Econ 203C Problem Set # 6 — Answer Key Problem 1. For arbitrary (possibly infinite) C, let L (C) denote the set of all lotteries with finite support. Generalize the vNM Theorem demonstrated in class to arbitrary infinite sets of consequences C. Suppose that  on L (C) satisfies the vNM axioms. We will derive the existence of an EU representation with u : C → R. The other parts of the vNM theorem generalize trivially. Fix any c, c0 ∈ C such that δc  δc0 . For any finite subset D ⊆ C, let L (D) denote the set of lotteries with support in D. Define a utility function u : C → R as follows. For any x ∈ C, consider the restriction of the preference relation  to L ({c, c0, x}). This restriction obviously satisfies the vNM axioms; hence there exists a vNM utility function representing  on L ({c, c0, x}) vx . By invariance of the representation to positive affine transformations, vx can be chosen such that vx (c) = 1 and vx (c0 ) = 0. Set u(x) := vx (x) . We need to show that u represents  on L (C) . To see this, take any two finite lotteries L = [pi , xi ] and L0 = [p0i , xi ] such that L  L0 ; let D denote the union of their supports. By the vNM Theorem,  on L (D) has a (unique) EU representation with vNM utility function vD satisfying vD (c) = 1 and vD (c0 ) = 0. In particular, X X pi vD (xi ) > p0i vD (xi ) . i i Since L (D) ⊇ L ({c, c0, x}) for all x ∈ D, clearly (clearly?) vD = u; hence X X pi u (xi ) > p0i u (xi ) , i i verifying the claim since L and L0 were arbitrary. 1 ¤ Problem 2. Assume C to be finite. A lottery is binary iff it assigns positive probability to at most two distinct outcomes; let L2 (C) denote the set of all binary lotteries on C. Prove the analogue to the von Neumann-Morgenstern theorem. That is prove in particular that if a preference relation on L2 (C) satisfies Weak Order, Independence, and the Archimedean axiom,  has an EU representation. 1. (a) Verify that Steps 1 and 2 in our proof can be used literally, so you don’t need to redo them (b) Why can’t you always apply Step 3-4 literally? (c) However, you can apply Step 3-4 sometimes; this is helpful in obtaining a workable proof in the general case. b) The difficulty is that the substitution of certain outcomes by bestworst lotteries may lead to non-binary lotteries. c) I will write binary lotteries as αδx + (1 − α) δy . Let x− and x+ denote the % −worst and -best outcomes, respectively, and write α for αδx+ + (1 − α) δx− . For any x ∈ X, let u (x) be the unique α such that δx ∼ α ; uniqueness and existence follows from Steps 1 and 2 in our proof of the von NeumannMorgenstern theorem (part a). Let αδx + (1 − α) δy denote any binary lottery. We need to show that αδx + (1 − α) δy ∼ αu(x)+(1−α)u(y) . (1) Consider first the special case that y = x− . By Independence, αδx + (1 − α) δx− ∼ α (u (x) δx+ + (1 − u (x)) δx− ) + (1 − α) δx− = αu(x) , (2) verifying (1). Consider now the general case. By (2), δy ∼ βδx + (1 − β) δx− , where u(y) . Hence by Independence, β = u(x) αδx +(1 − α) δy ∼ αδx +(1 − α) (βδx + (1 − β) δx− ) = (α + (1 − α) β) δx +(1 − α − (1 − α) β) δx− . By (2) again, the latter is indifferent to as was to be shown. ¤ 2 (α+(1−α)β)u(x) = αu(x)+(1−α)u(y) , Problem 3. c) Let p∗ denote your answer to b). If you are an EU maximizer with Bernouilli utility function u, u satisfies p∗ u(∞) + (1 − p∗ )u(1 − q) = u(1), 1 . By this where q = 0.1. With CRRA with ρ > 1, write u(x) = − xρ−1 normalization, u(1) = −1 and u(∞) = 0, and thus (1 − p∗ ) = 1, (1 − q)ρ−1 whence log(1 − p∗ ) = (ρ − 1) log(1 − q) and thus ρ= For smallish q and p∗ , thus log(1 − p∗ ) + 1. log(1 − q) ρ≈ p∗ + 1. q e) p(q)u(∞) + (1 − p(q))u(1 − q) = u(1) Hence we get for ρ = 2 p(q)0 + (1 − p(q)) 1 ≈ 1, and thus p(q) ≈ q. 1−q f) Two such axioms are the following (each is necessary and sufficient for boundedness of the utility function on its own). AXIOM 1. For all x, y ∈ R+ such that y > x there exists α ∈ (0, 1) such that, for all z ∈ R+ , δy % [α, z; (1 − α), x]. AXIOM 2. For all x ∈ R+ and all α ∈ (0, 1) there exists y ∈ R+ such that, for all z ∈ R+ , δy % [α, z; (1 − α), x]. Note the different order of quantifiers in each case; they make the content of the axiom quite different. 3
© Copyright 2025 Paperzz