Division of the Humanities
and Social Sciences
Monetary welfare measurement
KC Border
Fall 2008
Revised Fall 2014
v. 2016.11.09::18.03
One of the goals of consumer demand theory is to be able to measure welfare changes.
The ideal is to measure welfare changes in monetary units, so that they can be added up across
consumers to get an aggregate value. Unfortunately, there are problems with this approach that
were pointed out by Kaldor [15], Scitovsky [23], and culminating in Arrow’s famous impossibility
theorem [1]. Chipman and Moore [2, 3, 4, 5, 6, 7] provide an extensive analysis of the difficulties
involved.
1 Hicks’s Compensating and Equivalent Variations
Hicks [11, 12] assigns monetary values to budget changes. He defines the Compensating Variation [12, p. 126] to be “the loss of income which would just offset a fall in price, leaving the
consumer no better off than before.” That is, starting from a budget (p0 , w), considering a price
change to the new budget (p1 , w) he defines CV(p0 , p1 ) by
v(p1 , w − CV) = v(p0 , w),
where v is the indirect utility function. In terms of the expenditure function, letting υ0 =
v(p0 , w) and υ1 = v(p1 , w), we have w − CV = e(p1 , υ0 ) and w = e(p1 , υ1 ), so
CV(p0 , p1 ) = e(p1 , υ1 ) − e(p1 , υ0 ).
Similarly he defines the Equivalent Variation [12, p. 128] to be “the change in income which, if
it took place at [p0 ] prices, would have the same effect on satisfactions as is produced by the
change in prices from [p0 ] to [p1 ]” as
EV(p0 , p1 ) = e(p0 , υ1 ) − e(p0 , υ0 ).
See Figure 1 for the case where the price of y is held fixed and and the price of x decreases. I
hope it is apparent from the figure that the compensating and equivalent variations need not
be equal.
One difficulty with Hicks’ approach is that the expenditure function is not directly observable. Nevertheless, as we show in sections 3 and 9, there is a way to use ordinary demand
functions to recover these welfare measures.
1
KC Border
Monetary welfare measurement
2
y
e(p0 , υ

1 )











u = υ0
u = υ1
EV(p0 , p1 )













w





CV(p0 , p1 )





e(p1 , υ0 )
x̂(p0 , υ1 )
x∗ (p0 , w)
x∗ (p1 , w)
x̂(p1 , υ0 )
(p0 , w)
(p1 , w)
x
Figure 1. Compensating and equivalent variation for a decrease in px . (The figure is drawn to
scale for the case u(x, y) = x2/3 y 1/3 , w = 1, p0x = 1, p1x = 1/2, p0y = p1y = 1.)
v. 2016.11.09::18.03
KC Border
Monetary welfare measurement
3
2 The quasi-linear case
A utility function u(x1 , . . . , xn , y) is quasi-linear if it is of the form
u(x1 , . . . , xn , y) = y + f (x1 , . . . , xn ),
where f is concave. The indifference curves of a quasi-linear function are vertical shifts of one
another, and so typically intersect the x-axis, so corner solutions with y = 0 are common. For
interior points though, any additional income is spent on good y. In this case the CV and EV
are just vertical shifts of each other and so are equal—see Figure 2. But if corner solutions are
involved, this is no longer true—see Figure 3 for a particularly nasty example.
y
e(p0 , υ1 )
EV(p0 , p1 )


w

CV(p0 , p1 )
x̂(p0 , υ1 )

e(p1 , υ0 )
x∗ (p0 , w)
x∗ (p1 , w)
x̂(p1 , υ0 )
(p0 , w)
(p1 , w)
x
Figure 2. A nice quasi-linear case where CV = EV. (Here u(x, y) = y + (1/3) ln x, w = 1,
p0x = 1, p1x = 1/2, p0y = p1y = 1.)
