John Riley 6 December 2010 ANSWERS TO EVEN NUMBERED EXERCISES IN CHAPTER 7 Section 7.1 Exercise 7.1-2: Allais Paradox (a) Draw a tree diagram showing that the prospect C can be represented as a compound gamble between A and (0,0,1). Draw another tree diagram showing that the prospect D can be represented as a compound gamble between B and (0,0,1), where the probability of (0,0,1) is the same. (b) Show that the ranking of A and B must be the same as the ranking of C and D. Hence establish the original version of the Allais Paradox., that is C ~ D ⇔ A ~ E ANSWER (a) D = (ab, 0,1- ab) C = (0, b,1- b) p A = (0,1, 0) 1-p Z = (0, 0,1) p B = (a, 0,1- a) 1-p Z = (0, 0,1) (b) Then (0, b,1 − b) = p (0,1, 0) + (1 − p)(0, 0,1) if and only if p = b . Then (ab, 0,1 − ab) = p(a, 0,1 − a) + (1 − p)(0, 0,1) If A ∼ B ; then for any prospect Z and probabilities p1 , p2 > 0 and p1 + p2 = 1 , C = ( p1 , p2 ; A, Z ) ∼ ( p1 , p2 ; B, Z ) = D Answers to Chapter 7 page 1 John Riley 6 December 2010 Exercise 7.1-4: Rabin paradox You are offered either $100 for sure or one of the following risky alternatives. Which would you accept? Payoff if Number of green balls (out of 100) Payoff if green ball red ball 200 52 55 60 66 70 0 1100 13 20 33 48 57 0 5100 7 18 33 48 57 0 25100 7 18 33 48 57 0 Compare your answers to the answers you gave above in the discussion of the Rabin Paradox. Are the differences consistent with expected utility maximization? ANSWER Define ŵ = w + g where g = 100 . Then the choice is between a final wealth of ŵ and an uncertain payoff of ŵ + ng or ŵ − g depending on the color of the ball drawn from the urn, where n = 1,10,50, 250 . Note that this problem is identical in structure to the one analyzed above so the answers should be exactly the same. Section 7.2 Exercise 7.2-2: Relative risk aversion The degree of relative risk aversion R ( x ) = − xv′′( x ) / v′( x ) . (a) Show that if an individual exhibits constant relative risk aversion, his marginal rate of substitution M ( x1 , x2 ) is constant along a ray from the origin. Assuming that he can trade in claims to each state, show that the risk he will take on rises proportionally with his wealth. Answers to Chapter 7 page 2 John Riley 6 December 2010 (b) Show that if v′( x) = x −1/σ , σ > 0 , the individual exhibits constant relative risk aversion. Hence solve for the constant relative risk aversion utility functions. (c) It is usually argued that individuals exhibit increasing relative risk aversion and constant absolute risk aversion. What does this imply about the shape of wealth expansion paths? ANSWER (a) D B c2 Certainty line A c1 We must show that starting from a point A above the certainty line, the marginal rate of substitution increases along the ray AD under the assumption of increasing risk aversion. A point on the ray is x(θ ) = (1 + θ )( x1 , x2 ) . Then M ( x(1 + θ )) = π 1v′( x1 (1 + θ )) π 1v′( x2 (1 + θ )) Taking the logarithm and differentiating, 1 ∂M M ∂θ = θ =0 x1v′′( x1 ) x2 v′′( x2 ) − = R ( x2 ) − R ( x1 ) . v′( x1 ) v′( x2 ) Since c2 > c1 it follows that the marginal