Bilkent Economics, Ankara Econ 102, Spring 2025 G¸rkaynak

The Solow Growth Model1

Bilkent Economics
Econ 102 Introduction to Macroeconomics
Refet S. G¸rkaynak

Introduction and DeÖnitions

The study of macroeconomics falls under two broad headings: growth and business cycles. Growth
refers to the long-term, low frequency changes in economic activity while business cycles are shorter
duration áuctuations around the trend growth. This part of the class is concerned with the growth
question. Why do some countries have so much output per capita than others and why do the
output per capita in some countries grow so much faster than others? Notice that these are two
di§erent questions, one has to do with the level of output and the other with its growth rate. We
are interested in both but we will start with the former.
The stylized facts we have seen are that countries with higher savings rates have higher income
per capita and countries with higher population growth rates have lower income per capita. We also
know that countries with higher PISA scores have higher GDP per capita and some (but not all)
poorer countries tend to grow faster. Letís start with the Örst set of facts, the positive correlation
between savings rates and income per capita, and the negative correlation between population
growth rates and income per capita. Which way does the causality go? Is there a reason why
higher savings rates and lower population growth rates would cause higher income per capita?
To answer the question we will build a model. A model is a theoretical construct that allows
us to abstract away from a lot of the complexity of the world. We analyze the simpliÖed structure
to understand what the causality may be, how things may be related, and how policy actions may
change outcomes. Of course, the way we simplify the world a§ects the conclusions we reach, hence
most of the discussion in (theoretical) economics is about the assumptions we make to build the
models. If the model does a good job of Ötting the facts we are interested in, we may think that
the assumptions we made make sense and the model properly captures the relevant behavior in the
world. Then, we may be able to base policy advice on the model predictions, but it is a large and
conceptually complicated step to go from ìÖtting the factsî to ìmay be used for policy adviceî.
We will return to this issue when we talk about expectations and the Lucas critique later in the

1
These lecture notes are work in progress. BurÁin K¨sac¨koº
glu contributed to this set of notes. Please let me know
if you Önd typos or conceptual/expositional problems. Feedback will be welcomed. Please email refet@[Link]

1
Bilkent Economics, Ankara Econ 102, Spring 2025 G¸rkaynak

semester.
Economists routinely use mathematics in modeling phenomena. This is partly because our Öeld
is particularly amenable to quantiÖcation. After all, we are dealing with quantities and prices. This
comes with the large cost of leaving out anything we cannot express using math, which is a serious
cost. It does have the large beneÖt of making sure that the logic is airtight. That is, given the as-
sumptions one makes when deÖning the model, with correct mathematical analysis, the conclusions
always follow. Hence, we debate the assumptions, the building blocks of the model, rather than the
mechanics of reaching the conclusions from those assumptions. This makes economics very precise
in what is assumed and which assumption leads to which conclusion in each model.
Before we turn to the model, we need to deÖne two key concepts: equilibrium and steady state.
Unlike what you may be thinking, equilibrium is not when demand equals supply. One has to have
an understanding of what equilibrium is to understand why demand equaling supply at a given
price constitutes equilibrium.
Equilibrium is when no agent has an incentive to change her behavior given the behavior of
other agents. Russia and USA both having nuclear weapons and neither using them out of fear
that the other will retaliate is an equilibrium. (Notice that both using the weapons is also an
equilibrium. Lesson: there may be multiple equilibria and the world is not a safe place.) Demand
equaling supply at a given price is an equilibrium because at that price consumers do not want to
change their behavior (they are buying as much as they want as the point is on their demand curve)
and producers do not want to change their behavior (selling as much as they want as the point is on
their supply curve). At any other price the producers would be either producing too much, adding
to inventories and therefore having an incentive to decrease production, or would be producing too
little, selling their inventories and therefore having an incentive to increase production.
Equilibrium takes place at a point in time. But over time equilibria may change. For example
as technology improves the supply function changes, therefore the equilibrium price changes. Is
there a point when the equilibrium is no longer changing? If there is, it is called the steady state.
Understand that there is an equilibrium at all points in time but it does not have to be the same
equilibrium at all times. If, at some time, the equilibrium is no longer changing, the economy has
reached its steady state. There may never be a steady state for some variables and multiple steady
states for some others.

2
Bilkent Economics, Ankara Econ 102, Spring 2025 G¸rkaynak

The Aggregate Production Function

The key to understanding growth (levels and growth rates of trend output) is the aggregate produc-
tion function. A production function is an indispensable modeling tool. It is the function that maps
inputs into outputs. In microeconomics, a production function usually refers to a Örmís production
function, with the output being the good or service produced by that Örm. In macroeconomics, the
production function is more abstract. We imagine the GDP ìgoodî and think of the production
function of that. So the question is, what do we use to produce the GDP?
Denoting output (GDP) with Y , the production function we will assume is:

Y = F (A; K; L; H) (1)

where the arguments of the function are

A: Technology
K: Capital
L: Labor
H: Human Capital
We will Örst think of the function F itself.

