Chu Thanh Duc – MDE10 Chapter 4. Consumer Theory II Two essential tools of analysis in the modern treatment of consumer theory: The Indirect Utility Function The Expenditure Function 4.1 Indirect Utility and Expenditure 4.1.1 The Indirect Utility Function The direct Utility Function: The ordinary utility function, u(x), is defined over the consumption set X and represents the consumer’s preferences directly. The indirect Utility Function: The problem: max u ( x) s.t. y p.x, x 0 x the solution x* =x*(p,y) The Indirect Utility Function: u(x*) = u*(p,y) = v(p,y) The indirect Utility Function represent the relation between prices, income and the highest level of utility achieved. v : R n 1 R defined as: v( p, y ) max u ( x) s.t. y p.x 0, x 0. x v(p,y) is called indirect utility function. This function is clearly well-defined since, when preferences are monotonic and strictly convex, a unique solution x(p,y) to the consumer’s problem exists. In the maximization problems max u ( x) s.t. y p.x, x 0 , continuity of the x constraint function in the parameters is sufficient to guarantee that v(p,y) will be continuous in p and y. Theorem 4.1.1 Properties of the Indirect Utility Function Let preferences be monotonic and differentiable, and let p>>0 and y>0. Then v(p,y) has these properties: 1. Homogeneous of degree zero in p and y 2. Increasing in y 3. Non-increasing in p 4. Quansiconvex in p Geoffrey A. Jehle – Advanced Microeconomic Theory 1 Chu Thanh Duc – MDE10 Proof: 1. Homogeneous of degree zero in p and y: Equiproportionate changes in all p and y leave the consumer’s budget set unchanged. y p1 x1 p2 x2 ty tp1 x1 tp2 x2 (t 0) The set of feasible choices, and so the maximal level of utility the consumer can achieve, must therefore also remain the same. Changing all p an y by proportion t>0 must leave the maximal utilitly unchanged. 2. Increasing in y and Non-increasing in p Considering the Lagrangian for the utility maximization problem: L( x, ) u ( x) [ y p.x] By the Envelop Theorem, we have: v( p, y ) L pi pi x* 0 ( x*, ) v( p, y ) L 0 y y ( x*, ) as the proof. 3. Quansiconvex in p The Lagrangian multiplier ( ) will measure the sensitivity of the objective function u(x) to changes in the constraint constant (y). (See the Exercise 2.29). Thus the value of the Lagrangian multiplier at the solution measure the marginal utility of income. Let B1, B2, and Bt be the budget sets available when the consumer has income y and faces prices p1, p2, and p t tp1 (1 t ) p 2 , then: B 1 {x | p 1 .x y} B 2 {x | p 2 .x y} B t {x | p t .x y}. We need to show that: v( p t , y) max[ v( p1 , y), v( p 2 , y)] So we will show that every choice the consumer can possibly make when she faces budget B t is a choice which could have been made when she faced either budget B 1 or budget B 2 . It would be the case that every level of utility she can achieve faving B t is a level she could have achieved either when facing B 1 or when facing B 2 . Then the maximum level of utility Geoffrey A. Jehle – Advanced Microeconomic Theory 2 Chu Thanh Duc – MDE10 that she can achieve over B t could be no larger than at least one of the maximum level of utility that she can achieve over B 1 or the one she can achieve over B 2 . We want to show that if x B t , then x B 1 or x B 2 for all t [0,1] . It is easy to realize that if t=1 or t=0 x B t , then x B 1 or x B 2 . For t (0,1) . Suppose that if x B t , then x B 1 and x B 2 , then x. p1 y and x. p 2 y . Cos’s t (0,1) t>0 and (1-t)>0 t.x. p1 t. y and (1 t ).x. p 2 (1 t ) y t.x. p1 (1 t ).x. p 2 y x. p t y x B t contradicting our orginal assumption. if x B t , then x B 1 or x B 2 for all t [0,1] . v( p t , y) max[ v( p1 , y), v( p 2 , y)] v(p,y) is quansiconcave function in p ./. The indirect utility function tells us the maximal level of utility the consumer can achieve facing different prices and incomes. The demand functions give us the utility maximizing choices of each commodity he will make facing different prices and incomes. To get the indirect utility function, we simply substitute the demand functions into the direct utility function. Theorem 4.1.2 Roy’s Identity To get the indirect utility function, we simply substitute the demand functions into the direct utility function. There is a question that how to derive the direct utility function from the indirect utility function ? This theorem will answer this question. Theorem: Let v(p,y) be any indirect utility function satisfying the conditions of Theorem 4.1.1. Then, xi ( p, y ) v( p, y ) / pi v( p, y ) / y Geoffrey A. Jehle – Advanced Microeconomic Theory 3 Chu Thanh Duc – MDE10 Fifth Property of an