Lecture Notes on Elasticity of Substitution

Ted Bergstrom, UCSB Economics 210A

October 26, 2015

Today’s featured guest is “the elasticity of substitution.”

Elasticity of a function of a single variable

Before we meet this guest, let us spend a bit of time with a slightly simpler
notion, the elasticity of a a function of a single variable. Where f is a
differentiable real-valued function of a single variable, we define the elasticity
of f (x) with respect to x (at the point x) to be
xf 0 (x)
η(x) = . (1)
f (x)
Another way of writing the same expression 1 is
df (x)
x df (x) f (x)
η(x) = dx = dx . (2)
f (x) x

From Expression 2, we see that the elasticity of of f (x) with respect to x is

the ratio of the percent change in f (x) to the corresponding percent change
in x.
Measuring the responsiveness of a dependent variable to an independent
variable in percentage terms rather than simply as the derivative of the func-
tion has the attractive feature that this measure is invariant to the units in
which the independent and the dependent variable are measured. For exam-
ple, economists typically express responsiveness of demand for a good to its
price by an elasticity.1 In this case, the percentage change in quantity is the
1
Some economists find it tiresome to talk about negative elasticities and choose to define
the price-elasticity as the absolute value of the percentage responsiveness of quantity to
price.

1
same whether quantity is measured in tons or in ounces and the percentage
change in price is the same whether price is measured in dollars, Euros, or
farthings. Thus the price elasticity is a “unit-free” measure. For similar rea-
sons, engineers measure the stretchability of a material by an “elasticity” of
the length of the material with respect to the force exerted on it.
The elasticity of the function f at a point of x can also be thought of as
the slope of a graph that plots ln x on the horizontal axis and ln f (x) on the
vertical axis. That is, suppose that we make the change of variables u = ln x
and v = ln y and we rewrite the equation y = f (x) as ev = f (eu ). Taking
derivatives of both sides of this equation with respect to u and applying the
chain rule, we have
dv
ev = eu f 0 (eu ) (3)
du
and hence
dv eu f 0 (eu ) xf 0 (x)
= = = η(x), (4)
du ev f (x)
where the second equality in Expression 4 is true because eu = x and ev =
dv
f (x). Thus du is the derivative of ln f (x) with respect to ln x. We sometimes
express this by saying that

d ln f (x)
η(x) = . (5)
d ln x
It is interesting to consider the special case where the elasticity of f (x) with
respect to x is a constant, η that does not dependent on x. In this case,
integrating both sides of Equation 5, we have

ln f (x) = η ln x + a (6)

for some constant a. Exponentiating both sides of Equation 6, we have

f (x) = cxη (7)

where c = ea . Thus we see that f has constant elasticity η if and only if f is

a “power function” of the form 7.
In general, the elasticity of f with respect to x depends on the value of
−bx
x. For example if f (x) = a − bx, then η(x) = a−bx . In this case, as x ranges
from 0 to a/b, η(x) ranges from 0 to −∞.

2
Elasticity of inverse functions
Another useful fact about elasticities is the following. Suppose that the
function f is either strictly increasing or strictly decreasing. Then there is a
well defined inverse function φ, defined so that such that φ(y) = x if and only
if f (x) = y. It turns out that if η(x) is the elasticity of f (x) with respect to
x, then 1/η(x) is the elasticity of φ(y) with respect to y.
Proof. Note that since the function φ is the inverse of f , we must have
φ (f (x)) = x. Using the chain rule to differentiate both sides of this equation
with respect to x, we see that if y = f (x), then φ0 (y)f 0 (x) = 1 and hence
φ0 (y) = 1/f 0 (x). Therefore when y = f (x), the elasticity of φ(y) with respect
to y is
yφ0 (y) f (x)
= .
φ(y) φ(y)f 0 (x)
But since y = f (x), it must be that φ(y) = x, and so we have

yφ0 (y) f (x) 1

= = .
φ(y) xf 0 (x) η(x)

