Review of Basic Concepts:

Normal form

14.126 Game Theory

Muhamet Yildiz

Road Map
• Normal-form Games
• Dominance & Rationalizability
• Nash Equilibrium
– Existence and continuity properties
• Bayesian Games
– Normal-form/agent-normal-form

representations

– Bayesian Nash equilibrium—equivalence to

Nash equilibrium, existence and continuity

1
 Normal-form games
• A (normal form) game is a triplet (N, S, u):
– N = {1, . . . , n} is a (finite) set of players.
– S = S1 × … × Sn where Si is the set of pure
strategies of player i.
– u = (u1,…,un) where ui : S → R is player i’s
vNM utility function.
• A normal form game is finite if S and N are
finite.
• The game is common knowledge.

Mixed Strategies, beliefs

• Δ(X) = Probability distributions on X.
• Δ(Si) = Mixed strategies of player i.
• Independent strategy profile:
σ = σ1 × ... × σn ∈ Δ(S1) × ... × Δ(Sn)
• correlated strategy profile:
σ ∈ Δ(S)
• Δ(S-i) = possible conjectures of player i (beliefs
about the other players’ strategies). [σ-i ∈ Δ(S-i)]
– A player may believe that the other players’ strategies
are correlated!
• Expected payoffs:
ui(σ) = Eσ(ui) = Σs∈Sσ(s)ui(s)

2
 Rationality & Dominance
• Player i is rational if he maximizes his expected payoff
given his belief.
• si* is a best reply to a belief σ-i iff
∀si ∈ Si : ui(si*,σ-i) ≥ ui(si,σ-i).
• Bi(σ-i) = best replies to σ-i.
• σi strictly dominates si iff
∀s-i ∈ S-i : ui(σi,s-i) > ui(si,s-i).
• σi weakly dominates si iff
∀s-i ∈ S-i : ui(σi,s-i) ≥ ui(si,s-i) with a strict inequality.

Theorem: In a finite game, si* is never a best reply to a

(possibly correlated) conjecture σ-i iff si* is strictly
dominated (by a possibly mixed strategy).

Proof of Theorem

• Let • SHT: Let C and D be non-empty,

disjoint subsets of Rm with C closed.
– S-i = {s-i
1,…, s-i
m},
Then, ∃r∈Rm\{0} : ∀x∈cl(D) ∀y∈C,
– ui(si,.) = (ui(si,s-i1),…, ui(si,s-im)) r⋅x ≥ r⋅y.
– U = {ui(si,.)|si ∈ Si} • (<=) Define
– Co(U) = convex hull of U D = {x∈Rm|xk > ui(si*,s-ik) ∀k}.
= {ui(σi,.)|σi ∈ Δ(Si)}
• Assume si* is not strictly
• (=>) Assume si*∈Bi(σ-i). dominated.
⇒ ∀si, ui(si*,σ-i) ≥ ui(si,σ-i)
• Co(U) and D are disjoint.
⇒ ∀σi, ui(si*,σ-i) ≥ ui(σi,σ-i)
⇒ No σi strictly dominates si*. • By SHT, ∃r: ∀σi,
ui(si*,σ-i) ≥ ui(σi,σ-i)
where σ-i(s-ik) = rk/(r1+…+ rm)

3
 Iterated strict dominance &

Rationalizability

• S0 = S
• Sik = Bi(Δ(S-ik-1))
• (Correlated) Rationalizable strategies:
∞
Si∞ = I Sik
k =0

• Independent rationalizability: si∈Sik iff si∈

Bi(Πj≠iσj) where σj ∈ Δ(Sjk-1) ∀j.
• σi is rationalizable iff σi ∈ Bi(Δ(S-i∞)).
Theorem (fixed-point definition): S∞ is the largest
set Z1×…× Zn s.t. Zi ⊆ Bi(Δ(Z-i)) for each i. (si is
rationalizable iff si∈Zi for such Z1×…× Zn.)