indirect utility function: An indirect utility function is demand – generating.: any demand function can be generated from the indirect utility function. Proof: Let x* and solve the Kuhn – Tucker conditions. The solution ( p, y ) gives us the marginal utility of income at the consumer equilibrium and xi * xi ( p, y) gives the consumer’s demand function for good i. By the Envelope Theorem, we have: v( p, y ) L pi pi ( p, y ) x i ( p, y ) ( x*, ) v( p, y ) L ( p, y ) y y ( x*, ) xi ( p, y ) v( p, y ) / pi v( p, y ) / y 4.1.2 The Expenditure Function What is the minimum level of money expenditure, or outlay, which the consumer must make facing a given set of prices in order to achieve a given level of utility? In this construction, we ignore any limitations imposed by the consumer’s income and simply ask what the consumer would have to spend in order to achieve some particular level of utility. X2 X1 Iso – expenditure curve: e p1 x1 p2 x2 p.x The problem: e( p, u ) min x Geoffrey A. Jehle – Advanced Microeconomic Theory s.t. u u ( x) 0, x0 4 Chu Thanh Duc – MDE10 The solution is x h ( p, u) that depends on prices and utility level. If preferences are monotonic and strictly convex, the solution will be unique. The lowest expenditure necessary to achieve utility u at prices p will be equal to cost of the bundle x h ( p, u) : e( p, u ) p.x h ( p, u ) . We can use the consumer’s expenditure minimization problem to explore a very different kind of “demand behavior”, this one entirely unobservable or hypothetical. It is different from the Marshallian demand which is observable. If we fix the level of utility the consumer is permitted to achieve at some arbitrary level u, how will his purchases of each good behave as we change the prices he faces? Utility – constant demand functions Hicksian Demand Functions: Fix the utility level and solve the problem: e( p, u ) min p.x s.t. u u ( x) 0, x x0 to have consumption bundle will change: h x1 ( p, u ) : x1h ( p10 , p 20 , u ) , x 2h ( p10 , p 20 , u ) . Change price p1, the optimal choice h x 2 ( p, u ) : x1h ( p11 , p 20 , u ) , x 2h ( p11 , p 20 , u ) . Theorem 4.1.3 Properties of the Expenditure Function Let preferences be monotonic and let p>>0. Let u>u(0) and let e(p,u) be defined as in 4.1.3. Then e(p,u) is: 1. Increasing in u 2. Non-decreasing in p 3. Homogeneous of degree 1 in p 4. Concave in p 5. Also, the price partial derivatives of e(p,u) are the Hicksian demand functions e( p, u ) xih ( p, u ) pi Geoffrey A. Jehle – Advanced Microeconomic Theory 5 Chu Thanh Duc – MDE10 Proof: p.x The problem: e( p, u ) min x s.t. u u ( x) 0, x0 Lagrangian for this problem: L( x, ) p.x [u u ( x)] If x h x h ( p, u ) 0 solves the minimization problem, then x h and satisfy the KuhnTucker conditions: L u ( x h ) pi 0 xi xi xih L u ( x h ) xih [ pi ]0 xi xi L u u( x h ) 0 L [u u ( x h )] 0 u ( x h ) 0 for at least one From the Kuhn-Tucker conditions we must therefore have: p j x j j. By the monotonicity of preference, we have: pj u ( x h ) 0 0. x j u ( x h ) x j Prove the property 1: By the Envelop theorem: e( p, u ) L u u 0 ( x , ) h e(p,u) is increasing in u. Prove the properties 2 and 5: By the Envelop theorem we have: e( p, u ) L pi pi xih ( p, u ) 0 ( x h , ) e(p,u) is in-decreasing in p. Prove the property 3: e(p,u) is homogeneous of degree 1 e( p, u ) p1 .x1h p 2 .x 2h p n .x nh e(t. p, u ) t.e( p, u ) Geoffrey A. Jehle – Advanced Microeconomic Theory 6 Chu Thanh Duc – MDE10 e(p,u) is homogeneous of degree 1. Prove the property 4: e(p,u) is concave in p. We need to prove that: e( p t , u) t.e( p1 , u) (1 t ).e( p 2 , u) Let x1 , x 2 , x t minimize expenditure to achieve u when prices are p1 , p 2 , p t respectively. By the definition of e(p,u), we have: p 1 .x1 p 1 .x t tp1 .x 1 tp1 .x t p 2 .x 2 p 2 .xt (1 t ) p 2 .x 2 (1 t ) p 2 .x t t.e( p 1 , u ) (1 t ).e( p 2 , u ) e( p t , u ) 4.1.3 Relations between the Two From the definitions of the expenditure function: e( p, v( p, y )) min p.x s.t. u ( x) v( p, y ) e( p, v( p, y )) min p.x s.t. u ( x) max u ( x' ) s.t. x x p.x' y We must of course have: e( p, v( p, y )) y . From the definition of the indirect utility function: v( p, e( p, u )) max u ( x) s.t. x p.x e( p, u ) Substituting from the definition of the expenditure function, we have: v( p, e( p, u )) max u ( x) s.t. x p.x min p.x' s.t. u ( x' ) u x' We must have: v( p, e( p, u )) u Theorem 4.1.4 Identities Relating Indirect Utility and Expenditure Functions Let v(p,y) and e(p,u) be the indirect utility function and expenditure function for some consumer. Then the following relations between the two obtain for all prices