Lecture 2: Production Functions

Dongling Su*

1 The Source of Output

We reviewed some basic facts about the growth and level of output or GDP across coun-
tries in the last class. To explain those facts with a model, we must first specify the source
of output.

Definition: A production function is a function specifying how much output can be pro-
duced with a given combination of inputs.

Labor L and capital K are the two inputs we will consider most often, but there is no
reason in principle to restrict ourselves to this simple list of inputs. In general, microe-
conomics uses production functions to describe the output of an individual firm. In this
macroeconomics class, we will typically use a production function to instead describe the
output or GDP of the entire economy.

2 The Neoclassical Production Function

This workhorse of macroeconomics describes the overall output or GDP in an economy
Yt at time t as a function F of capital Kt and labor Lt

Yt = F (Kt , At Lt ).

There is no such thing as an aggregate production in reality. But we can use such a
mathematical tool to give reasonable approximations. Cobb and Douglas argue in their
famous paper (Cobb and Douglas, 1928 AER) that “The advantage in choosing a norm at
all seems to be that it involves us in logical consequences which may be compared with
the facts as we get the facts. It enables us to talk rightly or wrongly with more precision
and to draw conclusions which become hypotheses. ”
* These notes borrow heavily from notes by Adam Guren, Simon Gilchrist, Francois Gourio, and Stephen
Terry.

1
2.1 Technology
In the equation above At is some number which describes the level of technology in
the economy. It is also sometimes convenient to think of At as representing the level of
knowledge in an economy. For much of this class we will assume that At is exogenous,
i.e. we will take the level or path of At over time as given.
I have written technology At as labor augmenting above, which means that At enters
multiplicatively with respect to Lt . Therefore, increasing the level of technology makes
labor Lt more productive, and we refer to At Lt as the efficiency units of labor.
Production functions can also be written to assume capital augmenting technology,
F (At Kt , Lt ), or with Hicks neutral technology At F (Kt , Lt ). The distinction will typically
make little difference for our purposes, but for concreteness we will focus on labor aug-
menting technology.

2.2 Properties and Definitions

We will typically make several natural assumptions about the production function F . First,
we will typically assume that F is differentiable with respect to all its inputs and that the
higher order derivatives of F always exist. This allows us to lay out some additional
terminology and assumptions.

• The marginal product of an input is the partial derivative of output with respect to
that input. The marginal product of capital M P Kt is given by
∂
M P Kt = F (Kt , At Lt ) = F1 (Kt , At Lt ).
∂Kt
Holding At and Lt fixed, M P Kt represents the approximate increase in output Yt
that would result from an additional unit of capital. Similarly, the marginal product
of labor M P Lt is given by
∂
M P Lt = F (Kt , At Lt ) = F2 (Kt , At Lt )At .
∂Lt
The first equality is a definition, while the second equality results from application
of the Chain Rule. As with capital, M P Lt is the approximate increase in output Yt
resulting from an additional unit of labor, holding At and Kt fixed.

• We typically assume that both inputs are productive or that the production function
F is strictly increasing in both inputs:

M P Kt > 0, M P Lt > 0.

2
• We also typically assume diminishing returns to each input by itself:

∂ ∂2
M P Kt = F (Kt , At Lt ) < 0
∂Kt ∂Kt2

∂ ∂2
M P Lt = F (Kt , At Lt ) < 0
∂Lt ∂L2t
Holding capital fixed, the additional output generated by hiring an additional unit of
labor is decreasing in the level of labor already used. Similarly, holding labor fixed,
the additional output from an extra unit of capital declines with the level of capital.

• Often, we will also assume that capital and labor are complements. This means
that the marginal product of capital is increasing in the level of labor, and vice-versa,
i.e.
∂ ∂
M P Kt > 0, M P Lt > 0.
∂Lt ∂Kt
In practice, this means that workers with more capital are more productive, and
capital with more workers is more productive. It’s worth pausing for a moment to ask
whether either of these statements can be true if the other is false? The following
theorem is helpful.
Young’s Theorem: If the second partial derivatives of f (x, y) are continuous, then
∂2 ∂2
∂x∂y
f (x, y) = ∂y∂x f (x, y).

• We will also often assume that the production function F satisfies constant returns
to scale in capital and labor. This means that for any constant λ > 0 the function F
satisfies
F (λKt , At λLt ) = λF (Kt , At Lt ).
This implies, for example, that doubling both capital and labor doubles the output of
the economy as can be seen by setting λ = 2.