rate of substitution increases along the ray AD. Answers to Chapter 7 page 3 John Riley 6 December 2010 Hence as wealth increases, the consumer chooses a state claims bundle lying below the ray. It is therefore less risky relative to his wealth. (b) Differentiating, v′′( x) = − 1 σ x 1 − −1 σ . Therefore R ( x) = − xv′′( x) 1 = . Thus the CES v′( x) σ family of VNM utility functions exhibit constant relative risk aversion. If σ = 1 , 1− 1 x σ 1− 1 v′( x) = x −1 and so v( x) = ln x . If σ ≠ 1 v( x) = . σ (c) An almost identical argument to that made in (a) establishes that the marginal rate of substitution decreases along a line parallel to the certainty line. Therefore the consumption bundles with a marginal rate of substitution equal to that at A lie in the shaded region. Hence as wealth increases, the consumer chooses a state claims bundle with greater absolute risk but less relative risk. *Exercise 7.2-4 Wealth effects An individual with wealth W must decide how much to invest in a riskless asset with yield 1 + r1 and a risky asset with yield 1 + r2 where E{r2 ) > r1 . (a) If the individual exhibits constant absolute risk aversion, show that his investment in the risky asset is independent of his wealth. *(b) If the individual exhibits decreasing absolute risk aversion, show that he will invest more as his wealth increases. ANSWER (a) From section 7.2, if an individual invests x in the risky asset his marginal expected utility is S U ′( x) = ∑ π sθ s v′( w + xθ s ) , where θ s = r2 s − r1 and w = (1 + r1 )W s −1 We then ask what is the effect of an increase in wealth on the marginal utility of investing in the risky asset. S d U ′( x) = ∑ π sθ s v′′( w + xθ s ) dw s −1 . Answers to Chapter 7 page 4 John Riley 6 December 2010 Let x* be the optimal investment. If d U ′( x* ) > 0 , increasing wealth raises the marginal dw utility of investing in the risky asset. Then, at the higher wealth level, the individual invests more in the risky asset. (a) If absolute risk aversion is constant so that v′′(cs ) = − Av′(cs ) . Substitute this into the above equation, S d U ′( x) = − A∑ π sθ s v′( w + xθ s ) dw s −1 (D.7-1) At the optimum it follows from the first order condition that the right hand side is zero. Thus d U ′( x* ) = 0 dw and so a change in wealth has no effect on the optimal investment. (b) The key is to make use of our assumption that absolute risk aversion is diminishing. We index states so that the yield on the risky asset is lower in higher states. We also define state t to be the lowest state for which θ s < 0 . Substituting A(c) = − v′′(c) into (D.7-1), v′(c) S d U ′( x) = ∑ π sθ s v′′( w + xθ s ) dw s −1 t −1 S s =1 s =t = ∑ π s (−θ s ) A( w + xθ s )v′( w + xθ s ) − ∑ π sθ s A( w + xθ s )v′( w + xθ s ) If θ s > 0, A( w + xθ s ) < A( w) . Hence S ∑ s =t S π sθ s A( w + xθ s )v′( w + xθ s ) < ∑ π sθ s A( w)v′( w + xθ s ) s =t Also, if θ s < 0, A( w + xθ s ) > A( w) . Hence t −1 ∑ s =1 t −1 π s (−θ s ) A( w + xθ s )v′( w + xθ s ) > ∑ π s (−θ s ) A( w)v′( w + xθ s ) s =1 Combining these results, it follows that Answers to Chapter 7 