3 Money metric utility
A more general approach to assigning monetary values to budgets is the following. Define
(
)
µ(p∗ ; p, w) = e p∗ , v(p, w) ,
(⋆)
where p∗ is an arbitrary price vector. Since e is strictly increasing in v, this is an indirect
utility or welfare measure. That is,
v(p, w) ⩾ v(p′ , w′ ) if and only if
µ(p∗ ; p, w) ⩾ µ(p∗ ; p′ , w′ ),
v. 2016.11.09::18.03
KC Border
Monetary welfare measurement
4
y
e(p0 , υ

1)
EV(p0 , p1 )










































w
CV(p0 , p1 )
















e(p1 , υ0 )
x̂(p0 , υ1 )
(p0 , w)
(p1 , w)
x∗ (p0 , w) = x̂(p1 , υ0 )
x∗ (p1 , w)
x
Figure 3. A nasty quasi-linear case where CV ̸= EV. (Here u(x, y) = y + 0.6 ln x, w = 0.4,
p0x = 1, p1x = 1/2, p0y = p1y = 1. It may not look like it, but the red indifference curve is a
vertical shift of the blue one.)
v. 2016.11.09::18.03
KC Border
Monetary welfare measurement
5
but the units of µ are in dollars (or euros, or whatever currency). The money values depend on
the choice of p∗ . This function is variously called a money metric (indirect) utility or an
income-compensation function. This approach is taken by Varian [24] and Chipman and
Moore [5, 7], who adapt ideas from McKenzie [18]1 and Hurwicz and Uzawa [13].2
4 Compensating and equivalent variation in terms of money
metrics
Thus we may define the compensating and equivalent variations of arbitrary budget changes by
CV(p0 , w0 ; p1 , w1 ) = e(p1 , υ1 ) − e(p1 , υ0 ) = µ(p1 ; p1 , w1 ) − µ(p1 ; p0 , w0 )
(1)
EV(p , w ; p , w ) = e(p , υ1 ) − e(p , υ0 ) = µ(p ; p , w ) − µ(p ; p , w ).
(2)
0
0
1
1
0
0
0
1
1
0
0
0
5 A single price change
Now consider a decrease in the price of only good i. That is,
p1i < p0i ,
p1j = p0j = p̄j for j ̸= i
and
w0 = w1 = w̄.
Since the price of good i decreases, the new budget set includes the old one, so welfare will not
decrease. To make the analysis non-vacuous, let us assume that x∗i (p1 , w̄) > 0.
By definition,
CV(p0 , p1 ) = µ(p1 ; p1 , w1 ) −µ(p1 ; p0 , w0 )
|
{z
}
w̄
EV(p0 , p1 ) = µ(p0 ; p1 , w1 ) − µ(p0 ; p0 , w0 )
|
{z
}
w̄
We now use the following trick (which applies whenever w0 = w1 = w̄):
µ(p0 ; p0 , w0 ) = w0 = w̄ = w1 = µ(p1 ; p1 , w1 ).
1
McKenzie defines an expenditure function in terms of a preference relation ≽ by µ(p, x0 ) = min{p·x : x ≽ x0 }.
This enables him to dispense with a the utility function.
2
Varian attributes the money-metric function µ to Samuelson [22], but this seems a stretch to me. The closest
Samuelson seems to come is found on page 1262 and again on page 1273, where he uses McKenzie’s expenditure
function. The discussion of how this can be used as an indirect utility is confusing at best.
v. 2016.11.09::18.03
KC Border
Monetary welfare measurement
6
This enables us to rewrite the values as
CV = µ(p0 ; p0 , w0 ) −µ(p1 ; p0 , w0 )
|
{z
}
w̄
EV = µ(p0 ; p1 , w1 ) − µ(p1 ; p1 , w1 )
|
{z
}
w̄
Or in terms of the expenditure function, we have:
CV = e(p0 , υ0 ) − e(p1 , υ0 ).
