Game Theory: lecture of April 2, 2019
Chiara Mocenni
Course on Game Theory
Symmetric Games with two players (1/2)
A game is symmetric when the two players obtain the same payoff under the same conditions.
A symmetric game involves two players with the same number of strategies and the payoff function of any strategy is
independent on the location of the player that is playing at a given time.
Assuming the equivalence of the payoff functions of the pure strategies is equivalent to assume that the payoff bi-matrix of the second player is the transpose of the payoff matrix of the first player, that is: B = AT.
Symmetric Games with two players (2/2)
With K = {1, 2, ..., k} we indicate the set of pure strategies With x ∈ ∆ e y ∈ ∆ we indicate the mixed strategies of the first and the second player, where
∆ = {x ∈ R+k :P
i ∈Kxi = 1}, while Θ = ∆2.
The set of best reply β(z) is calculated with respect to one strategy z ∈ ∆ and it is the same for both players:
β(z) = {x ∈ ∆ : π(x , z) ≥ π(y , z) ∀y ∈ ∆}.
Symmetric Nash Equilibria
A couple of strategies (x , y ) ∈ Θ = ∆2 (strategy profile) is a Nash equilibrium, (x , y ) ∈ ΘNE, is and only if x ∈ β(y ) and y ∈ β(x ). If x = y , the equilibrium (x , y ) is said symmetric.
The subset of strategies x ∈ ∆ that are in equilibrium with themselves is:
∆NE = {x ∈ ∆ : (x , x ) ∈ ΘNE}
The Nash equilibria of a symmetric game can be asymmetric, but each symmetric game has at least one symmetric Nash equilibrium.
Theorem
For each finite symmetric game with two players,, ∆NE 6= ∅.
Classification of 2x2 symmetric games
In this section we analyze games where players have only two pure strategies available. Let be the generic payoff matrix of a 2 × 2 symmetric game:
A =
a11 a12 a21 a22
Subtracting a21 from the first column and a12 from the second, we have:
A0 =
a11− a21 0 0 a22− a12
In this way we obtain a symmetric matrix and then a symmetric game, with payoff matrix:
A0=
a1 0 0 a2
Where a1 = a11− a21 and a2 = a22− a12.
Graphical representation
The obtained matrix is identified by one point a = (a1, a2) ∈ R2. This point will belong to one of the orthants of this cartesian plane, allowing us to identify the category to which the game corresponds.
Calculus of symmetric Nash equilibria
If x ∈ ∆NE (x is a symmetric Nash equilibrium), then x ∈ β(x ). In other words:
π(x , x ) ≥ π(y , x ) ∀y ∈ ∆, or
xTAx ≥ yTAx ∀y ∈ ∆.
We can rewrite the equation in terms of sums:
k
X
i =1
xi[Ax ]i ≥
k
X
i =1
yi[Ax ]i ∀y ∈ ∆, (1) where [Ax ]i indicates l’i -th component of vector Ax .
Theorem for pure equilibria
Theorem
Let x be a pure strategy (x = eh). If [Ax ]h≥ [Ax]i ∀i 6= h, then x ∈ ∆NE.
Proof.
Let M = [Ax ]h. We have that:
k
X
i =1
xi[Ax ]i = xh[Ax ]h= M, and
k
X
i =1
yi[Ax ]i = yh[Ax ]h+
k
X
i =1,i 6=h
yi[Ax ]i = yhM +
k
X
i =1,i 6=h
yi[Ax ]i.
... cont
Since M ≥ [Ax ]i ∀i 6= h, then:
k
X
i =1,i 6=h
yi[Ax ]i ≤
k
X
i =1,i 6=h
yiM = (1 − yh)M.
Therfore:
Pk
i =1xi[Ax ]i = M, e Pk
i =1yi[Ax ]i ≤ M.
From this it follows that
k
X
i =1
xi[Ax ]i ≥
k
X
i =1
yi[Ax ]i ∀y ∈ ∆.
Theorem for the mixed equilibria
Theorem
Let x be a strategy belonging to int∆, that is it is a mixed strategy such that xi > 0 ∀i . If [Ax ]i = [Ax ]j ∀i , j, then x ∈ ∆NE. Proof.
Let M = [Ax ]i ∀i . Then:
k
X
i =1
xi[Ax ]i =
k
X
i =1
xiM = M
k
X
i =1
xi = M · 1 = M, and
k
X
i =1
yi[Ax ]i =
k
X
i =1
yiM = M
k
X
i =1
yi = M · 1 = M.
... cont
Then, equation (1) holds always in non strict form, that is:
k
X
i =1
xi[Ax ]i =
k
X
i =1
yi[Ax ]i ∀y ∈ ∆.
This means that x is a symmetric Nash equilibrium. Moreover, this equilibrium is non strict.
Theorem for pure/mixed equilibria
Theorem
Let x be a mixed strategy with only one null component (xh= 0, xi > 0 ∀i 6= h). If [Ax ]i = [Ax ]j ∀i 6= h, j 6= h, and [Ax ]i ≥ [Ax]h ∀i 6= h then x ∈ ∆NE.
Proof. Let M = [Ax ]i ∀i 6= h. Then:
k
X
i =1
xi[Ax ]i =
k
X
i =1
xiM = M
k
X
i =1
xi = M · 1 = M, and
k
X
i =1
yi[Ax ]i =
k
X
i =1,i 6=h
yiM + yh[Ax ]h= M
k
X
i =1,i 6=h
yi + yh[Ax ]h=
= M(1 − yh) + yh[Ax ]h.
... cont
In the second equation we have used the fact that
k
X
i =1
yi = 1 ⇒
k
X
i =1,i 6=h
yi = 1 − yh. By assumption, we have that M ≥ [Ax ]h. It follows that:
M ≥ [Ax]h ⇒
yhM ≥ yh[Ax ]h ⇒
yhM + M − M ≥ yh[Ax ]h ⇒
M − M(1 − yh) ≥ yh[Ax ]h ⇒ M ≥ M(1 − yh) + yh[Ax ]h ⇒
k
X
i =1
xi[Ax ]i ≥
k
X
i =1
yi[Ax ]i ∀y ∈ ∆.
Since x satisfy eq. (1) for each y , then x is a symmetric Nash eq.