p, incomes y, and utility level u: e( p, v( p, y )) y v( p, e( p, u )) u This theorem points us to an easy way to derive either one directly from knowledge of the other, thus requiring us to solve only one optimization problem and giving us the choice of which one we care to solve. Geoffrey A. Jehle – Advanced Microeconomic Theory 7 Chu Thanh Duc – MDE10 v(p,u) is strictly increasing in y. v( p, e( p, u )) u . Invert the indirect utility function in its income variable, we have: e( p, u) v 1 ( p : u) e( p, u ) is strictly increasing in u. e( p, v( p, u )) u . Invert the expenditure function in its utility variable, we have: v( p, u) e 1 ( p : y) Example: The CES direct utility function gives the indirect utility function: v( p, y) ( p1r p2r ) 1 / r . y For an income level equal to e(p,u), we must have: v( p, e( p, u)) ( p1r p2r ) 1 / r .e( p, u) By the theorem, we have ( p1r p2r ) 1 / r .e( p, u) u e( p, u ) ( p1r p2r )1 / r .u Theorem 4.1.5 Identical Relations Between Marshallian and Hicksian Demand Functions For any prices p, income y, and utility level u, the following identical relations hold between the consumer’s Hicksian and Marshallian demand functions: xi ( p, y) xih ( p, v( p, y)) for all p, y and i 1,2,, n xih ( p, u ) xi ( p, e( p, u )) for all p, y and i 1,2,, n Example: The Hicksian demands are: xih ( p, u ) ( p1r p2r )1/ r )1 pir 1u r r 1 / r y The indirect utility function is: v( p, y ) ( p1 p 2 ) We have: xi ( p, y ) xih ( p, v( p, y ) ( p1r p 2r ) (1 / r )1 pir 1 ( p1r p 2r ) 1 / r y pir 1 y p1r p 2r 4.2 Properties of Consumer Demand In statistically estimating consumer demand systems, characteristics of demand behavior predicted by the theory are used to provide restrictions on the values which estimated parameters are allowed to take, thereby ensuring that the empirical estimates are at least logically consistent with the underlying theory from which they are constructed. Geoffrey A. Jehle – Advanced Microeconomic Theory 8 Chu Thanh Duc – MDE10 Relative Price and Real Income Economists generally prefer to measure important variables in real. Relative prices and real income are two such real measures Relative price: By the relative price of some good, we mean the number of units of some other good which must be foregone in order to acquire 1 unit of the good in question. pi $ / unit i unit j p j $ / unit j unit i measure the units of good j forgone per unit of good i acquired. Consumer’s real income: We mean the total number of units of some commodity which could be acquired if the consumer spent his entire money income on that commodity. Real income interm of good j: y $ units of j p j $ / unit of j Theorem 4.2.1 Homogeneity The consumer’s demand function xi ( p, y ) , are homogeneous of degree zero in all prices and income. Proof: Equiproportionate changes in all p and y leave the consumer’s budget set unchanged. y p1 x1 p2 x2 ty tp1 x1 tp2 x2 (t 0) So the optimal point is unchanged. xi ( p, y) xi (t. p, t. y) the consumer’s demand function xi ( p, y ) is homogeneous of degree zero. Application: With t 1 0 pn We have: xi ( p, y) xi (tp, ty) xi ( Geoffrey A. Jehle – Advanced Microeconomic Theory p p1 y ,, n1 ,1, ) pn pn pn 9 Chu Thanh Duc – MDE10 Demand for each of the n goods depends only on (n-1) relative prices and the consumer’s real income. 4.2.2 Income and Substitution Effects Ordinarily, a consumer will buy more of a good when its price declines, and less when its price increases. However, these cases are not necessarily true. Substitution effect: Since all goods are taken to be desirable by the consumer, even if the consumer’s total command over goods were unchanged, we would expect him to substitute more of the good which has become relatively cheapter for less of the goods which are now relatively more expensive. The substitution effect is that (hypothetical) change in consumption which would occur if relative prices were to change to their new level but the maximum level of utility the consumer can achieve were kept the same as before the price changes. Change price of good 1 from x10 to x11 , the problem is that: min p11 x1 p20 x2 s.t. u( x) u 0 Income effect: When the price of any one good declines, the consumer’s total command over all resources is effectively increased, allowing him to change his purchases of all goods in any way he sees fit. The effect on quantity demanded of this generalized increase in purchasing power is called the income effect. The income effect is defined as the residual out of the total effect which is left after the substitution effect. Total effect: Change price of good 1 from x10 to x11 , the problem