Returns to scale

For simplicity, think of the production function as Y = F (K; L; H), omitting technology. We will
do a thought experiment involving scaling of inputs. Imagine multiplying all factors of production
(all inputs) by a factor x, that is, increasing all inputs x-fold. A production function is said to have

Constant returns to scale if F (xK; xL; xH) = xF (K; L; H)

Increasing returns to scale if F (xK; xL; xH) > xF (K; L; H)

Decreasing returns to scale if F (xK; xL; xH) < xF (K; L; H)

It is important to note that returns to scale here refers to scaling all inputs by the same factor.
We will soon make extensive use of the constant returns to scale production function where, say,
increasing all inputs by 35% increases output by 35% as well.
Returns to scale of the production function is distinct from the returns to scaling each individual
input. We usually expect to see diminishing returns in individual factors of production. That is,
even if the aggregate production function is constant returns to scale, increasing one factor while

3
Bilkent Economics, Ankara Econ 102, Spring 2025 G¸rkaynak

keeping others constant will lead to ever decreasing (but positive) additions to output. Why?
Having more hands (labor) without more machinery (capital) will help increase output, but as
capital becomes more diluted (less per capita) having more labor will help less and less. That is
the essence of diminishing returns (diminishing marginal product, in this case).
l!sc! z b!r !s!
g!nde yap!yars

y
d!sen !s
h!s! bas!na azcar ,
!sc! artarsa .
U !sk! 3
g!nde yapa

The Solow Growth Model

The Solow model is the seminal growth model in modern growth theory. We will study a discrete
time version of the model which requires only basic math. The neoclassical production function
and the simple factor accumulation employed in the original Solow (1956) paper help shed light on
key growth facts.
For the moment, we will simplify the production function (1) to consist only of K and L and
assume a neoclassical production function (you do not yet know enough history of economic thought
to know why this is a neoclassical production function or what alternatives there may be, but it
is good to get acquainted with the terminology) that is constant returns to scale, F (xK; xL) =
xF (K; L):
We will assume that the function F () is twice continuously di§erentiable and assume that

FK () > 0; FKK () < 0; (2)

FL () > 0; FLL () < 0: (3)

The conditions in (2) imply that the marginal product of capital, @F=@K denoted by FK , is positive
and there are diminishing returns. That is, the marginal product is decreasing as capital increases
(holding the labor input constant). This is the second derivative with respect to capital, FKK ,
being negative. We make the same assumptions for labor as well.
Further, we assume that the function satisÖes the Inada conditions:

lim FK = 1; lim FK = 0 (4)

K!0 K!1

lim FL = 1; lim FL = 0 (5)

L!0 L!1

Notice that the Inada conditions are consistent with (2) and (3). Marginal product is always
positive, and decreasing (marginal product starting at inÖnity, going to zero). The Inada conditions
deÖne the range over which marginal product is changing. The assumptions we make here are

4
Bilkent Economics, Ankara Econ 102, Spring 2025 G¸rkaynak

stronger than needed for the particular model we will work on, but in essence they are indispensable.
We will need (you will see below why) marginal product to be high enough at zero input and low
enough at inÖnite input.

The Cobb-Douglas Functional Form

To make life easy we will work with the Cobb-Douglas (1928) functional form:

Yt = Kt L1
t (6)

We use subscripts to denote time so that Yt is output in time t, Yt1 is output in time t  1, etc.
In this functional form  is a parameter between zero and one and captures shares of capital and
labor in output (this will not be obvious to you yet). Cobb-Douglas is extremely useful for a variety
of reasons that we will not go into in this class but observe that it is constant returns to scale and
conforms to conditions (2)-(5). To see the constant returns observe that for any constant x

F (xKt ; xLt ) = (xKt ) (xLt )1 = x x1 Kt L1

t = xKt L1
t = xF (Kt ; Lt ) (7)

Letís quickly show (2)-(5) for capital:

@Y
FK = = Kt1 L1
t >0 (8)
@K  
@ @Y
FKK = = (  1)Kt2 L1
t <0 (9)
@K @K

which shows positive and diminishing marginal product of capital as 0 <  < 1. We will use (8) to
verify the Inada conditions. Notice, trivially, that

1
FK = Kt1 L1
t = 1
1 Lt
Kt

hence in the limit where the denominator (K) is approaching zero marginal product goes to inÖnity
and in the limit where K is approaching inÖnity marginal product is zero. So

lim FK = 1; lim FK = 0
K!0 K!1

which are the Inada conditions we were looking to verify. Thus, we see that the constant returns
to scale Cobb-Douglas production function is an example of the neoclassical production function

5
Bilkent Economics, Ankara Econ 102, Spring 2025 G¸rkaynak

that we wanted to work with.

