner.
 5

Examples

Cournot Competition with Asymmetric Information

This application is adapted from Gibbons Section 3.1.A. However, there are some issues
with the notations in the text book which may lead to serious confusion while reading.
Kindly see a re-written version of the game below.

• Players: 1, 2

• Types: T1 = {c} and T2 = {cH , cL }.

• Common Prior: cost is cH with probability θ and cL with 1 − θ and cL < cH .

• Actions: A1 = A2 = [0, ∞), that is, q1 , q2 .

• Strategy: For Firm 1, its simply q1 ∈ A1 and for Firm 2 is a pair (q2 (cH ), q2 (cL ))
which indicates choice of Firm 2’s quantity for each its type, where (q2 (cH ), q2 (cL ) ∈
A2 .

• Payoffs: u1 (q1 , q2 , c) = [(a−q1 −q2 )−c]q1 , u2 (q1 , q2 , cH ) = [(a−q1 −q2 )−cH ]q2
and u2 (q1 , q2 , cL ) = [(a − q1 − q2 ) − cL ]q2 .

Let q2 (cH ) and q2 (cL ) denote the quantity chosen by Firm 2 if its cost is high and low
respectively. Let q1 denote th quantity choice of firm 1.

If Firm 2’s cost is high, it will choose q2 (cH ), given q1 and solve:

maxq2 (cH ) [(a − q1 − q2 (cH )) − cH ]q2 (cH )

Similarly, Firm 2’s choice when its cost is low would be:

maxq2 (cL ) [(a − q1 − q2 (cL )) − cL ]q2 (cL )

Firm 1 knows the distribution on the cost of Firm 2. Therefore, Firm 1 maximizes the
expected payoff:

maxq1 θ([(a − q1 − q2 (cH )) − c]q1 ) + (1 − θ)[(a − q1 − q2 (cL )) − c]q1

6

The first order conditions gives us the reaction functions/best response functions for
both the firms:
a−q1 −cH
q2 (cH ) = B2 (q1 ; cH ) = 2

a−q1 −cL
q2 (cL ) = B2 (q1 ; cL ) = 2

θ[a−q2 (cH )−c]+(1−θ)[a−q2 (cL )−c]

q1 = B1 (q2 ) = 2

Note that q1 depends on q2 (cH ) and q2 (cL ). These are the actions of Firm 2 and NOT
the type. The action of Firm 1 q1 does not depend on the type of the other Firm. This is
a private value game.

Assuming that the solution exists, solving the system of equations, we get the Bayesian-
Nash Equilibrim of the game:

a−2cH +c
q2∗ (cH ) = 3
+ 1−θ
6
(cH − cL )

a−2cL +c
q2∗ (cL ) = 3
− 6θ (cH − cL )
a−2c+θcH +(1−θ)cL
q1∗ = 3

Please read the interpretation and comparisons on page 146 of Gibbons (paragraph
above Section 3.1.B)

Modified Prisoner’s Dilemma

Consider the game of Prisoner’s Dilemma, but now there is uncertainty about the type
of Player 1. Player 1 can either be altruistic or he can be rational. However, Player 2 is
rational. We may think that Nature selects one of the two following games:

M F M F

M -2, -2 -5, -1 M -2,-2 -5,-1

F -3,-5 -7, -4.5 F -1,-5 -4,-4

Game 1: Player 1 is altruistic Game 2: Player 1 is rational

Figure 0.2: Modified Prisoner’s Dilemma

 7

• Players: 1, 2

• Types: T1 = {a, r} and T2 = {r}

• Common Prior: Player 1 is altruistic with probability θ and rational with 1 − θ

where θ ∈ [0, 1]

• Actions: A1 = A2 = {M, F }

After the Nature chooses one of the game from Game 1 and Game 2, Player 1 knows
his type, which Player 2 only has a belief about the type of Player 1.

When Player 1 mover, he knows his own type, but he does not know the action chosen
by Player 2. Remember that this is a simultaneous move game.

When Player 2 mover, he neither knows which of the types of Player 1 he is facing (he
only knows the distribution on T1 ), nor he knows the choice of Player 1.

It is as if Player 2 faces two possible players and nature decided against which player
he is playing the game.

Let us see what are the strategies of Player 1 in this game. Player 1 must have four pure
strategies: each of his type has two actions to choose from. Let us define strategy of
Player 1 as xy ∈ S1 where S1 = {M M, M F, F M, F F }. x describes the action of the
altruistic type of Player 1 and y describes the action of the rational type of Player 1.
(Go back and see the definition of Strategy).

Once we have the pure strategies for each player, each pair of pure strategies will give
rise to a path of play that starts with Nature’s choice and then follows with the simulta-
neous actions of both the players.

In this example, say, if Player 1 plays M if he is altruistic and F is he is rational, that

is, M F and if Player 2 plays M , then with probability θ the outcome yield payoffs of
(−2, −2), and with probability 1 − θ the payoffs will be (−1, −5). Thus, the pair of
payoffs from the pair of strategies (M F, M ) is:

u1 = θ ∗ (−2) + (1 − θ) ∗ (−1) and u2 = θ ∗ (−2) + (1 − θ) ∗ (−5).

8

We can find expected payoff for both the players for all strategies. These are given in
figure ?? below

M F

MM -2, -2 -5, -1

MF −1 − θ, 3θ − 5 −4 − θ, 3θ − 4

FM −θ − 2, −2 − 3θ −5 − 2θ, −1 − 3.5θ

FF −1 − 2θ, −5 −4 − 3θ, −4 − 0.5θ

Figure 0.3: Modified Prisoner’s Dilemma: Expected Payoffs

We can now use the tools we already know and find the equilibrium. We can find Nash
Equilibrium of this game. Notice that for any value of θ ∈ [0, 1] (M F, F ) is strictly
dominant strategy equilibrium.

We cannot perform equilibrium analysis unless we assume that each player knows the
distribution of his opponent’s types, and once he assumes some behavior of these dif-
ferent types, then he can calculate his own expected utility from his different actions.
In this way Harsyani changed the complex and challenging concept of incomplete in-
formation, into a well known game of imperfect information, in which Nature chooses
the players’ types, and we then use our standard tools of analysis (add figure).

Let us expand this example further with two players and two types, Ti = {tai , tri },
i = 1, 2 and like before each player has two pure actions, Ai = {M, F }. Now consider
the case where Player 1 believes that Player 2 is using the following strategy:


M if t2 = ta2


s2 (t2 ) =  (1)
F if t2 = tr2


If Player 1 is of type ta1 , then from his own perspective, his expected utility from playing
M will be:
 9

u1 (M, s2 (t2 ); ta1 ) = p1 (ta2 |ta1 ).u1 (M, s2 (ta2 ); ta1 ) + p1 (tr2 |ta1 ).u1 (M, s2 (tr2 ); ta1 )
= p1 (ta2 |ta1 ).u1 (M, M ; ta1 ) + p1 (tr2 |ta1 ).u1 (M, F ; ta1 )

After we take a static game of incomplete information and transform it into a Bayesian
game, then the “Bayesian" part is infact that for each player’s type realization, we com-
pute his beliefs about the actions of his opponents using the Baye’s Rule. Payoffs from
the strategy profile are the expected utility that is derived from the strategies played by
other players, and the mixing that occurs is due to the randomization of nature that each
player faces through his belief pi (t−i |ti ).

Entry Game

Fudenberg & Tirole (Ch 6, Page 209 - 216)

Mixed Strategies Revisited

Gibbons Section 3.2.A