Application to monopolist’s revenue function

One of the most common applications of the notion of elasticity of demand
is to monopoly theory, where a monopolist is selling a good and the quantity
of the good that is demanded is a function D(p) of the monopolist’s price p.
The monopolist’s revenue is R(p) = pD(p). Does the monopolist’s revenue
increase of decrease if he increases his price and how is this related to the
price elasticity? We note that R(p) is increasing (decreasing) in price if
and only if ln R(p) increases (decreases) as the log of price increases. But
ln R(p) = ln p + ln D(p). Then

d ln R(p) d ln D(p)
=1+ = 1 + η(p)
d ln p d ln p
where η(p) is the price elasticity of demand. So revenue is an increasing
function of p if η(p) > −1 and a decreasing function of p if η < −1. In the
former case we say demand is inelastic and in the latter case we say demand
is elastic.

3
Elasticity of substitution
Now we introduce today’s main event–the elasticity of substitution for a func-
tion of two variables. The elasticity of substitution is most often discussed in
the context of production functions, but is also very useful for describing util-
ity functions. A firm uses two inputs (aka factors of production) to produce
a single output. Total output y is given by a concave, twice differentiable
function y = f (x1 , x2 ). Let fi (x1 , x2 ) denote the partial derivative (marginal
product) of f with respect to xi . While the elasticity of a function of a sin-
gle variable measures the percentage response of a dependent variable to a
percentage change in the independent variable, the elasticity of substitution
between two factor inputs measures the percentage response of the relative
marginal products of the two factors to a percentage change in the ratio of
their quantities.
The elasticity of substitution between any two factors can be defined
for any concave production function of several variables. But for our first
crack at the story it is helpful to consider the case where there are just two
inputs and the production function is homogeneous of some degree k > 0.
We also assume that the production function is differentiable and strictly
quasi-concave.
Fact 1. If f (x1 , x2 ) is homogeneous of some degree k and strictly quasi-
concave, then the ratio of the marginal products of the two factors is deter-
mined by the ratio x1 /x2 and f1 (x1 , x2 )/f2 (x1 , x2 ) is a decreasing function of
x1 /x2 .
Proof. If f is homogeneous of degree k, then the partial derivatives of f are
homogeneous of degree k − 1.2 Therefore
x1
 
f1 (x1 , x2 ) = xk−1
2 f 1 , 1
x2
and
x1
 
f2 (x1 , x2 ) = xk−1
2 f 2 , 1 .
x2
It follows that
 
f1 (x1 , x2 ) xk−1 x1
2 f 1 x2 , 1
= k−1  x  (8)
f2 (x1 , x2 ) x2 f2 x12 , 1
2
To prove this, note that if f is homogeneous of degree k, then f (λx) = λk f (x).
Differentiate both sides of this equation with respect to xi and arrange terms to show that
fi (λx) = λk−1 fi (x).

4
  
x1
f1 x
,1
=  2 . (9)
f2 xx12 , 1

Therefore the ratio of marginal products is determined by the ratio x1 /x2 .

Let us define this ratio as
x1 f1 (x1 , x2 )
 
g = .
x2 f2 (x1 , x2 )
Since strict quasi-concavity implies diminishing marginal rate of substitution,
it must be that g is a strictly decreasing function of x1 /x2 .

Since g is strictly decreasing, it must be that the function g has a well-

defined inverse function. Let’s call this inverse function h.
Let prices of the two inputs be given by the vector p = (w1 , w2 ). Suppose
that the firm always chooses factors so as to minimize its costs, conditional
on its output level. Then it must be that at prices p, the firm uses factors in
the ratio x1 /x2 such that
 