is that: max u( x) s.t. p11 x1 p20 x2 y Slutsky Equation – Fundamental Equation of Demand Theory: the general analytical relationships between total effect, substitution effect, and income effect are summarized by the Slutsky Equation. Geoffrey A. Jehle – Advanced Microeconomic Theory 10 Chu Thanh Duc – MDE10 Theorem 4.2.2 The Slutsky Equation Let x(p,y) be the consumer’s Marshallian demand system. Let u* be the level of utility the consumer achieves at prices p and income y. Then, x j ( p, y ) p i x hj ( p, u*) x j ( p, y ) x i ( p, y ) p y i TE IE SE Proof: The identity linking the Marshallian and Hicksian demand function: x hj ( p, u*) x j ( p, e( p, u*)) x hj ( p, u*) pi x j ( p, e( p, u*)) pi x j ( p, e( p, u*)) e( p, u*) e( p, u*) pi (1) By the assumption , u* is the level of utility the consumer achieves facing prices p and having income y (see the theorem) u*=v(p,y) the minimum expenditure at prices p and utility u* will therefore be the same as the minimum expenditure at price p and utility v(p,y) e( p, u*) e( p, v( p, y )) y (2) We have: e( p, u*) xi ( p, y ) (3) pi Substitute (2) and (3) into (1), we have: x j ( p, y ) p i TE x hj ( p, u*) x j ( p, y ) x i ( p, y ) p y i SE IE Theorem 4.2.3 Negativity of Own-Substitution Terms: Let xih ( p, u ) be the Hicksian demand for good i. Then, xih ( p, u ) 0 pi Proof: Geoffrey A. Jehle – Advanced Microeconomic Theory 11 Chu Thanh Duc – MDE10 e( p, u ) xih ( p, u ) pi 2 e( p , u ) pi 2 xih pi By the theorem 4.1.3, the expenditure function is a concave function of p./. Norminal or superior goods: consumption increases as real income increases, holding relative prices constant. Inferior goods: consumption decreases as real income increases, holding relative prices constant. Theorem 4.2.4 The Law of Demand Let preferences be complete, transitive, reflexive, monotonic, and strictly convex. If a good is a normal good, then a decrease in price will cause an increase in quantity demanded. If a decrease in price causes a decrease in quantity demanded, then the good must be an inferior good. Theorem 4.2.5 Symmetry of Substitution Terms Let x h ( p, u) be the consumer’s system of Hicksian demands. Then, h xih ( p, u ) x j ( p, u ) p j pi Proof: xih ( p, u ) 2 e( p, u ) p j pip j By the Young’s Theorem: 2 e( p , u ) 2 e( p , u ) pi p j p j pi h xih ( p, u ) x j ( p, u ) p j pi Geoffrey A. Jehle – Advanced Microeconomic Theory 12 Chu Thanh Duc – MDE10 Theorem 4.2.6 Negative Semi-Definiteness of the Substitution Matrix x h ( p, u ) Let x h ( p, u) be the consumer’s system of Hicksian demands, and let i p j i , j 1n represent the entire matrix of Hicksian substitution terms. Then this matrix is negative semi-definite. Proof: xih ( p, u ) 2 e( p, u ) p j pip j The expenditure functions is concave in prices. From the theorem 2.1.3, the matrix of secondorder partials (the Hessian) of a concave function is negative semi-definite Theorem 4.2.7 Negative Semi – definiteness of the Slutsky Matrix Let x(p,y) be the consumer’s Marshallian demand system. Define the Slutsky matrix as the n n matrix of price and income responses given by: xi ( p, y ) x ( p, y ) x j ( p, y ) i y p j i , j 1,n Then the theory of the preference-maximizing, atomistic consumer requires that the Slutsky matrix be negative semi – definite. 4.2.3 Some Elasticity Relations If preferences are monotonic, at least one good will be bought in a positive amount. Definition 4.2.1 Demand Elasticities and Income Shares Let xi ( p, y ) be the consumer’s Marshallian demand for good i. Then let Income elasticity: i Price elasticity: ij xi ( p, y) y y x i ( p, y ) xi ( p, y ) p j p j xi ( p, y ) And Income Share: si p i x i ( p, y ) so that si 0 and y n s i 1 i 1. Theorem 4.2.8 Aggregation in Consumer Demand Geoffrey A. Jehle – Advanced Microeconomic Theory 13 Chu Thanh Duc – MDE10 Let x(p,y) be the consumer’s Marshallian demand system. Let i , ij , and si be income elasticity, cross-price elasticity and income share. Then the following relations must hold between income shares, price, and income elasticities of demand: n Engel Aggregation: s i i 1 i 1 n Cournot Aggregation: s i 1 i ij s j Proof: Prove (1) y p.x( p, y ) n 1 pi i 1 n n xi ( p, y) p x x ( p, y) y i i i si i y y y xi i 1 i 1 Prove (2) y p.x( p, y ) . Differentiating both sides with respect to p j n 0 pi i 1 n xi ( p, y ) xj p j x j pi i 1 xi ( p, y ) p j Multiply both sides of the equation by p j / y and get: xj pj y n i 1 n n pi xi ( p, y ) p x x ( p, y ) p j pj i i i si ij y p j y p j xi i 1 i 1 n si ij s j i 1 4.3 Duality In Consumer Theory There is a question: Starting with an expenditure or an indirect