Theorem

If player ii's preference relation i\succsim_i over the set of compound lotteries ˆ\hat{\mathcal{L}} is complete and transitive, and satisfies the four von Neumann-Morgenstern axioms, then this preference relation can be represented by a linear utility function.

von Neumann-Morgenstern axioms

Assume i\succsim_i defined over the set of compound lotteries ˆ\hat{\mathcal{L}}. Player ii's utility function, representing his preference relation i\succsim_i, is therefore a function ui:ˆu_i : \hat{\mathcal{L}} \to \mathbb{R} satisfying

ui(Lˆ1)ui(Lˆ2)Lˆ1iLˆ2,Lˆ1,Lˆ2ˆu_i(\hat{L}_1) \geq u_i(\hat{L}_2) \iff \hat{L}_1 \succsim_i \hat{L}_2, \quad \forall \hat{L}_1, \hat{L}_2 \in \hat{\mathcal{L}}

Axiom of Continuity

For every triplet of outcomes AiBiCA \succsim_i B \succsim_i C, there exists a number θi[0,1]\theta_i \in [0,1] such that

Bi[θi(A),(1θi)(C)]B \approx_i [\theta_i(A), (1-\theta_i)(C)]

(where i\approx_i denotes an indifference relation)

Axiom of Monotonicity

Let α,β\alpha, \beta be numbers in [0,1][0,1], and suppose that AiBA \succ_i B (strict preference). Then,

[α(A),(1α)(B)]i[β(A),(1β)(B)][\alpha(A) , (1-\alpha)(B)] \succsim_i [\beta(A) , (1-\beta)(B)]

if and only iff αβ\alpha \geq \beta.

Theorem

If a preference relation satisfies the Axioms of Continuity and Monotonicity, and if AiBiCA \succsim_i B \succsim_i C, and AiCA \succ_i C, then the value of θi\theta_i defined in the Axiom of Continuity is unique.

Corollary

If a preference relation i\succsim_i over ˆ\hat{\mathcal{L}} satisfies the Axioms of Continuity and Monotonicity, and if AKiA1A_K \succ_i A_1, then for each k=1,2,...,Kk = 1,2,...,K there exists a unique θik[0,1]\theta_i^k \in [0,1] such that

Aki[θik(AK),(1θik)(A1)]A_k \approx_i [\theta_i^k (A_K), (1-\theta_i^k)(A_1)]

The corollary and the fact that A1i[0(AK),1(A1)]A_1 \approx_i [0 (A_K), 1(A_1)] and Aki[1(AK),0(A1)]A_k \approx_i [1 (A_K), 0(A_1)] imply that

θi1=0,θiK=1\theta_i^1 = 0, \quad \theta_i^K = 1

Axiom of Simplification of Compound Lotteries

For each j=1,...,Jj = 1,...,J, let LjL_j be the simple lottery

Lj=[p1j(A1),p2j(A2),...,pKj(AK)]L_j=[p_1^j (A_1), p_2^j (A_2),...,p_K^j (A_K)]

and let Lˆ\hat L be the compound lottery

Lˆ=[q1(L1),q2(L1),...,qJ(LJ)]\hat L = [q_1(L_1), q_2(L_1), ..., q_J(L_J)]

For each k=1,...,Kk=1,...,K, define the overall probability that the outcome of Lˆ\hat L will be AkA_k,

rk=q1pk1+q2pk2+...+qJpkJr_k = q_1 p_k^1 + q_2 p_k^2 + ... + q_J p_k^J

Consider simple lottery

L=[r1(A1),r2(A2),...,rK(AK)]L = [r_1(A_1),r_2(A_2),...,r_K(A_K)]

Then,

LˆiL\hat L \approx_i L

Axiom of Independence

Let Lˆ=[q1(L1),q2(L1),...,qJ(LJ)]\hat L = [q_1(L_1), q_2(L_1), ..., q_J(L_J)] be a compound lottery, and let MM be a simple lottery. If LjiML_j \approx_i M then

Lˆi[q1(L1),...,qj1(Lj),qj(M),qj+1(Lj+1),...,qJ(LJ)]\hat L \approx_i [q_1(L_1),...,q_{j-1}(L_j),q_j(M),q_{j+1}(L_{j+1}),...,q_J(L_J)]

Notes


References

  1. M. Maschler, E. Solan, and Shmuel Zamir, Game Theory, Cambridge University Press, 2013, pp. 14-17.
  2. https://en.wikipedia.org/wiki/Von_Neumann–Morgenstern_utility_theorem#The_theorem
  3. Neumann, John von and Morgenstern, Oskar, Theory of Games and Economic Behavior. Princeton, NJ. Princeton University Press, 1953.
  4. https://isa-afp.org/entries/Neumann_Morgenstern_Utility.html