• We also typically assume that the production function satisfies the following techni-
cal conditions:

– F (Kt , 0) = 0 and F (0, At Lt ) = 0

– F (Kt , 1) is concave
∂
– limKt →∞ ∂Kt
F (Kt , 1) =0
∂
– limKt →0 ∂Kt
F (Kt , 1) =∞

The first condition simply states that both inputs are essential for positive produc-
tion. Later on, the other conditions will help to guarantee a positive and finite level
of output per capita in the long run.

3
2.3 The Intensive Form of the Production Function
With a constant returns to scale production function, we will often work with the intensive
form of the production function. This form expresses output per efficiency units of labor
as a function of capital per efficiency units of labor. We will use undercase variables to
refer to these quantities: yt = AYt Lt t and kt = AKt Lt t . With this notation in place, we can write

   
Yt 1 Kt At Lt Kt
yt = = F (Kt , At Lt ) = F , =F , 1 = F (kt , 1).
At Lt At Lt At Lt At Lt At Lt
We can now define a new function, the intensive form of the production function,
f (kt ) = F (kt , 1). With constant returns to scale we therefore have

yt = f (kt ).

Because we assumed that capital is productive and has diminishing returns in the original
function F , we can immediately conclude that the intensive form f satisfies the same
properties
∂ ∂
f (kt ) = F (kt , 1) > 0
∂kt ∂kt
∂2 ∂2
f (k t ) = F (kt , 1) < 0
∂kt2 ∂kt2
The intensive form is convenient because it allows us to summarize a two-input constant
returns to scale production function using a single input. If we are only interested in the
output per efficiency unit of labor yt , then we only need to track the ratio of capital to the
efficiency units of labor kt .

2.4 Example: Cobb-Douglas Production Function

The Cobb-Douglas production function, first proposed by Charles W. Cobb and Paul Dou-
glas in the 1920s, is given by
Yt = Ktα (At Lt )1−α
for some α ∈ (0, 1). The constant α is often referred to as the “capital elasticity” or
“capital share” for reasons that will become clear later. We can easily compute each of
the quantities above and verify each of the assumptions above for this functional form:

• The marginal products of capital and labor are given by

∂ Yt
M P Kt = Ktα (At Lt )1−α = αKtα−1 (At Lt )1−α = α
∂Kt Kt
∂ Yt
M P Lt = Ktα (At Lt )1−α = (1 − α)Ktα (At Lt )−α At = (1 − α)
∂Lt Lt

4
 • The expressions above imply that M P Lt > 0 and M P Kt > 0 for any Kt , Lt > 0, so
both capital and labor are productive inputs.

• Both capital and labor have diminishing returns. Note that

∂ ∂ Yt
M P Kt = αKtα−1 (At Lt )1−α = α(α − 1)Ktα−2 (At Lt )1−α = −α(1 − α) 2
∂Kt ∂Kt Kt
∂ ∂ Yt
M P Lt = (1−α)Ktα (At Lt )−α At = (1−α)(−α)Ktα (At Lt )−α−1 A2t = −α(1−α) 2 .
∂Lt ∂Lt Lt
∂ ∂
Since 0 < α < 1, ∂Kt
M P Kt < 0 and ∂Lt
M P Lt < 0.

• Labor and capital are complements. Note that

∂ ∂
M P Kt = αKtα−1 (At Lt )1−α = α(1 − α)Ktα−1 (At Lt )−α At .
∂Lt ∂Lt
∂
Since 0 < α < 1, ∂Lt
M P Kt > 0. By Young’s theorem M P Lt is also increasing in Kt .

• The Cobb-Douglas production function exhibits constant returns to scale. For λ > 0

F (λKt , At λLt ) = (λKt )α (At λLt )1−α = λα λ1−α Ktα (At Lt )1−α = λKtα (At Lt )1−α = λF (Kt , At Lt ).

• The Cobb-Douglas production function satisfies our technical conditions, since

– F (0, At Lt ) = 0α (At Lt )1−α = 0 and F (Kt , 0) = Ktα 01−α = 0

∂2
– F (Kt , 1) = Ktα is concave with Kα
∂Kt2 t
= −α(1 − α)Ktα−2 < 0
– limKt →∞ ∂
∂Kt
F (Kt , 1) = limKt →∞ αKtα−1 = 0
– limKt →0 ∂
∂Kt
F (Kt , 1) = limKt →0 αKtα−1 = ∞

At this point, it is useful to point out a few facts about the Cobb-Douglas production func-
tion which will come in handy later:

• The marginal product of each input is proportional to the average product:

Yt Yt
M P Kt = α ∝ = Average Product of Capital
Kt Kt
Yt Yt
M P Lt = (1 − α) ∝ = Average Product of Labor
Lt Lt
• Because the Cobb-Douglas production function satisfies constant returns to scale,
the intensive form will often be useful. The intensive form is given by

f (kt ) = ktα .