utility function, can we “work backwards” to discover the underlying direct utility function that would have generated it? This question is answered by following the mathematical “duality” between various optimization problems used to describe consumer behavior. Geoffrey A. Jehle – Advanced Microeconomic Theory 14 Chu Thanh Duc – MDE10 4.3.1 Expenditure and Consumer Preferences 4.3.2 Indirect Utility and Consumer Preferences The duality between direct and indirect utility functions. Suppose we are given a continuous function v(p,y), homogeneous of degree zero in p and y, increasing in y, non-increasing and quasiconvex in p. It can be shown that there exists a non-decreasing, quasiconcave direct utility function u*(x) which rationalizes v(p,y). v( p, y ) is homogeneous of degree zero in p and y v(tp, ty) v( p, y ) . Let t 1 / y v( p / y,1) v( p, y ) v( ~ p ) v ( p, y ) p ) is called normalized indirect utility function, and depends on normaliz v( ~ price alone. With direct utility function u(x), the normalized indirect utility function is defined as: v( ~ p ) max u ( x) s.t. ~ p.x 1 x Thetheorem 4.3.4 Duality Between Direct and Indirect Utility Let v( p, y ) be any indirect utility function, and form the normalized indirect utility p ) . Then the implied direct utility function is given by the following minimum function v ( ~ value function: u( x) min v( ~ p) s.t. ~ p.x 1 . ~ p Proof: Theorem 4.3.5 (Hotelling, Wold) Duality and the System of Inverse Demands Let u(x) be the consumer’s direct utility function. Then the inverse demand function for good i is given by: ~ pi ( x) u ( x) / xi n x j 1 j (u ( x) / x j ) Proof: p) From the normalized indirect utility function, v ( ~ Geoffrey A. Jehle – Advanced Microeconomic Theory 15 Chu Thanh Duc – MDE10 u( x) min v( ~ p) s.t. ~ p.x 1 ~ p The associated Lagrangian is: L v( ~ p ) [1 ~ p.x] u ( x) L Applying the Envelope Theorem: x j x j * ( x). ~ p j * ( x) (1) ~ p*( x ), *( x ) Multiplying both side by x j and summing from j 1,..., n gives: n u ( x) xj * ( x ) ~ p j * ( x).x j * ( x) (2) x j j 1 j 1 n From (1) and (2), we have: ~ p j * ( x) u ( x) / x j n x (u( x) / x ) j 1 j j 4.4 Uncertainty Certainty: the consumer knows the prices of all commodities and knows that any feasible consumption bundles can be obtained with certainty. Many eoconomic decisions contain some element of uncertainty: future income, future prices… 4.4.1 Preferences Beforem, the consumer was assumed to have a preference ordering over all consumption bundles x in the consumption set X. Implicit in our statement that “bundle xi is preferred to bundle xj” was the assumption the individual chooes between xi with certainty and xj with certainty. Instead of ordering consumption bundles, the individual is assumed to have a preference ordering over gambles. Let’s first define an outcome as a result of some uncertain situation. For example, outcomes of betting are win and loss. Geoffrey A. Jehle – Advanced Microeconomic Theory 16 Chu Thanh Duc – MDE10 A {a1 , a2 ,..., an } : the set of all mutually exclusive ultimate outcomes that an individual could endup with. a i could be an m-dimensional commodity vector, or alternatively, a scalar. The way we characterize A depend upon the nature of the particular problem we wish to address. Gambles: G [ p1oa1p2 oa2 pn an ] denotes the entire gamble involving a1 with probability p1 and a 2 with probability p 2 and so forth. Where: p1oa1 denotes the outcome a1 with its probability of occurrence p1 . denotes “and” – the logical symbol. Definition 4.4.1 The Space of Gambles, g(A) The space g(A) is the set of all possible gamble which can be constructed from the outcome set A by varying the probabilities 0 pi 1 of each ai A while ensuring that n i 1 pi 1 . Since each a i is the special gamble in g(A) where pi 1 and p j 0 , i j , the set of all ultimate outcomes A is itself a subset of g(A). The problem of choice under uncertainty can be veiwed as a choice between alternative gambles in g(A). We can then define a binary relation on g(A). Where the symbol stands for the statement “is atleast as well as”. We again suppose that these preferences obey certain rules which we’ll lay down in the form of axiom, called the “axioms of choice under uncertainty”. Axiom G1: (Completeness) For any distinct gambles G1 and G2 in g(A), either G1 G2 or G2 G1. Axiom G2: (Reflexivity) For any gamble G g ( A), G G . Axiom G3: (Transitivity) For any three gambles G1, G2, and G3 in g(A), if G1 G2 and G2 G3 , then G1 G3 . With the addition of Axiom G3, gives a complete ordering of gambles. One important consequence of this Axiom is that there must