x1
f1 x
,1 w1
g(x1 /x2 ) =  2  = (10)
f2 xx12 , 1 w2

or equivalently such that

!
x1 p1 w1
 
= g −1 =h . (11)
x2 p2 w2
Definition 1. Where there are two factors and a homogeneous production
function, the elasticity of substitution σ(w1 /w2 ) measures the responsiveness
of the ratio in which factors are used to the ratio of factor prices. Thus it is
the absolute value of the ratio of the percentage by which a cost-minimizing
firm changes the ratio in which it uses the factors in response to a change in
the ratio of factor prices to he percentage change in the ratio of factor prices.
The elasticity of substitution is just the negative of the elasticity of the
function h with respect to its argument w1 /w2 . That is,
   
w1 0 w1 w1
w1 w2
h w2
d ln hw
σ( ) = − w  =−  2 . (12)
w2 h w12 w1
d ln w2

5
 As we remarked in our earlier discussion, the elasticity of an inverse func-
tion is just the inverse of the elasticity of a function. The function g defined
in Equation 10 is the inverse of the function h defined in Equation 11 and so
where
x1 w1
 
=h
x2 w2
it must be that the elasticity σ(x1 /x2 ) of the function g satisfies the equations
f1 (x1 ,x2 )
1 d ln f2 (x1 ,x2 )
=− (13)
σ(w1 /w2 ) d ln xx12

Constant Elasticity of Substitution

A very interesting special class of production functions is those for which
the elasticity of substitution is a constant σ. These have come to be known
as CES production functions. This class of functions was first explored in a
famous paper published in 1961 by Arrow, Chenery, Minhas, and Solow [1].3
These authors prove that a production function with n inputs has constant
elasticity of substitution σ between every pair of inputs if and only if the
production function is either of the functional form
n
!k/ρ
λi xρi
X
f (x1 , . . . , xn ) = A (14)
i=1

or else of the Cobb-Douglas form

n
λi
Y
A xi (15)
i=1

where A > 0, k > 0, where λi ≥ 0 for all i, i λi = 1 and where ρ is a

P

constant, possibly negative.

You should be able to prove the following facts about ces functions:
3
This paper is a notable example of the interaction of theoretical advances and empirical
data for understanding economic phenomena. Though in retrospect, its results seem pretty
straightforward, they were a revelation at the time. Two of the authors, Arrow and Solow,
went on to win Nobel prizes.

6
Fact 2. The function as in Equation 14 has a constant elasticity of substi-
tution
1
σ= .
1−ρ
The function in Equation 15 has constant elasticity of substitution 1.

Fact 3. The function defined in Equation 14 is homogeneous of degree k.

The function defined in Equation 15 is homogeneous of degree ni=1 λi .
P

Fact 4. A CES function as in Equation 14 is quasi-concave if and only if

ρ ≤ 1. Such a function is concave if and only if it is quasi-concave and k ≤ 1.

Fact 5. The elasticity of substitution of the function in Equation 14 is greater

than one if and only if ρ > 0 and less than one if and only if ρ < 0.

This function is readily seen to be homogeneous of degree k. It is also easy

to check that the form in equation 14 has constant elasticity of substitution
σ = 1/(1−ρ) between any two variables and that in equation 15 has constant
elasticity σ = 1. The proof of the converse result that a CES function must
be of one of these two forms is not very difficult, but we will not show it here.

Cost functions for CES production functions

In this discussion, we assume there are only two factors. These results extend
in a pretty obvious way to the case of a CES function with n factors. Since
f has a constant elasticity of substitution, it must be that:
!
x2 f1 (x)
 
ln = σ ln + µ. (16)
x1 f2 (x)

where σ is the constant elasticity of substitution and µ is a constant.

