Definition (equivalent mixed and behavior strategies)

A mixed strategy σi\sigma_i and behavior strategy bib_i of player ii in an extensive-form game are equivalent to each other if for every mixed/behavior strategy vector σi\sigma_{-i} of the players N{i}N \setminus \{i\} and every vertex xx in the game tree,

ρ(x;σi,σi)=ρ(x;bi,σi)\rho(x; \sigma_i, \sigma_{-i}) = \rho(x; b_i, \sigma_{-i})

i.e. the mixed strategy σi\sigma_i and behavior strategy bib_i are equivalent if for every mixed/behavior strategy vector σi\sigma_{-i} the two strategy vectors (σi,σi)(\sigma_i, \sigma_{-i}) and (bi,σi)(b_i, \sigma_{-i}) induce the same probability of arriving at each vertex in the game tree, in particular for each leaf xx.

Theorem (equivalent utility of equivalent mixed and behavior strategies)

If mixed strategy σi\sigma_i is equivalent to behavior strategy bib_i, then for every mixed/behavior strategy vector σi\sigma_{-i} of the other players and every player jNj \in N,

uj(σi,σi)=uj(bi,σi)u_j(\sigma_i, \sigma_{-i}) = u_j(b_i, \sigma_{-i})

Corollary

Let σ=(σi)iN\sigma = (\sigma_i)_{i \in N} be a vector of mixed strategies. For each player ii let bib_i be a behavior strategy that is equivalent to σi\sigma_i, and denote b=(bi)iNb = (b_i)_{i \in N}. Then for each player ii,

ui(σ)=ui(b)u_i(\sigma) = u_i(b)

Conditions for existence of equivalent mixed strategy to any behavior strategy

Definition (action)

Let xx be a vertex in the game tree that is not the root, and x1x_1 a vertex on the path from the root to xx. The (unique) edge emanating from x1x_1 on the path from the root to xx is called the action at x1x_1 leading to xx.

Corollary

If there exists a path from the root to some vertex x that passes at least twice through the same information set UiU_i of player ii, and if the action leading in the direction of xx is not the same action at each of these information sets, then player ii has a behavior strategy that has no equivalent mixed strategy.

Theorem

Let Γ=(N,V,E,v0,(Vi)iN{0},(px)xV0,(𝒰iN),O,u)\Gamma = (N,V,E,v_0, (V_i)_{i \in N \cup \{0\}}, (p_x)_{x \in V_0}, (\mathcal{U}_{i \in N}), O, u) be an extensive form game that satisfies the condition that at every vertex there are at least two actions. Every behavior strategy of player ii has an equivalent mixed strategy if an only if each information set of player ii intersects every path emanating from the root at most once.


References

  1. M. Maschler, E. Solan, and Shmuel Zamir, Game Theory, Cambridge University Press, 2013, pp. 223, 226.