EV = e(p0 , υ1 ) − e(p1 , υ1 )
Compare these expression to the definitions (2) and (1). In terms of the expenditure function,
definition (2) is equivalent to EV(p0 , w0 ; p1 , w1 ) = e(p0 , υ1 ) − e(p0 , υ0 ), which by the trick is
equal to e(p0 , υ1 )−e(p1 , υ1 ). In my opinion, using the trick obscures the true comparison, which
is of the budgets (p0 , w0 ) and (p1 , w1 ) using the money metric defined by p0 . So why did Hicks
rewrite things this way? So he could use the Fundamental Theorem of Calculus and the fact
that ∂e/∂pi = x̂i to get
EV(p , p ) = e(p , υ1 ) − e(p , υ1 ) =
0
1
0
∫
1
CV(p0 , p1 ) = e(p0 , υ0 ) − e(p1 , υ0 ) =
p0i
p1i
∫
p0i
p1i
x̂i (p, p̄−i , υ1 ) dp
(2′ )
x̂i (p, p̄−i , υ0 ) dp,
(1′ )
where the notation p, p̄−i refers to the price vector (p̄1 , . . . , p̄i−1 , p, p̄i+1 , . . . , p̄n ).
That is, for a change in the price of good i, the equivalent and the
compensating variation are the areas under the Hicksian compensated demand curves for good i corresponding to utility levels υ1
and υ0 respectively.
Also since p0i > p1i , the integrals above are positive if x̂i is positive.
Assume now that good i is not inferior, that is, assume
∂x∗i
>0
∂w
for all (p, w). Recall the Slutsky equation,
(
)
∂ x̂i p, v(p, w)
∂x∗i (p, w)
∂x∗ (p, w)
=
+ x∗j (p, w) i
.
∂pj
∂pj
∂w
Under the assumption that good i is normal and that x∗j > 0, this implies that
(
)
∂ x̂i p, v(p, w)
∂x∗i (p, w)
>
.
∂pj
∂pj
v. 2016.11.09::18.03
(S)
KC Border
Monetary welfare measurement
7
We know that the Hicksian compensated demands for good i are downward-sloping as a function
of pi (other prices held constant), that is, ∂ x̂i /∂pi < 0, so we have for j = i
(
)
∂ x̂i p, v(p, w)
∂x∗i (p, w)
0>
>
.
∂pi
∂pi
(3)
Now by the equivalence of expenditure minimization and utility maximization we know that
x∗ (p1 , w̄) = x̂(p1 , υ1 ) and
x∗ (p0 , w̄) = x̂(p0 , υ0 ).
(4)
Then (3) tells us that
0>
∂ x̂i (p1 , υ1 )
∂x∗i (p1 , w̄)
>
∂pi
∂pi
and 0 >
∂ x̂i (p0 , υ0 )
∂x∗i (p0 , w̄)
>
∂pi
∂pi
That is, as a function of pi . the ordinary demand x∗i is steeper (more negatively sloped) than
the Hicksian demand x̂i , where it crosses the Hicksian demand. Since x∗i is downward sloping
and p1i < p0i it must be the case that the Hicksian demand x̂i (·, υ1 ) lies above the Hicksian
demand x̂i (·, υ0 ). See Figure 4. Thus, for a price decrease, if good i is not inferior, then
EV > CS > CV > 0.
The inequalities are reversed for a price increase. Also note that if
demand curves coincide.
∂x∗i
∂w
= 0, then the three
6 “Deadweight loss”
Consider a simple problem where the good 1 is subjected to an ad rem tax of t per unit, but
income and other prices remain unchanged. The original price vector is p0 and the new one is
p1 = (p01 + t, p02 , . . . , p0n ). Clearly the consumer is worse off under the price vector p1 . But how
much worse off?
We shall compare the tax revenue
T = tx∗1 (p1 , w)
(5)
to the welfare cost measured by the money metric utility. Specifically, we shall compare tax
revenue to the equivalent variation of the price change. The equivalent variation is negative,
since the consumer is worse off. We shall show that
− EV > T.
That is, the dollar magnitude of the welfare loss as measured by the equivalent variation exceeds
the tax revenue raised. This excess of the welfare loss over the tax revenue is often referred to
as the deadweight loss3 from ad rem taxation.