exist a best and a worst outcome in A. Note that best and worst outcomes need not be unique. Geoffrey A. Jehle – Advanced Microeconomic Theory 17 Chu Thanh Duc – MDE10 G [ p1oa1p2 oa2 pn an ] if all p j 0 and pi 0 ( i j ) G ai By the Axiom 3, there therefore must exist a best and a worst outcome in A. A best outcome aB A satistfies a B a j for all a j A . A worst outcome aW A statisfies a w a j for all a j A . Axiom G4: (Continuity) For any gamble G g ( A) , there exists some probability z, 0 z 1, such that G ~ [ z o a B (1 z ) o aW ] . Indifference probability Best – Worst gamble [ z o aB (1 z ) o aW ] For any gamble G there is some other gamble, involving only the best and the worst outcomes in A, which the agent ranks indifferent to G. Axiom G5. (Monotonicity) For any two best-worst gambles, G1 [ poa B (1 p)oaW ] and G2 [qoa B (1 q)oaW ] , we have G1 G2 if and only if p q . Axiom G5 states that given the choice between any two best – worst gambles with different probabilities attached to the same best outcome, an individual will never prefer the gamble with the lower probability of the best outcome. Together, Axiom G4 and G5 rule out some kinds of behavior which, at first glance, might appear quite reasonable. Example, let A = {“death”, $10, $1000}; aW death , a B $1000 and a B $10 aW . Consider the intermediate gamble G3 = $10. According to axiom G4, there must be some best-worst gamble such that G3 ~ [ z o a B (1 z ) o aW ] and (1-z)>0. If there is no strictly positive probability of death at which you would prefer the gamble [ z o $1000 (1 z ) o death] to $10 with certainty, then the preferences violate the combined implications of Axiom G4 and G5. Axiom G6. (Substitutability) For any outcome ai A and any gamble G j g ( A) , if ai ~ G j , then [ p1oa1 pi oai pn oan ] ~ [ p1oa1 pi oG j pn oan ] Axiom G6 states that if the individual is indifferent between and outcome promised with certainty and some gamble, then he mus also be indifferent between two otherwise identical gambles offering each of these with the same probability. Axiom G7. (Net Probability Rule) Let the gamble Geoffrey A. Jehle – Advanced Microeconomic Theory 18 Chu Thanh Duc – MDE10 Gi [ p1oa1 pi oG j p n oa n ] Where the gamble G j [q1oa1 qi oai q n oa n ] Then Gi ~ [( p1 q1 pi )oa1 pi qi oG j ( p n q n pi )oa n ] 4.4.2 Von Neumann – Morgenstern Utility Whether we can represent those preferences with a continuous real valued function? Say Yes! Axiom G1, G2, G3 and some kind of continuity assumption should be sufficient to ensure the existence of a simple numerical function representing . Instead of asking whether there is a certain kind of function, with a certain specific mathematical property, representing . Let G [ p1oa1 pi oai p n oa n ] be any gamble in g(A), and suppose that the function U : g ( A) R represents the preference by assigning larger numbers to preferred gambles and equal numbers to indifferent gambles. Then, of course, U is a utility function in the usual sence. But if, in addition, the numbers assigned by U to gambles G satisfy: n U (G ) p i U (ai ) , we say that the utility function U possesses the extra, expected i 1 utility property. A Utility Function possesses the Expected Utility Property if and only if the Utility number it assigns to any gamble can be expressed as the Expected Value of the Utility numbers it assigns to the Ultimate outcomes in that gamble. Theorem 4.4.1 Existence of a Von Neumann – Morgenstern (VNM) Utility Function over Gambles Let preferences over gambles, , satisfy Axiom G1 through G7. Then there exists a function U : g ( A) R such that, for all G1 and G2 in g(A), G1 G2 if and only if, U (G1 ) U (G2 ) , and where, moreover, for any gamble G [ p1oa1 pi oai p n oa n ] , n U (G ) p i U (ai ) . i 1 Proof: Geoffrey A. Jehle – Advanced Microeconomic Theory 19 Chu Thanh Duc – MDE10 Let G be any gamble in g(A), where: G [ p1oa1 pi oai pn oan ] (P.1) By the Axiom G1, G2, and G3, there is a complete ordering over g(A), and we can therefore identify a best outcome a B and worst outcome a w in A. By Axiom G4, there exists an “indifference probability”, 0 z i 1 , for each outcome a i which satisfies ai ~ [ z i oaB (1 z i )oaW ] (P.2). It is easy to prove that these indifference probabilities are unique (using Axiom G5). According to Axiom G6, we can substitute from the right-hand side of (P.2) of each a i in (P.1): G ~ p1o[ z1oa B (1 z1 )oaW ] pn o[ z n oa B (1 z n )oaW ] (P.3) Using transitivity and Axiom G7, we obtain n n G ~ pi z i oa B 1 pi z i oaW (P.4) (remind that i 1 i 1 p i 1 ). We now propose a mapping from grambles to the real line. n We let (propose): U (G ) piU (ai ) i 1 (P.5) where: U (ai ) z i , i 1,, n (P.6) Take note that this mapping is indeed a function, since the indifference probabilities from which it is constructed always exist and are unique. We need to show that this function represents . Consider any two gambles in g(A) where: G1 [q1oa1 qn oan ] (P.7) G2 [r1oa1 rn oan ] (P.8) Applying the mapping in (P.5) and (P.6) to the gamble G1, we obtain: n n i 1 i 1 U (G1 ) qiU (ai ) qi zi From the (P.4), we obtain: G1 ~ U (G1 )oaB 1 U (G1 ) oaW . With completely analogous steps, we obtain that: Geoffrey A. Jehle – Advanced Microeconomic Theory 20 Chu Thanh Duc – MDE10 G2 ~ U (G2 )oaB 1 U (G2 )oaW Note that 0 U (G1 ) 1 and 0 U (G2 ) 1 . By the Axiom G5, we have: G1 G2 U (G1 ) U (G2 ) The theorem is proved./. The proceduce to construct U(G). 1. Identify p i such that G [ p1oa1 pi oai pn oan ] . 2. Identify z i such that: ai ~ [ z i oaB (1 z i )oaW ] 3. Let U (ai ) z i 4. U (G) piU (ai ) pi zi Example 4.4.1 Suppose that A = {$10,$4,-$2). We can reasonable suppose that a B = $10, aW = -$2. Suppose we find that: $10 ~ [1o$10 0o($2)] U ($10) 1 (E.1) $4 ~ [.6o$10 .4o($2)] U ($4) .6 (E.2) $2 ~ [0o$10 1o($2)] U ($2) 0 (E.3) Under this mapping, the utility of the best outcome must be (identically) 1, and that of the worst outcome must be (identically) 0. The utility assigned to intermediate outcomes will depend on the individual’s attitude toward taking risks. Consider two gambles: G1 [.2o$4 .8o$10] (E.4) G2 [.07o $2 .03o$4 .9o$10] (E.5) U (G1 ) .2U ($4) .8U ($10) .2 * .6 .8 *1 .92 U (G2 ) .07U ($2) .03U ($4) .9U ($10) .918 G1 G2 Geoffrey A. Jehle – Advanced Microeconomic Theory 21 Chu Thanh Duc – MDE10 We can rank any of the infinite number of gambles that could be constructed from the three outcomes in A./. The VNM mapping does its assignment of numbers to gambles in two distinct stages: 1. First, all gambles in g(A) that offer one outcome with certainty are assigned utility numbers that reflect th agent’s ordering of those alternatives with certainty. 2. Then, all other gambles in g(A) are assigned utility numbers via the expected utility calculation. The VNM utility numbers assigned to ultimate outcomes must not only properly reflect the agent’s ranking of those particular outcomes relative to each other, they must also be capable of properly reflectig the agent’s ranking of gambles comprised of them through the (special) expected utility calculation. It should not, therefore, be terribly suprising that we are less free to transform VNM utility functions if the ranking of every gamble is to be preserved. Theorem 4.4.2 VNM Utility Functions are Unique Up to Positive Affine Tranformations Let satisfy Axiom G1 through G7, and suppose that the VNM utility function U(G) represents . Then the VNM utility function, V(G), represents those same preferences if, and only if, V (G ) U (G ) , for some arbitrary scalar and some scalar >0. Proof: Sufficient condition is obvious Necessary condition: We need to prove that if V(G) is another utility function V (G ) U (G ) with , 0 . Let A {a1 ,, an } and G { p1oa1 pn oan } . By the proof of Theorem 4.4.1 that if the VNM utility function U(G) represents , it possesses the expected utility property and so, for any G G ( A) , we can write: n n i 1 i 1 U (G ) piU (ai ) pi z i where z i satisfies: ai ~ [ z i oa B (1 z i )oaW ] . Suppose that V(G) is another VNM utility function which represents , we must have the following: Geoffrey A. Jehle – Advanced Microeconomic Theory 22 Chu Thanh Duc – MDE10 V (G ) piV (ai ) pi [ z i oa B (1 z i )oaW ] pi z i V (a B ) 1 pi z i V (aW ) n n i 1 i 1 U (G ).V (a B ) (1 U (G )).V (aW ) V (aW ) V (a B ) V (aW )U (G ) For any outcome set A and VNM utility function V, the numbers V (a B ) and V (aW ) are constants, with V (a B ) > V (aW ) . Setting V (aW ) and V (a B ) V (aW ) 0 , so the theorem is proved./. Theorem 4.4.2 tells us that VNM utility functions are not completely unique, nor are they entirely ordinal. We can still find an infinite number of them that will rank gambles in precisely the same order and also possess the expected utility property. However, unlike ordinary utility functions, here we must limit the posivite transformation of VNM utility function under the form of V (G ) U (G ) with , 0 . Yet the less than complete ordinality of the VNM utility function must not tempt us into attaching undue significance to the absolute level of a gamble’s utility, or to the differene in utility between one gamble and another. With what little weve required of the agent’s binary comparisons between gambles in the underlying preference ordering, we still cannot use VNM utility functions for interpersonal comparisons of well – being, nor can we measure the “intensity” with which one gamble is perferred to another. 