The cost function c(·, ·) corresponding to the production function f (·) is
defined so that c(w, y) = min{wx|f (x) = y}. (This is mathematically the
same notion as the expenditure function e(·, ·), corresponding to the utility
function u(·), where e(p, u) = min{px|u(x) = u}.) Constant returns to
scale implies c(p, y) = c(w, 1)y where c(w, 1) is the cost of producing one
unit. By Shepherd’s Lemma the conditional factor demand for good i is
given xi (w, y) = ci (w, 1)y where ci (w, 1) is the partial derivative of c(w, 1)

7
with respect to wi . Cost minimization requires that the ratio of marginal
products is equal to the ratio of prices, so

f1 (x(w, y)) w1
= . (17)
f2 (x(w, y)) w2

According to Shepherd’s lemma,

c2 (w, 1) x2 (w, 1)
= . (18)
c1 (w, 1) x1 (w, 1)

From Equations 16 and 18 it follows that

!
c2 (w, 1) w1
 
ln = σ ln +µ (19)
c1 (w, 1) w2
Rearranging terms of equation 19, we find that
!
w2 1 c1 (w, 1) µ
 
ln = ln + (20)
w1 σ c2 (w, 1) σ

This proves the following result:

Fact 6. If the production function f (x1 , x2 ) has constant elasticity of substi-

tution σ between factors 1 and 2, then the cost function c(w1 , w2 ) must have
constant elasticity of substitution, 1/σ between the prices of factors 1 and 2.

Finding a CES cost function

Suppose that
f (x1 , x2 ) = (λ1 xρ1 + λ2 xρ2 )1/ρ
is a CES production function with λ1 + λ2 = 1. Then
! !
f1 (x1 , x2 ) x1 λ1
 
ln = (ρ − 1) ln + ln . (21)
f2 (x1 , x2 ) x2 λ2

Rearranging terms, we have

! !
x1 1 f1 (x1 , x2 ) 1 λ1
 
ln = ln − ln . (22)
x2 ρ−1 f2 (x1 , x2 ) ρ−1 λ2

8
Noting that ln(x2 /x1 ) = − ln(x1 /x2 ), we can write this as:
! !
x2 1 f1 (x1 , x2 ) 1 λ2
 
ln = ln + ln
x1 1−ρ f2 (x1 , x2 ) 1−ρ λ1
! !
f1 (x1 , x2 ) λ2
= σ ln + σ ln , (23)
f2 (x1 , x2 ) λ1

where σ = 1/(1 − ρ) is the elasticity of substitution of f .

Let c(w1 , w2 )y be the corresponding cost function. We have shown that
c(w1 , w2 ) is a constant elasticity of substitution function with elasticity of
substitution 1/σ. Therefore the function c must be of the form

c(w1 , w2 ) = (a1 w1r + a2 w2r )1/r . (24)

We have seen that a function of this form has elasticity of substitution 1/(1−
r). We have shown that the elasticity of substitution of c must also equal 1/σ
where σ is the elasticity of substitution of the original production function.
Therefore
1 1
= . (25)
σ 1−r
We can solve for r in terms of the parameter ρ of the original production
function. From Equation 25 it follows that σ = 1 − r and hence
1 ρ
r =1−σ =1− = .
1−ρ ρ−1
We still have a bit more work to do. How are the constants a1 and a2 in
the cost function related to the coefficients in the production function? One
way to find this out is as follows. At the wage rates, w1 = λ1 and w2 = λ2 ,
the cheapest way to produce 1 unit is to set

x1 (λ1 , λ2 ) = x2 (λ1 , λ2 ) = 1

and the cost of producing one units is c(λ1 , λ2 ) = λ1 + λ2 = 1. (To see this,
note that when one unit of each factor is used, the ratio of marginal products
is equal to λ1 /λ2 and total output is equal to λ1 + λ2 = 1.)
Thus, we know that

c(λ1 , λ2 ) = (a1 λr1 + a2 λr2 )1/r = 1. (26)

9
 Calculating derivatives, we have
!r−1
c1 (λ1 , λ2 ) a1 λ1
= (27)
c2 (λ1 , λ2 ) a2 λ2

We also have, from Shepherd’s lemma

c1 (λ1 , λ2 ) x1 (λ1 , λ2 )
= =1 (28)
c2 (λ1 , λ2 ) x2 (λ1 , λ2 )

From Equations 27 and 28 it follows that

!1−r
a1 λ1
= (29)
a2 λ2

From Equations 26 and 29 it then follows that a1 = λ11−r and a2 = λ21−r .