3
I don’t know why the term “deadweight” is used. Musgrave [19] uses the term “excess burden” in 1959,
which dates back at least to Joseph [14] in 1939, who claims the concept was known to Marshall [16, 8th edition]
in 1890. Harberger [10] uses the term “deadweight loss” in 1964, and claims the analysis of the concept goes
back at least to Dupuit [9] in 1844.
v. 2016.11.09::18.03
KC Border
Monetary welfare measurement
8
xi
x∗i (pi ; p̄−i , w̄)
CS
EV
x̂i (pi ; p̄−i , υ1 )
CV
p1i
x̂i (pi ; p̄−i , υ0 )
p0i
pi
Figure 4. Illustration of a single price decrease. (Graphs are for a Cobb–Douglas utility.)
N.B. The horizontal axis is the price axis and the vertical axis is quantity axis.
The equivalent variation is the area under the Hicksian demand curve for utility level υ0 .
The compensating variation is the area under the Hicksian demand curve for utility level υ1 .
The consumer’s surplus is the area under the ordinary demand curve.
v. 2016.11.09::18.03
KC Border
Monetary welfare measurement
9
Let’s abuse the notation slightly to get rid of some of the visual noise by setting
x(p) = x̂1 (p, p02 , . . . , p0n , υ1 ),
where p is just a scalar that represents p1 . Then
− EV −T =
∫
p01
∫
=
p01 +t
p01
∫
=
p01 +t
p01
∫
=
p01 +t
p01 +t
p01
x(p) dp − T
by (2′ )
x(p) dp − tx∗1 (p1 , w)
by (5)
x(p) dp − tx̂1 (p1 , υ1 )
x(p) dp −
∫
p01 +t
p01
x̂1 (p1 , υ1 ) dp.
The last equality come from integrating the constant x̂1 (p01 + t, p0−1 , υ1 ) over an interval of
length t. But Hicksian compensated demands are downward sloping, so for p01 ⩽ p ⩽ p11 = p01 +t,
we have
x(p) = x̂1 (p, p02 , . . . , p0n , υ1 ) > x̂1 (p11 , p12 , . . . , p1n , υ1 ),
so the last expression is > 0.
Therefore a lump-sum tax leaves the consumer better off than an ad
rem tax that raises the same revenue.
The amazing thing is not so much that the ad rem tax is inferior to
the lump-sum tax, but that some taxes are worse than others at all,
even when they collect the same amount of revenue! This would not
be apparent without our theoretical apparatus.
7 Revealed preference and lump-sum taxation
Recall that x is revealed preferred to y if there is some budget containing both x and y and
x is chosen. If the choice function is generated by utility maximization, then if x is revealed
preferred to y, we must have u(x) ⩾ u(y).
So consider an ad rem tax t on good 1 versus a lump-sum tax, as above. Assume both taxes
raise the same revenue T . The ad rem tax leads to the budget (p1 , w), and the lump-sum to
the budget (p0 , w − T ), where
p11 = p01 + t and
p1j = p0j , j = 2, . . . , n.
Let x1 be demanded under the ad rem tax and x0 be demanded under the lump-sum tax. Then
x0 is revealed preferred to x1 :
w ⩾ p1 · x1 = p0 · x1 + tx11 = p0 · x1 + T,
v. 2016.11.09::18.03
KC Border
Monetary welfare measurement
10
so
p0 · x1 ⩽ w − T,
which says that x1 is in the budget (p0 , w − T ), from which x0 is chosen. Thus
u(x0 ) ⩾ u(x1 ),
so the lump-sum tax is at least as good as the ad rem tax.
This argument is a lot simpler than the argument above, but we don’t get a dollar value
of the difference. Of course the previous argument gave us two or three different dollar values,
depending on how we chose p∗ for the money metric.
8 Money metrics and recovering utility from demand: A little
motivation
It is possible to solve differential equations to recover a utility function from a demand function.
The general approach may be found in Samuelson [20, 21], but the following discussion is based
on Hurwicz and Uzawa [13].
Consider the demand function
x∗ : Rn++ × R++ → Rn+
derived by maximizing a locally nonsatiated utility function u. Let v be the indirect utility,
that is,
(
)
v(p, w) = u x∗ (p, w) .