5
 • The elasticities of output with respect to capital and labor are given by α and 1 − α,
respectively. To see this, note that
∂Yt Kt ∂ log Yt
= =α
∂Kt Yt ∂ log Kt
∂Yt Lt ∂ log Yt
= = 1 − α.
∂Lt Yt ∂ log Lt
For each equation, the first expression on the left hand side above is the definition
of the elasticity. The first equality then follows from the argument below. The sec-
ond equality follows from taking the logarithm of the production function to get the
equation log Yt = α log Kt + (1 − α) log At + (1 − α) log Lt .
Elasticities: Economists love elasticities, as they are unit-free. It measures how 1%
changes in X would lead to ϵ% changes in Y, and is defined as

∆Y /Y ∆Y X ∂Y X
ϵ≡ = · ≈ · ,
∆X/X ∆X Y ∂X Y

where the last equality holds if changes in X are sufficiently small.

Log Differentials & Elasticities: Let Y = f (X) for differentiable X. Then define
the transformations x = log X and y = log Y . We have

Y = f (X) (assumed)
ey = f (ex ) (substitute)
∂ y ∂
e = f (ex ) (differentiate)
∂x ∂x
y ∂y
e = f ′ (ex )ex (Chain Rule)
∂x
∂ log Y X
= f ′ (X) (re-arrange and substitute for elasticity formula)
∂ log X Y

2.5 Cobb and Douglas (1928, AER) and the History of Production
Function
In Figure 1, Paul Douglas plotted the time series of output, capital and labor in log scale.
And he found the curve for product to lie, in general, approximately one-quarter of the dis-
tance between the curve for labor and the curve of capital. He spoke with mathematician
and colleague Charles Cobb, who suggested the now-famous function form. Estimating
the production function using data from 1899 to 1922 led to Figure 2, which showed that
the Cobb-Douglas production fitted data pretty well.

6
 Figure 1: Time series of Y , K and L in log.

Figure 2: The estimated Cobb-Douglas production function seem to fit data well.

7
 The production function analysis faced harsh critics initially: “...so fews observations
were involved that any mathematical relationship was purely accidental and not causal ...
all past work should be torn up and consigned to the wastepaper basket.” And Douglas
was like “I must admit that I was discouraged by this criticism and thought of giving up the
effort, but there was something which told me that I should hold on. ”
But they then moved on to use cross-section American studies (1904, 1909, 1914, and
1919)1 involving 1,490 observations at industry level. They assume production function
takes the form of P = bLk C j . Therefore, k and j were independently estimated. As a
result, they can also test whether the economy was subject to constant returns to scale
by comparing k + j with 1.0. The results were shown in Figure 3. As can be seen, the
estimates of k and j were pretty precise and the constant returns to scale seemed to hold.

Figure 3: Estimates from American Cross-section Studies.

Douglas (1976, JPE) summarized the followed-up studies that used (macro or micro)
data from western coutries and concluded that

• Cobb-Douglas production fitted aggregate output reasonably well.

• The sum of k and j approached unity, or constant returns, but were nearly always
slightly below unity. The values of labor share were around 0.6 to 0.7.

• The standard errors of k were comparatively slight, raning from 1/10 to 1/15 the
value of k

2.6 Labor share in US and China

• The labor share in US is fairly stable, which supports the formulation of Cobb-
Douglas production function.
1
Note there are still pros and cons. Good: short time span, less worries about changes in the production
function over time. Bad: Need to assume identical functions across industries.

8
Figure 4: Labor Share in US (left) and China (right). Note: the left panel is taken from
Karabarbounis and Neiman (2018, factorless income); the right panel is taken from Chang
et al. (2016, Trends and Cycles)

• Recent empirical evidence shows that the labor share in US (or more broadly for
other countries as well) is slightly declining.

• The labor share in China declined significantly around 2000s. Chang et al. (2016)
argues this is due to the deveopment of the heavy industry.