page 5 John Riley 6 December 2010 S d U ′( x) > ∑ − π sθ s A( w)v′( w + xθ s ) dw s =1 Since A(w) is independent of the state, we can take it outside the summation. Then d U ′( x) > − A( w)U ′( x) dw In particular, at x* , d U ′( x* ) > − A( w)U ′( x* ) = 0 . dw Thus investment in the risky asset rises with wealth. Exercise 7.2.6: Concave function A function u (c) where c ∈ is strictly concave if for any c1 and c3 > c1 c2 = (1 − λ )c1 + λ c3 , 0 < λ < 1 ⇒ u (c2 ) > (1 − λ )u (c1 ) + λu (c3 ) (a) Rearrange these two expressions and hence show that u (c) is concave if λ (c3 − c2 ) − (1 − λ )(c2 − c1 ), 0 < λ < 1 ⇒ λ (u (c3 ) − u (c2 )) − (1 − λ )(u (c2 ) − u (c1 )) < 0 (b) Hence show that the concavity of u (c) is equivalent to the statement that for all c1 and c3 > c1 and c2 ∈ (c1 , c3 ) c2 − c1 u (c2 ) − u (c1 ) u (c2 ) − u (c1 ) u (c3 ) − u (c2 ) > < . . That is, c3 − c2 u (c3 ) − u (c2 ) c2 − c1 c3 − c2 ANSWER (a) c2 = (1 − λ )c2 + λ c2 = (1 − λ )c1 + λ c3 and u (c2 ) = (1 − λ )u (c2 ) + λu (c2 ) > (1 − λ )u (c1 ) + λ u (c3 ) . Rearranging these statement yields the desired result. (b) Hence c2 − c1 u (c2 ) − u (c1 ) u (c2 ) − u (c1 ) u (c3 ) − u (c2 ) λ and so > . = < c3 − c2 1 − λ u (c3 ) − u (c2 ) c2 − c1 c3 − c2 Answers to Chapter 7 page 6 John Riley 6 December 2010 Section 7.3 Exercise 7.3-2: First order stochastic dominance (a) For S = 2 and S = 3 confirm the following identity. S ∑ s =1 S −1 t t =1 s =1 π s us = Π S uS − ∑ Π t (ut +1 − ut ) where Π t ≡ ∑ π s . (4.7-2) Prove by induction that the identity holds for all S. Consider the two consequences (c, π A ), (c, π B ) where cs +1 > cs . The second probability distribution stochastically dominates the first there is less weight in the left tail, that is Π sB ≤ Π sA , s = 1,..., S . (b) If v(⋅) is an increasing function, show that the individual prefers B over A. s (c) Define H s = ∑ Π t and Δut = ut +1 − ut . Appeal to the identity in (a) to show that t =1 S ∑ t =1 S Π t Δut = H S −1 − ∑ ( H S −1 (Δut +1 − Δut ) . s =1 Substitute this expression into (4.7-2), hence prove Proposition 7.3-2. ANSWER (a) π 1u1 + π 2u2 = π 1u1 + u2 (Π 2 − π 1 ) = Π 2u2 − π 1 (u2 − u1 ) = Π 2u2 − Π1 (u2 − u1 ) Suppose that the Proposition is true with S states. We must show that the proposition must also be true with S + 1 states. Let (π 1 ,...., π S , π S +1 ) be the probability vector with S + 1 states. Then π π S 1 ( ,...., ) 1 − π S +1 1 − π S +1 is a probability vector over S states and so π u ΠS 1 S −1 π 1u1 + .... + S S = uS − ∑ Π t (ut +1 − ut ) . 1 − π S +1 1 − π S +1 1 − π S +1 1 − π S +1 t =1 Therefore S −1 π 1u1 + .... + π S uS = uS − ∑ Π t (ut +1 − ut ) . t =1 Answers to Chapter 7 page 7 John Riley 6 December 2010 Hence S −1 π 1u1 + .... + π S uS + π S +1uS +1 = π S +1uS +1 + Π S uS − ∑ Π t (ut +1 − ut ) t =1 S −1 = uS +1 (Π S +1 − Π s ) + Π s uS − ∑ Π t (ut +1 − ut ) t =1 S = Π S +1uS +1 − ∑ Π t (ut +1 − ut ) . t =1 (b) Note that Π SA = Π SB = 1 . Therefore S ∑ s =1 S S −1 S −1 s =1 t =1 t =1 π sBu (cs ) − ∑ π sAu (cs ) = −∑ Π tB (ut +1 − ut ) + ∑ Π tA (ut +1 − ut ) S −1 = ∑ (Π tA − Π