From Equation 24 it follows that if the production function is

f (x1 , x2 ) = (λ1 xρ1 + λ2 xρ )1/ρ

then the cost of producing one unit of output is given by the function
 1/r
c(w1 , w2 , 1) = λ11−r w1r + λ21−r w2r (30)

where
ρ
r= .
ρ−1
We previously found that r = 1 − σ where σ is the elasticity of sub-
stitution of the production function. Therefore the following result fol-
lows from Equation 30 and the fact that with constant returns to scale,
c(w1 , w2 , y) = c(w1 , w2 , 1)y.

Fact 7. The production function

f (x1 , x2 ) = (λ1 xρ1 + λ2 xρ2 )1/ρ

with λ1 + λ2 = 1 has constant elasticity of substitution σ = 1/(1 − ρ). The

corresponding cost function has elasticity of substitution 1/σ = (1 − ρ) and
is given by
 1/(1−σ)
c(w1 , w2 , y) = λσ1 w11−σ + λσ2 w21−σ y.

10
 We can also work backwards. If we know that if there is a CES cost
function with parameter r, then it corresponds to a CES production function
with parameter ρ = r/(1 − r).
So for example if ρ = −1, r = 1/2 and if r = 1/2, ρ = −1.
Remark: We have solved for the cost function if

f (x1 , x2 ) = (λ1 xρ1 + λρ2 )1/ρ

where r ≤ 1. It is not hard now to find the cost function for the more general
CES function
f (x1 , x2 ) = A (λ1 xρ1 + λ2 xρ2 )k/ρ
where A > 0 and k > 0. Hint: If f is homogeneous of degree 1, how is the
cost function for the production function g such that g(x) = Af (x)k related
to the cost function for f ?

Generalized means, CES functions, and limit-

ing cases
These notes are based on the presentation by Hardy, Littlewood and Polya
in their classic book Inequalities [2]. Consider a collection of numbers x =
{x1 , . . . , xn } such that xi ≥ 0 for all i. We define the mean of order r of
these numbers as !1/r
1X r
Mr (x) = x (31)
n i i
If r < 0, then the above definition of Mr (x1 , . . . , xn ) does not apply when
xj = 0 for some j, since 0r is not defined for r < 0. Therefore for r < 0,
we define Mr (x1 , . . . , xn ) as in Equation 31 if xi > 0 for all i and we define
Mr (x1 , . . . , xn ) = 0 if xj = 0 for some j.
The special case where r = 1 is the familiar arithmetic mean of the
numbers x1 , . . . , xn . The case where r = −1 is known as the harmonic mean.
So far, we have not defined M0 (x1 , . . . , xn ). We can demonstrate the
following useful fact.

Fact 8. n
X 1/r
ai
ai xri
Y
limr→0 = xi .
i=1

11
Proof. To prove this, note that
X 1/r 1 X 
ln ai xri = ln ai xri (32)
r
Applying L’Hospital’s rule to the expression on the right, we have
d
ai xri )
P !
1 X  (
lim ln ai xri = ln lim dr
d
r→0 r r→0 r
dr
ai xri ln xi
X
= lim
r→0
X
= ai ln xi . (33)

Therefore
1 X 
ai xri =
X
lim
ln ai ln xi .
r→0 r

Since the exponential function is continuous, it must be that

1/r 1
P
ln( ai xri )
X
limr→0
lim ai xri = e r (34)
r→0
P
ai ln xi
= e
n
ai
Y
= xi
i=1

The special case where r = 0 for all i is known as the geometric mean.
Two other interesting limiting cases of means of order r are the limits as
r approaches ∞ and −∞. We have the following result:
Fact 9.
lim Mr (x1 , . . . , xn ) = max{x1 , . . . , xn }
r→∞

and
lim Mr (x1 , . . . , xn ) = min{x1 , . . . , xn }
r→−∞

Here are some more useful facts about generalized means.