Evolution of Labor and Capital

We now have a production function that maps the inputs, labor and capital, into output, GDP. To
understand why GDP per capita is higher in some countries and is growing faster in some (other)
countries we need to have an understanding of the evolution (change over time) of the inputs. We
will assume that the population growth rate, savings rate, depreciation rate and the initial levels
of labor and capital are exogenously given. (That is, these values are not determined inside the
model.)
To make things as simple as possible the Solow model has exogenously given, constant popula-
tion growth at rate n. Further, we assume that everyone works so that population is equal to the
labor force. Given the initial population, L0 , growth rate n implies that population at any time t
is
Lt = (1 + n)Lt1 = (1 + n)2 Lt2 = ::: = (1 + n)t L0 (10)

Next, we have to think about what happens to output. We assume that this is a closed economy,
N X = M = X = 0, and there is no government spending, G = 0. Then, the GDP identity is

Yt = Ct + It (11)

We will take a mechanical approach to the model and assume that a social planner tells everyone
to work2 (so labor is equal to population) and save a constant fraction s of the total output (which,
as you will remember from basic GDP accounting, is the total income). We will denote savings
with St . Hence we have consumption equal to a fraction (1  s) of output. We then have

Savingst  St = sYt (12)

But (11) tells us that

It = Yt  Ct = Yt  (1  s)Yt = sYt (13)

2
We could have also assumed households inelastically supplying their labor in competitive markets, earning a wage
equal to marginal product, and renting capital to Örms, earning a competitive return on that as well. That is, we can
have an alternative, competitive markets interpretation of the model where Örms (but not households) optimize. And
the results would have been identical (with some use of Inada conditions). The mechanical interpretation is easier as
we do not need to solve for equilibrium wages and returns to capital, and helps build intuition. We will proceed with
that.

6
Bilkent Economics, Ankara Econ 102, Spring 2025 G¸rkaynak

Putting (12) and (13) together, we have

St = It (14)

Equation (14) is of fundamental importance. It tells us that in a closed economy with no

government, savings have to equal investment. We will see later in the semester that this holds for
closed economies with government as well. In a sense this is an accounting identity. Goods that are
produced, if they are not consumed (and cannot be exported or used by government in this model),
have to be either willingly invested, that is added to the capital stock, or added to inventories,
which is also measured under investment. But a more fundamental insight is that eq (14) is also an
equilibrium condition. Interpreting It as desired investment (there will be no undesired inventory
changes in equilibrium, as producers would change behavior if their inventories were going up or
down outside their wishes), in equilibrium savings will be equal to desired investment. This is the
sense in which this model is always in equilibrium.
We now understand that countries that save more get to invest more. Remember that investment
is the addition to the capital stock, hence countries that save more will end up having more capital.
This is the Örst glimpse into the link between saving rates and output per capita. More capital,
as the marginal product of capital is positive, will lead to more output. Will it also lead to more
output per person? We will see soon.
Investment adds to capital but capital also depreciates. Every period, when producing the
GDP, some of capital is ìlostî because of wear and tear due to usage. The reduction in capital due
to this wear and tear is called depreciation. We assume capital depreciates at a constant rate  so
that Kt is lost at any period t due to production.
Taking all of these together, the law of motion of capital is

Kt+1 = Kt  Kt + It = (1  )Kt + sYt (15)

In any period t+1, (1)Kt of capital from last period is available. The savings out of last periodís
output, sYt is investment and is added to the capital stock, hence the total amount of capital used
in production in t + 1 is the right hand side of (15).
This completes the description of the model. Now we can ask whether this model has predictions
about the level and growth rate of per capita GDP that are consistent with the data we see.

7
Bilkent Economics, Ankara Econ 102, Spring 2025 G¸rkaynak

Steady State of the Solow Model

We want to think about the steady state of this model. Remember that the steady state is when
the equilibrium is no longer changing. The Solow model will not admit a (nontrivial) steady state
in output as the economy will grow without bound with constantly increasing labor. (K0 = 0 is a
trivial and uninteresting steady state. It is a steady state because output is zero if capital is zero
regardless of the amount of labor (see (6)) and with zero output savings and investment are zero
so the capital stock does not change and output remains at zero. This is an unstable steady state,
if there is a small but positive amount of capital then output and investment will be positive and
capital stock will increase, moving output away from the zero steady state. We can be sure that
a small amount of capital will lead to large enough output to allow investment to be greater than
depreciation because marginal product of capital is large (indeed inÖnite) when the capital stock
is close to zero. This is the use of the Inada condition.)
We will convert the model into per capita values. This is for two reasons, Örst because we are
interested in output per capita, second because while total output does not have a (nontrivial)
steady state, per capita output may have one. That is, since the lack of a steady state comes from
increasing labor, we may be able to side step this issue by transposing the model into per capita
values. We will denote per capita values using lower case letters