Since u is locally nonsatiated, the indirect utility function v is strictly increasing in w. The
Hicksian expenditure function e is defined by
e(p, υ) = min{p · x : u(x) ⩾ υ}.
and the income compensation function µ is defined by
(
)
µ(p; p0 , w0 ) = e p, v(p0 , w0 ) ,
Set υ 0 = v(p0 , w0 ). From the Envelope Theorem we know that
(
)
∂e(p, υ 0 )
= x̂i (p, υ 0 ) = x∗i p, e(p, υ 0 ) .
∂pi
Suppressing υ0 , this becomes a total differential equation
(
)
e′ (p) = x∗ p, e(p) .
What does it mean to solve such an equation, and what happened to υ 0 ?
v. 2016.11.09::18.03
(6)
KC Border
Monetary welfare measurement
11
An aside on solutions of differential equations
You may recall from your calculus classes that, in general, differential equations have many
solutions, often indexed by “constants of integration.” For instance, take the simplest differential equation,
y′ = a
for some constant a. The general form of the solution is
y(x) = ax + C,
where C is an arbitrary constant of integration. What this means is that the differential
equation y ′ = a has infinitely many solutions, one for each value of C. The parameter υ in
our problem can be likened to a constant of integration.
You should also recall that we rarely specify C directly as a condition of the problem,
since we don’t know the function y in advance. Instead we usually specify an initial
condition (x0 , y 0 ). That is, we specify that
y(x0 ) = y 0 .
In this simple case, the way to translate an initial condition into a constant of integration
is to solve the equation
y 0 = ax0 + C =⇒ C = y 0 − ax0 ,
and rewrite the solution as
y(x) = ax + (y 0 − ax0 ) = y 0 + a(x − x0 ).
In order to make it really explicit that the solution depends on the initial conditions, differential equations texts may go so far as to write the solution as
y(x; x0 , y 0 ) = y 0 + a(x − x0 ).
In our differential equation (6), an initial condition corresponding to the “constant of integration” υ is a pair (p0 , w0 ) satisfying
e(p0 , υ) = w0 .
From the equivalence of expenditure minimization an utility maximization under a budget
constraint, this gives us the relation
(
)
υ = v(p0 , w0 ) = u x∗ (p0 , w0 ) .
Following Hurwicz and Uzawa [13], define the income compensation function in terms
of the Hicksian expenditure function e via
(
)
µ(p; p0 , w0 ) = e p, v(p0 , w0 ) .
Observe that
µ(p0 ; p0 , w0 ) = w0
and
(
)
(
)
∂e(p, υ 0 )
∂µ(p; p0 , w0 )
=
= x̂i (p, υ 0 ) = x∗i p, e(p, υ 0 ) = x∗i p, µ(p; p0 , w0 ) .
∂pi
∂pi
v. 2016.11.09::18.03
KC Border
Monetary welfare measurement
12
In other words, the function e defined by
e(p) = µ(p; p0 , w0 )
solves the differential equation
(
)
e′ (p) = x∗ p, e(p) .
subject to the initial condition
e(p0 ) = w0 .
We are now going to turn the income compensation function around and treat (p0 , w0 ) as the
variable of interest. Fix a price, any price, p∗ ∈ Rn++ and define the function v̂ : Rn++ × R++ →
R by
(
)
v̂(p, w) = µ(p∗ ; p, w) = e p∗ , v(p, w) .
The function v̂ is another indirect utility. That is,
v̂(p, w) ⩾ v̂(p′ , w′ ) ⇐⇒ v(p, w) ⩾ v(p′ , w′ ).
We can use w to find a utility U , at least on the range of x∗ by
U (x) = µ(p∗ ; p, w) where x = x∗ (p, w).
9 Recovering utility from demand: The plan
The discussion above leads us to the following approach. Given a demand function x∗ :
1. Somehow solve the differential equation
(
)
∂µ(p)
= x∗i p, µ(p) .
∂pi
Write the solution explicitly in terms of the intial condition µ(p0 ) = w0 as µ(p; p0 , w0 ).
2. Use the function µ to define an indirect utility function v̂ by
v̂(p, w) = µ(p∗ ; p, w).
3. Invert the demand function to give (p, w) as a function of x∗ .
4. Define the utility on the range of x∗ by
U (x) = µ(p∗ ; p, w)
where x = x∗ (p, w).