tB )(u (ct +1 ) − u (ct )) > 0 . t =1 (c) From the identity, S −1 ∑ t =1 S −2 Π t Δut = H S −1ΔuS −1 − ∑ H t (Δut +1 − Δut ) . t =1 Therefore S ∑ s =1 S −1 S −2 t =1 t =1 π s us = uS − ∑ Π t (ut +1 − ut ) = us − H S −1ΔuS −1 + ∑ H t (Δut +1 − Δut ) . If ct − ct −1 = δ > 0 and u is concave then Δut +1 < Δut . Therefore if ( H1B ,..., H SB−1 ) < ( H1A ,..., H SA−1 ) for probability vectors π B and π A , it follows that S ∑ s =1 S π sBus > ∑ π sAus . s =1 Exercise 7.3.4 Equivalent definition of conditional stochastic dominance (a) If the probability distribution π exhibits conditional strict stochastic dominance over πˆ , show that 1− ˆ Π t −1 Π > 1 − t −1 ˆ Πt Π t Hence establish the “only if” part of the following proposition Answers to Chapter 7 page 8 John Riley 6 December 2010 Proposition 7.3-5: Conditional strict stochastic dominance The probability distribution π exhibits strict conditional stochastic dominance over πˆ if πs and only if Πs > πt Πt , ∀t and ∀s > t . (4.7-3) (b) Show that if (4.7-3) holds then ˆ Π s −1 Π < s −1 . Use this to establish strict conditional ˆ Πs Π s stochastic dominance. ANSWER (a) ONLY IF If π exhibits conditional stochastic dominance over πˆ than for all s < t s ˆ Πs Π ≤ s where Π s ≡ ∑ π s . ˆ Πt Π θ =1 t In particular, ˆ ˆ Π t −1 Π π Π Π πˆ ≤ t −1 hence t = 1 − t −1 ≥ 1 − t −1 = t . ˆ ˆ ˆ Πt Πt Πt Π Π Π t t t (b) IF πt Πt ≥ πˆt , t = 1,..., S . Then ˆ Π t π πˆ 1 − t ≤ 1 − t , t = 1,..., S , ˆ Πt Π t Suppose that Hence ˆ Π t −1 Π ≤ t −1 , t = 1,..., S . ˆ Πt Π t Therefore t t ˆ ˆ Πs Π Π Π t −1 = × τ −1 ≤ × = s. ˆ ˆ Π t τ = s +1 Πτ τ = s +1 Π Π t t Section 7.4 Exercise 7.4-2: Pareto efficiency and monotonicity with more than one binding incentive constraint Answers to Chapter 7 page 9 John Riley 6 December 2010 Extend the analysis to show that as long as all the binding constraints are associated with less costly actions, the efficient wage is higher in states where output is higher. ANSWER Suppose that the set of binding constraints is indexed by B. That is constraint i is binding if i ∈ B . By hypothesis, xi < x* , i ∈ B . The new maximization problem is Max{U P | U A ( x* , w) ≥ U A , U A ( x* , w) ≥ U A ( xi , w), for all i ∈ B} . w The Lagrangian of this optimization problem is L = U P + λ (U A ( x* , w) − U A ) + ∑ μi (U A ( x* , w) − U A ( xi , w)) i∈B = U P + (λ + ∑ μi )(U A ( x* , w) − ∑ μi ) μiU A ( x, w)) + a constant i∈B i∈B S S S s =1 s =1 = ∑ π s ( x* )( ys − ws ) + (λ + ∑ μi ∑ π s ( x* )v( ws ) − μ ∑ π s ( xi )v( ws ) s =1 i∈B plus constant terms. The first order conditions are therefore ∂L = −π s ( x* ) + (λ + ∑ μiπ s ( x* )v′( ws* ) − ∑ μiπ s ( xi )v′( ws* ) = 0 , s = 1,..., S ∂ws i∈B i∈B Hence π (x ) 1 = λ + ∑ μi − ∑ μi s *i where xi < x* * v′( ws ) π s (x ) i∈B i∈B By hypothesis, the likelihood ratio π s ( x* ) is increasing in the output state s. Thus the π s ( x) right hand side of this expression increases with s. Therefore, since v(⋅) is concave, ws is increasing in s. That is, for efficiency, the higher the output, the higher is the payment to the agent. Answers to Chapter 7 page 10