3 An Argument for an Aggregate Production Function

The assumption that we can use a single production function F to describe the output of
an entire economy is clearly quite strong. A more palatable starting point might be to use
the same production function to describe the output of a single firm. However, it turns out
that under the assumption of constant returns to scale there is a natural link between the
firm and aggregate production functions.
First, let’s review the standard argument in support of constant returns to scale which
is known as the replication argument. It relies on the following thought experiment.
First, take an entire economy, say Mexico, which uses some given levels of capital, labor,
and technology to produce output. Then, imagine that we snap our fingers and magically
duplicate Mexico to create a second, identical economy alongside the first one. The output
of the two Mexicos together should simply be twice the output of the original Mexican
economy, as long as they are using the same production practices and technology to
generate output. This result is exactly the constant returns to scale definition with λ = 2.
If we accept the replication argument and hence the constant returns to scale assumption,
we can now given an example of conditions under which there is a tight link between the
firm-level and aggregate production functions.

9
3.1 Example Problem: The Irrelevance of Microeconomics
There are N firms in an economy indexed by i = 1, ..., N . Each firm produces output Yi
using capital Ki and labor Li with access to the same production function Yi = F (Ki , ALi )
featuring fixed technology A. The total amount of capital and labor in the economy are
given by K = N
P PN
i=1 Ki and L = i=1 Li , respectively. Assume that F exhibits constant
returns to scale. Furthermore, assume that each firm uses capital and labor in the same
Ki
proportion satisfying AL i
= κ for some constant κ. Prove that “microeconomics is ir-
PN
relevant,” i.e. show that the total level of output Y = i=1 Yi must always satisfy the
aggregate production function Y = F (K, AL) regardless of the allocation of inputs across
firms {(Ki , Li )}N
i=1 .

Solution: Under these assumptions of identical factor proportions Ki = κALi for all firms.
Write
N
X XN XN
K= Ki = κALi = κA Li = κAL.
i=1 i=1 i=1

Therefore, the aggregate capital and labor inputs satisfy the same proportionality as the
K
firm-by-firm inputs with κ = AL . Then, we can write
N
X
Y = Yi (definition)
i=1
N
X
= F (Ki , ALi ) (definition)
i=1
N
X Ki
= ALi F ( , 1) (constant returns to scale)
i=1
ALi
N
X
= ALi F (κ, 1) (assumption)
i=1
N
X
= AF (κ, 1) Li (constants)
i=1
= ALF (κ, 1) (definition)
K
= ALF ( , 1) (result above)
AL
= F (K, AL) (constant returns to scale)

This relationship, Y = F (K, AL) is exactly the desired aggregate production function. ■

3.2 Profit Maximization with Capital and Labor

The aggregation example 3.1 above relies upon the assumption of identical input ratios at
firms. So far we have only discussed possibilities for production, so we do not yet have a

10
framework for predicting the choice of inputs. In this section, we’ll offer one simple set of
predictions for firm behavior relying upon the assumption of profit maximization.
Consider a firm i with a constant returns to scale production function F (Ki , ALi ) facing
a fixed technology level A. The firm can hire labor at the wage rate W per unit and may
rent capital at the rental rate R. The profits of the firm are given by output minus input
costs, i.e. F (Ki , ALi ) − W Li − RKi . This expression embeds the assumptions of perfect
competition or price taking in both the output and input markets. Under these conditions,
a firm which seeks to maximize its profits will choose optimal levels of capital and labor
by solving
max F (Ki , ALi ) − W Li − RKi .
Ki ,Li

It is straightforward to show that the first-order conditions for optimality in this problem are

M P Ki = R, M P Li = W.

The firm must set the marginal product of each input equal to its marginal cost, which is
the rental rate R for capital and the wage rate W for labor.
If we are willing to assume a Cobb-Douglas form for the production function, we can
proceed further. In that case, Yi = F (Ki , ALi ) = Kiα (ALi )1−α . Above we showed that with
a Cobb-Douglas production function the marginal product of an input is proportional to its
average product:
Yi Yi
M P Ki = α , M P Li = (1 − α) .
Ki Li
The first-order conditions for profit maximization can then be written
Yi Yi
α = R, (1 − α) = W.
Ki Li
Taking the ratio of the two first-order conditions, rearranging, and multiplying by 1/A yields

Ki Wα
= .
ALi AR(1 − α)

If we consider N firms i = 1, ..., N which all face the same wage W and rental rate R, then
each firm’s optimal choice of capital per efficiency unit of labor is equal to the constant
Wα Wα
AR(1−α)
. Setting κ = AR(1−α) yields profit-maximizing inputs at each firm exactly satisfying
the assumptions of Example 3.1 and implying the existence of an aggregate production
function.

11