This is easier said than done, and there remain a few questions. For instance, how do we know
that the differential equation has a solution? If a solution exists, how do we know that the
“utility” U so derived generates the demand function x∗ ? We shall address these questions
presently, but I find it helps to look at some examples first.
v. 2016.11.09::18.03
KC Border
10
Monetary welfare measurement
13
Examples
In order to draw pictures, I will consider two goods x and y. By homogeneity of x∗ , I may take
good y as numéraire and fix py = 1, so the price of x will simply be denoted p.
10.1
Deriving the income compensation function from a utility
For the Cobb–Douglas utility function
u(x, y) = xα y β
where α + β = 1, the demand functions are
x∗ (p, w) =
αw
,
p
y ∗ (p, w) = βw.
The indirect utility is thus
v(p, w) = wβ
β
The expenditure function is
e(p, υ) = υβ
−β
( )α
α
p
.
( )α
p
α
.
Now pick (p0 , w0 ) and define
(
)
µ(p; p0 , w0 ) = e p; v(p0 , w0 )
(
( )α )
( )α
α
p
0 β
−β
= w β
β
0
p
α
( )α
p
= w0 0
.
p
Evaluating this at p = p0 we have
µ(p0 ; p0 , w0 ) = w0 .
That is, the point (p0 , w0 ) lies on the graph of µ(·; p0 , w0 ). Figure 5 shows the graph of this
function for different values of (p0 , w0 ). For each fixed (p0 , w0 ), the function µ(p) = µ(p; p0 , w0 )
satisfies the (ordinary) differential equation
[
]
(
)
dµ
αµ(p)
= α w0 (p0 )−α pα−1 =
= x∗ p, µ(p) .
dp
p
Note that homogeneity and budget exhaustion have allowed us to reduce the dimensionality
by 1. We have n − 1 prices, as we have chosen a numéraire, and the demand for the nth good
∑
∗
is gotten from x∗n = w − n−1
i=1 pi xi .
v. 2016.11.09::18.03
KC Border
Monetary welfare measurement
14
µ, w
3
2
1
p
1
2
3
4
Figure 5. Graph of µ(p; p0 ; w0 ) for Cobb–Douglas α = 2/5 utility and various values of (p0 , w0 ).
10.2
Examples of recovering utility from demand
Let n = 2, and set p2 = 1, so that there is effectively only one price p, and only one differential
equation (for x1 )
(
)
µ′ (p) = x p, µ(p) .
1 Example In this example
x(p, w) =
αw
.
p
(This x is the demand for x1 . From the budget constraint we can infer x2 = (1 − α)w.)
The corresponding differential equation is
µ′ =
αµ
p
or
µ′
α
= .
µ
p
(For those of you more comfortable with y-x notation, this is y ′ = αy/x.) Integrate both sides
of the second form to get
ln µ = α ln p + C
so exponentiating each side gives
µ(p) = Kpα
where K = exp(C) is a constant of integration. Given the initial condition (p0 , w0 ), we must
have
w0
w0 = K(p0 )α ,
so K = 0 α ,
(p )
or
w0
µ(p; p0 , w0 ) = 0 α pα .
(p )
v. 2016.11.09::18.03
KC Border
Monetary welfare measurement
15
For convenience set p∗ = 1, to get
v̂(p, w) = µ(p∗ ; p, w) =
w
.
pα
To recover the utility u, we need to invert the demand function, that is, we need to know
for what budget (p, w) is (x1 , x2 ) chosen. The demand function is
x1 =
αw
,
p
x2 = (1 − α)w,
so solving for w and p, we have
w=
x1 =
x2
α 1−α
p
x2
1−α
=⇒
p=
α x2
.
1 − α x1
Thus
u(x1 , x2 ) = v̂(p, w)
(
)
α x2 x2
= v̂
,
1 − α x1 1 − α
=
=
x2
1−α
(
)α
α x2
1−α x1
(
)1−α ( )α
x2
x1
1−α
= cxα1 x1−α
,
2
α
where c = (1 − α)1−α αα , which is a Cobb–Douglas utility.
□
2 Example In this example we find a utility that generates a linear demand for x. That is,
x(p, w) = β − αp.