Fact 10. For any real number r, Mr (x1 , . . . , xn ) is homogeneous of degree
one and is a non-decreasing function of each of its arguments. Moreover,

min {x1 , . . . , xn } ≤ Mr (x1 , . . . , xn ) ≤ max {x1 , . . . , xn }.

12
Fact 11. If r < s, then unless all of the xi are equal,
Mr (x) < Ms (x).
Immediate consequences of Fact 11 are that for any collection of numbers
that are not all equal, the geometric mean is less than the arithmetic mean
and the harmonic mean is less than the geometric mean.
We have demonstrated the following fact in previous discussions.
Fact 12. The function Mr (x) is a concave function if r ≤ 1 and a convex
function if r ≥ 1.
This result is sometimes known as Minkowski’s inequality.
There are at least four different (equally weighted) generalized means that
are sufficiently commonly used to have common names. The first three listed
below are known as the Pythagoran means. The final one is known as the
root mean square.

• the familiar “arithmetic mean” (order 1)

n
1X
M1 (x1 , . . . , xn ) = xi
n i

• the geometric mean (order 0)

n
!1/n
Y
M0 (x1 , . . . , xn ) = xi
i

which is the nth root of the product of the xi ’s

• the harmonic mean (order -1)
1
M−1 (x1 , x2 , . . . , xn ) = 1

1 1

n x1
+ ... + xn

which is the reciprocal of the average of the reciprocals of the xi ’s.

• the root mean square (order 2)
1/2
1 2
 
M2 (x1 , . . . , xn ) = x1 + . . . + x2n .
n
This is the square root of the average of the squares.

13
 A simple generalization of the means of order r is the weighted mean of
order r. Where a = a1 , . . . , an ) such that ai ≥ 0 for all i and i ai = 1, we
P

define the function Mr (a, x) of x = {x1 , . . . , xn } so that

X 1/r
Mr (a, x) = ai xri . (35)

Thus the complete family of CES functions used by economists can be

regarded as weighted means of some order r.
The properties that we have found for ordinary means extend in a straight-
forward way to weighted means. Hardy, Littlewood and Polya point out that
not only are ordinary means special cases of weighted means, but weighted
means with rational weighs can be constructed as ordinary means by “re-
placing every number (in x) by an appropriate set of equal numbers.”

Exercises
Exercise 1.
A) Suppose that x1 = x2 = x3 = 4. Find Mr (x1 , x2 , x3 ) for r = 1, r = 0,
and r = −1.
B) Suppose that x1 = 1, x2 = 4 x3 = 16. Find Mr (x1 , x2 , x3 ) for r = 1,
r = 0, and r = −1. Verify that Minkowski’s inequality holds.
Exercise 2. Suppose that a vehicle travels from point A to point B at a
speed of S1 and travels back from point B to point A at a speed of S2 . The
average speed of travel is defined to be the ratio of total distance travelled
to total time spent traveling. Show that the average speed of travel for this
vehicle is the harmonic mean of S1 and S2 . Suppose that a vehicle travels at
speed S1 for two hours and speed S2 for two hours. Show that the average
speed of the vehicle is the arithmetic mean of S1 and S2 .
Exercise 3. Suppose that investment 1 offers a rate of return of r0 in odd
numbered years and r1 in even numbered years. Investment 2 offers a rate
of return r every year. If, after 100 years, the total return from investment
1 is the same as that from investment 2, how is r related to r0 and r1 ?

14
References
[1] Kenneth J. Arrow, Hollis B. Chenery, Bagicha S. Minhas, and Robert M.
Solow. Capital-labor substitution and economic efficiency. Review of
Economics and Statistics, 43(3):225–250, August 1961.

[2] G. Hardy, J.E. Littlewood, and Pólya G. Inequalities. Cambridge Uni-

versity Press, Cambridge, U.K., 1934.

15