Foundations of rationalizability
• If the game and rationality are common
knowledge, then each player plays a
rationalizable strategy.
• Each rationalizable strategy profile is the
outcome of a situation in which the game and
rationality are common knowledge.
• In any “adaptive” learning model the ratio of
players who play a non-rationalizable strategy
goes to zero as the system evolves.

4
 Rationalizability in Cournot

Duopoly

q2
Simultaneously,
each firm i∈{1,2}
produces qi units at 1-c
marginal cost c,
and sells it at price
P = max{0,1-q1-q2}.
1− c
2

q1

1− c
2
1-c

Rationalizability in Cournot
duopoly
• If i knows that qj ≤ q, then qi ≥ (1-c-q)/2.
• If i knows that qj ≥ q, then qi ≤ (1-c-q)/2.
• We know that qj ≥ q0 = 0.
• Then, qi ≤ q1 = (1-c-q0)/2 = (1-c)/2 for each i;
• Then, qi ≥ q2 = (1-c-q1)/2 = (1-c)(1-1/2)/2 for each
i;
• …
• Then, qn ≤ qi ≤ qn+1 or qn+1 ≤ qi ≤ qn where
qn+1 = (1-c-qn)/2 = (1-c)(1-1/2+1/4-…+(-1/2)n)/2.
• As n→∞, qn → (1-c)/3.

5
 Nash Equilibrium
• The following are equivalent:
– σ* =(σ1*,…,σn*) is a Nash Equilibrium
– ∀i, σi* ∈ Bi(σ-i*), where Bi contains mixed best replies
– ∀i, ∀si ∈ Si : ui(σi*,σ-i*) ≥ ui(si,σ-i*),
– ∀i, supp(σi*) ⊆ Bi(σ-i*).
• Aumann & Brandenburger: In a 2-person game, if game,
rationality, and conjectures are all mutually known, then
the conjectures constitute a Nash equilibrium.
• For n>2 players, we need common prior assumption and
common knowledge of conjectures.
• Steady states of any adaptive learning process are Nash
equilibria.

Existence and continuity

• For any correspondence F : X → 2Y, where X compact
and Y bounded, F is upper-hemicontinuous iff F has
closed graph:
[xm → x & ym → y & ym ∈ F(xm)] ⇒ y ∈ F(x).
• Berge’s Maximum Theorem (existence and continuity of
individual optimum): Assume f : X×Z→Y is continuous
and X, Y, Z are compact. Let
F(x) = arg maxz∈Zf(x,z).

Then, F is non-empty, compact-valued, and upper­

hemicontinuous.

• Kakutani’s Fixed-point theorem: Let X be a convex,

compact subset of Rm and let F:X→ 2X be a non-empty,
convex-valued correspondence with closed graph. Then,
there exists x∈X such that x∈F(x).

6
 Existence of Nash Equilibrium
Theorem: Let each Si be a convex, compact
subset of a Euclidean space and each ui be
continuous in s and quasi-concave in si. Then,
there exists a Nash equilibrium s∈S.
Corollary: Each finite game has a (possibly mixed)
Nash equilibrium σ*.
Proof of corollary: Each Δ(Si) ⊆ Rm is convex and
compact. Each ui(σ) is continuous, and linear in
σi. Then, the game with strategy spaces Δ(Si)
has a NE σ* ∈ Δ(S1) × … × Δ(Sn).

Proof of Existence Theorem

• Let F : S → 2S be the “best reply”
correspondence:

Fi(s) = Bi(s-i)

• By the Maximum Theorem, F is non-empty

and has closed graph.
• By quasi-concavity, F is convex valued.
• By Kakutani fixed-point theorem, F has a
fixed point: s* ∈ F(s*).
• s* is a Nash equilibrium.