(Note the lack of w.) The differential equation is
µ′ = β − αp.
This differential equation is easy to solve:
µ(p) = βp −
α 2
p +C
2
2
For initial condition (p0 , w0 ) we must choose C = w0 − βp0 + α2 p0 , so the solution becomes
µ(p; p0 , w0 ) = βp −
α 2
α 2
p + w0 − βp0 + p0 .
2
2
So choosing p∗ = 0 (not really allowed, but it works in this case), we have
v̂(p, w) = µ(p∗ ; p, w) = w − βp +
v. 2016.11.09::18.03
α 2
p .
2
KC Border
Monetary welfare measurement
16
Given (x, y) (let’s use this rather than (x1 , x2 )), we need to find the (p, w) at which it is
chosen. We know
x = β − αp, y = w − px = w − βp + αp2 ,
so
β−x
,
p=
α
(
β−x
β−x
w = y + βp − αp = y + β
−α
α
α
)2
2
.
Therefore
(
u(x, y) = v̂(p, w) = w
(
β−x
β−x
β−x
,y + β
−α
α
α
α
(
= y+β
|
β−x
β−x
−α
α
α
)2
{z
w
= y−
−β
}
)2 )
β−x α
+
α } 2
| {z
p
(
|
β−x
α
{z
)2
}
p2
(β − x)2
.
2α
Note that the utility is decreasing in x for x > β. Representative indifference curves are shown
in Figure 6. The demand curve specified implies that x and y will be negative for some values
of p and w, so we can’t expect that this is a complete specification. I’ll leave it to you to figure
out when this makes sense.
□
5
4
3
2
1
1
2
3
4
5
6
7
8
9
10 11 12
Figure 6. Indifference curves for Example 2 (linear demand) with β = 10, α = 5.
11
A general integrability theorem
Hurwicz and Uzawa [13] prove the following theorem, presented here without proof.
3 Hurwicz–Uzawa Integrability Theorem Let ξ : Rn++ × R+ → Rn+ . Assume
(B) The budget exhaustion condition
p · ξ(p, w) = w
is satisfied for every (p, w) ∈ Rn++ × R+ .
(D) Each component function ξi is differentiable everywhere on Rn++ × R+ .
v. 2016.11.09::18.03
KC Border
Monetary welfare measurement
17
(S) The Slutsky matrix is symmetric, that is, for every (p, w) ∈ Rn++ × R+ ,
Si,j (p, w) = Sj,i (p, w)
i, j = 1, . . . , n.
(NSD) The Slutsky matrix is negative semidefinite, that is, for every (p, w) ∈ Rn++ × R+ , and
every v ∈ Rn ,
n ∑
n
∑
Si,j (p, w)vi vj ⩽ 0.
i=1 j=1
(IB) The function ξ satisfies the following boundedness condition on the partial derivative with
respect to income. For every 0 ≪ a ≪ ā ∈ Rn++ , there exists a (finite) real number Ma,ā
such that for all w ⩾ 0
a ≦ p ≦ ā
=⇒
∂ξi (p, w) ⩽ Ma,ā
∂w
i = 1, . . . , n.
Let X denote the range of ξ,
X = {ξ(p, w) ∈ Rn+ : (p, w) ∈ Rn++ × R+ }.
Then there exists an upper semicontinuous monotonic utility function u : X → R on the range
X such that for each (p, w) ∈ Rn++ × R+ ,
ξ(p, w) is the unique maximizer of u over the budget set {x ∈ X : p · x ⩽ w}.
Moreover u has the following property (which reduces to strict quasiconcavity if X is itself
convex): For each x ∈ X, there exists a p ∈ Rn++ such that if y ̸= x and u(y) ⩾ u(x), then
p · y > p · x.
I have more extensive notes on this topic, including most of a sketch of the proof here.
References
[1] K. J. Arrow. 1950. A difficulty in the concept of social welfare. Journal of Political Economy
58(4):328–346.
http://www.jstor.org/stable/1828886
[2] J. S. Chipman and J. C. Moore. 1971. The compensation principle in welfare economics.