4.4.3 Risk Aversion The VNM utility function we created reflected some desire to avoid risk. We shall assume that the VNM utility function is both increasing and differentiable over the appropriate domain of wealth concerned. We let the possible wealth outcomes be denoted A {w1 ,, wn } . n The expected value of G: E[G ] pi wi . i 1 Now suppose that the agent is given a choice between accepting the gamble G on the one hand, or receiving with certainty the expected value of G on the other. We can evaluate these two alternative as follows: The utility of the gamble: U (G) piU (wi ) and Geoffrey A. Jehle – Advanced Microeconomic Theory 23 Chu Thanh Duc – MDE10 The utility of the gamble’s expected value: U ( E[G]) U ( pi wi ) . When someone would rather receive the expected value of a gamble with certainty than face the risk inherent in the gamble itself, we say they are risk averse. Definition 4.4.2 Risk Aversion, Risk Neutrality, and Risk Loving Let A {w1 ,, wn } . Then at G g ( A) , an individual is said to be locally: 1. Risk averse whenever U ( E[G ]) U (G ) 2. Risk neutral whenever U ( E[G ]) U (G ) 3. Risk loving whenever U ( E[G ]) U (G ) If these relationships hold for all gambles G g ( A) , these definitions apply globally. Each of these attitudes toward risk is equivalent to a particular property of the VNM utility function. We will be asked to show that the agent is risk averse, risk neutral, or risk loving over some subset of gambles if, and only if, their VNM utility function is strictly concave, linear, or strictly convex, respectively, over the appropriate domain of wealth. Consider a simple gamble involving two outcomes: G [ p ow1 (1 p)ow2 ] and E[G] pw1 (1 p)w2 . The individual is offered a choice between receiving an amount of wealth equal to E[G] pw1 (1 p)w2 with certainty, or receiving the gamble G itself. We can compare the alternatives as follows: U (G) p.U (w1 ) (1 p)U (w2 ) and U ( E[G]) U ( pw1 (1 p)w2 ) . U(w2) U(E[G]) U(G) U(w1) P w1 CE E(G) w2 Risk aversion and strict concavity of the VNM utility function Geoffrey A. Jehle – Advanced Microeconomic Theory 24 Chu Thanh Duc – MDE10 We can see that, strict concavity of the VNM utility function, U(E[G])>U(G), so the individual is risk averse. Certainty Equivalent (CE): Amount of wealth we could offer with certainty that would make him indifferent between accepting that wealth with certainty and facing the gamble G. When a person is risk CE < E(G). A risk – averse person will “pay” some positive amount of wealth in order to avoid the gamble’s inherent risk. This willingness to pay to avoid risk is measured by the Risk Premium. Definition 4.4.3 Certainty Equivalent and Risk Premium The Certainty Equivalent of any gamble G is an amount of wealth, CE, offered with certainty, such that U (G ) U (CE ) . The Risk Premium is an amount of wealth, P, such that U (G ) U ( E[G ] P) . Clearly, the two are related, and P E[G ] CE . Example 4.4.2 Suppose that U ( w) log( w) U is strictly concave in wealth, the individual is risk averse. Let G offers 50-50 odds of winning or losing some amount of wealth, h, so that: G [.5o( w h).5o( w h)] where w is current wealth, and E[G] = w. Log(CE) = (1/2)log(w+h) + (1/2)log(w-h) = log (w2 – h2)1/2. Thus CE = (w2 – h2)1/2 < E[G] and P = w - (w2 – h2)1/2 > 0. Risk aversion and concavity of the VNM utility function in wealth are equivalent. The sign of the second derivative U’’(w) does tell us whether the individual is risk averse, risk loving, or risk neutral, its size is entirely arbitrary. The size of U’’(w) depends on the positive affine transformations of U(w) and the units in which the outcome is measured. Arrow (1965) and Pratt (1964) have proposed a measure a risk aversion which is based on the second derivative, but which avoids these non-uniqueness problems. Geoffrey A. Jehle – Advanced Microeconomic Theory 25 Chu Thanh Duc – MDE10 Definition 4.4.4 The Arrow – Pratt Measure of Absolute Risk Aversion The Arrow – Pratt measure of absolute risk aversion is given by Ra ( w) U ' ' ( w) U ' ( w) Ra (w) is positive, negative, or zero as the agent is Risk Averse, Risk Loving, or Risk Neutral repectively. Any positive affine transformation of utility will leave the measure unchanged. Changing units of measurement of outcome leaves Ra (w) unaffected Ra (w) is only a local measure of risk aversion, so it need not be the same at every level wealth. Geoffrey A. Jehle – Advanced Microeconomic Theory 26