7
 Upper-hemicontinuity of NE
• X, S are compact metric spaces
• ux(s) is continuous in x∈X and s ∈ S.
• NE(x) is the set of Nash equilibria of (N,S,ux).
• PNE(x) is pure Nash equilibria of (N,S,ux).
Theorem: NE and PNE are upper­
hemicontinuous.
Corollary: If S is finite, NE is non-empty, compact-
valued, and upper-hemicontinuous.
Proof:
• Δ(Si) is compact and ux(σ) is continuous in (x,σ).
• Suppose: xm→x, σm∈NE(xm), σm→σ∉NE(x).
• ∃i, si: ux(si,σ-i) > ux(σ).
• u m (si , σ −mi ) > u m (σ m ) for large m.
x x

Bayesian Games
• A Bayesian game is a list (N, A, Θ, T, u, p):
– N = {1, . . . , n} is a (finite) set of players;
– A = A1 × … × An; Ai is the set of actions of i;
– Θ is the set of payoff relevant parameters;
– T = T1 × … × Tn; Ti is the set of types of i;
– u = (u1,…,un); ui : Θ×A → R is i’s vNM utility function;
– p ∈ Δ(Θ ×T) is a common prior.
• A Bayesian game is a list (N, A, Θ, T, u, p) as
above except ui : Θ ×T ×A → R.
• A Bayesian game is a list (N, A, T, u, p) as above
except ui :T ×A → R.
Fact: All three formulations are equivalent (as long
as you know what you are doing).
Fact: We can replace p with p1,…, pn, dropping CPA.

8
 Normal-form representations
• Given a Bayesian game Γ=(N,A,Θ,T,u,p),
• Normal Form: G(Γ)= (N,S,U):
– Si = {functions si: Ti → Ai }
– Ui(s) = Ep[ui(θ,s1(t1),…,sn(tn))].
• Agent-Normal Form: AG(Γ)= (N,S,U):
N = T1 ∪L∪Tn
S ti = Ai for each ti ∈Ti
U ti (s) = E p [ui (θ , s1 (t1 ),K, sn (t n )) | ti ]

Bayesian Nash equilibrium

Definition: σ* =(σ1*,…,σn*) is a Bayesian Nash
Equilibrium iff for each i, ti,
σ i* (ai | ti ) > 0 ⇒ ai ∈ arg max a E p [ui (θ , ai , σ −*i (a−i | t −i )) | ti ]
i

Fact: σ* is a Bayesian Nash equilibrium of Γ iff the

profile σ*(·|ti), ti∈Ti, i∈N is a Nash equilibrium of
AG(Γ).
Fact: If σ* is a Bayesian Nash equilibrium of Γ,
then σ* is a Nash equilibrium of G(Γ). Ιf p(ti)>0
for each ti, the converse is also true.

9
 Existence of BNE
Consider Γ=(N,A,Θ,T,u,p) with finite N and T.
Theorem: If
• each Ai is compact and convex
• each ui is bounded, continuous in a, concave
in ai,
then Γ has a pure Bayesian Nash equilibrium.
Proof: AG(Γ) has a pure Nash equilibrium.
Corollary: If A is finite, Γ has a (possibly
mixed) Bayesian Nash equilibrium.

Upper-hemicontinuity of BNE
• A, T finite and Θ, X compact.
• uix(θ,a) continuous in (x,θ,a)
• BNE(x) Bayesian NE of Γx = (N,A,Θ,T,ux,p).
• BNE(p) Bayesian Nash equilibria of (N,A,Θ,T,u,p).

Theorem: BNE is upper-hemicontinuous.

Proof: BNE(x) = NE(AG(Γx)).

Theorem: Assume p(ti) > 0 ∀p∈P, ∀ti∈Ti, for compact

P ⊆ Δ(Θ×T). BNE(p) is upper-hemicontinuous on P.

Proof: Ui(s;p) = Ep[ui(θ,s1(t1),…,sn(tn))] is continuous;
BNE(p) = NE(G((N,A,Θ,T,u,p))).

10
MIT OpenCourseWare
http://ocw.mit.edu

14.126 Game Theory

Spring 2010

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.