In A. M. Zarley, ed., Papers in Quantitative Economics, volume 2, pages 1–77. Lawrence,
Kansas: University of Kansas Press.
[3]
. 1973. Aggregate demand, real national income, and the compensation principle.
International Economic Review 14(1):153–181.
http://www.jstor.org/stable/2526050
[4]
. 1978. The new welfare economics 1939–1974. International Economic Review
19(3):547–584.
http://www.jstor.org/stable/2526326
v. 2016.11.09::18.03
KC Border
Monetary welfare measurement
18
[5]
. 1980. Compensating variation, consumer’s surplus, and welfare. American Economic Review 70(5):933–949.
http://www.jstor.org/stable/1805773
[6]
. 1980. Real national income with homothetic preferences and a fixed distribution
of income. Econometrica 48(2):401–422.
http://www.jstor.org/stable/1911104
[7]
. 1990. Acceptable indicators of welfare change, consumer’s surplus analysis, and
the Gorman polar form. In J. S. Chipman, D. L. McFadden, and M. K. Richter, eds.,
Preferences, Uncertainty, and Optimality: Essays in Honor of Leonid Hurwicz, pages 1–77.
Boulder, Colorado: Westview Press.
[8] P. A. Diamond and D. L. McFadden. 1974. Some uses of the expenditure function in public
finance. Journal of Public Economics 3:3–21.
[9] J. E. J. Dupuit. 1844. De la l’utilité and de sa mesure. Collezione di scritti inediti o rari
di economisti. Turin: La Riforma Soziale. Collected and reprinted by Mario di Bernardi
and Luigi Einaudi, La Riforma Soziale, Turin, 1932.
[10] A. C. Harberger. 1964. The measurement of waste. American Economic Review 54:58–
76.
http://www.jstor.org/stable/1818490
[11] J. R. Hicks. 1936. Value and capital. Oxford, England: Oxford University Press.
[12]
. 1942. Consumers’ surplus and index-numbers. Review of Economic Studies
http://www.jstor.org/stable/2967665
9(2):126–137.
[13] L. Hurwicz and H. Uzawa. 1971. On the integrability of demand functions. In J. S.
Chipman, L. Hurwicz, M. K. Richter, and H. F. Sonnenschein, eds., Preferences, Utility,
and Demand: A Minnesota Symposium, chapter 6, pages 114–148. New York: Harcourt,
Brace, Jovanovich.
[14] M. F. W. Joseph. 1939. The excess burden of indirect taxation. Review of Economic
Studies 6(3):226–231.
http://www.jstor.org/stable/2967649.pdf
[15] N. Kaldor. 1939. Welfare propositions of economics and interpersonal comparisons of
utility. Economic Journal 49(195):549–552.
http://www.jstor.org/stable/2224835
[16] A. Marshall. 1890. Principles of economics. London.
[17] A. Mas-Colell, M. D. Whinston, and J. R. Green. 1995. Microeconomic theory. Oxford:
Oxford University Press.
[18] L. W. McKenzie. 1957. Demand theory without a utility index. Review of Economic
Studies 24(3):185–189.
http://www.jstor.org/stable/2296067
[19] R. A. Musgrave. 1959. The theory of public finance. New York: McGraw–Hill.
[20] P. A. Samuelson. 1947. Foundations of economic analysis. Cambridge, Mass.: Harvard
University Press.
v. 2016.11.09::18.03
KC Border
[21]
385.
[22]
Monetary welfare measurement
19
. 1950. The problem of integrability in utility theory. Economica N.S. 17(68):355–
http://www.jstor.org/stable/2549499
. 1974. Complementarity: An essay on the 40th anniversary of the Hicks–Allen
revolution in demand theory. Journal of Economic Literature 12(4):1255–1289.
[23] T. Scitovsky. 1941. A note on welfare propositions in economics. Review of Economic
Studies 9(1):77–88.
http://www.jstor.org/stable/2967640
[24] H. R. Varian. 1992. Microeconomic analysis, 3d. ed. New York: W. W. Norton & Co.
[25] J. A. Weymark. 1985. Money-metric utility functions. International Economic Review
26(1):219–232.
http://www.jstor.org/stable/2526537
v. 2016.11.09::18.03