Yt Kt
yt = ; kt =
Lt Lt

Using this notation and dividing both sides of the production function (6) with Lt we have

Yt 1  1
= K L
Lt Lt t t
) yt = kt ; (16)

which tells us that output per capita is a concave function of capital per capita only.3 We know that
dividing a concave function with a constant will maintain concavity but take the two derivatives
and verify that the right hand side of (16) is concave anyway.
We learned something important here: what matters for output per capita is capital per capita.
Hence, China having more labor than Turkey and also having more capital than Turkey are un-

3
In general when the production function is linearly homogenous (constant returns to scale), Yt =Lt =
(1=Lt )F (Kt ; Lt ) = F (Kt =Lt ; 1), often denoted as yt = f (kt )

8
Bilkent Economics, Ankara Econ 102, Spring 2025 G¸rkaynak

informative about the relative output per capita levels. We need to know which country has the
larger capital to labor ratio to know which one would produce more per worker. Note that in a
more realistic model we would also have technology levels and human capital making a di§erence.
This model, as we have built it, only allows us to talk about capital and labor and has so far told
us that what matters for the determination of output per capita is capital per capita, not capital
and labor independently.
We will similarly transform the law of motion of capital into per capita values by dividing both
sides of (15) with Lt . That gives

(1 + n)kt+1 = (1  )kt + skt (17)

Kt+1 Kt+1 Lt+1 Kt+1 Lt+1

(1 + n) appears on the left hand side because Lt = Lt : Lt+1 = Lt+1 : Lt = kt+1 (1 + n). We can
further manipulate the equation above to have an expression for the change in per-capita capital.
Subtract (1 + n)kt from both sides to have:

(1 + n)kt+1 = skt  (n + )kt (18)

where kt+1 = kt+1  kt is the change in per capita capital between two periods. For the per capita
capital stock to be constant, that is for kt+1 = 0, we need skt = (n + )kt . This is called the
ìbreakeven investmentî. To keep the per capita capital stock level constant, the economy needs to
invest just enough to o§set the e§ects of depreciation and population growth.
Remember that yt = kt so understanding the evolution of kt alone lets us also understand
how yt is behaving. For that reason, (17) is the backbone of the Solow model. Notice that this
equation tells us that other things equal, higher population growth rate leads to lower capital per
capita (hence lower output per capita).4 This is a typical economistsí moment. We laid out a
model (made modeling assumptions) and carried out correct logical operations on the assumptions
(used correct mathematics) which led to a conclusion that is apparently correct, but we do not
yet understand the economics behind it. Why does a higher population growth rate lead to lower
output per capita?
The answer is quite lovely. Higher population growth rate does not lead to lower capital per
capita today, it leads to lower capital per capita tomorrow. Notice that it is kt+1 that is a§ected.
Why? Because we do investment using todayís savings, which comes from todayís production,

4
kt+1 = 1
(1+n)
((1  )kt + skt ) so higher n implies lower kt+1 .

9
Bilkent Economics, Ankara Econ 102, Spring 2025 G¸rkaynak

which was done using todayís labor. That determines tomorrowís capital stock. But if there are
↑

more people around tomorrow because of population growth, there will be more people to use that
capital. That, e§ectively, dilutes the capital; there is less capital per head to be used. Hence,
on average, we all work with less capital because new additions to the labor force, who had not
contributed to investment last period, also use capital. That is why higher population growth rate
leads to lower capital per capita. And that is why the Solow model tells us that countries with
quickly increasing populations will have lower output per capita, which is what we saw in the data.
We want to formalize this argument and for that we need to Önd the steady state capital per
capita, if a steady state exists. Weíll do this constructively, that is, we will assume there is a steady
state and will try to Önd it. (If there is no steady state we will fail.5 ) Denote the steady state value
of capital per capita with k ss . At steady state

k ss = kt = kt+1

that is, once we reach the steady state for any t and t + 1 (or t  1 and t, etc.) the capital per
capita is the same. (16) says in that case output per capita will be unchanging as well, hence yt
will also be at steady state. Setting the per capita capital stock equal to k ss on both sides of (17)
we Önd the steady state per capita capital stock to be S! = 45X
n = 45t
  1
s 1
k ss = (19)
(n + )

This veriÖes the intuition we have built so far. Countries that save more have more capital per
capita at steady state because they invest more, and countries with faster population growth rates
have less capital per capita because population growth lowers the amount of capital available to
each worker. Output per capita at steady state is correspondingly =>
nufusu veren Allch

   sermayey! de
ver!yor mut
s 1
y ss =
(n + )

which follows directly from the capital per capita.

We see that the Solow growth model helps us make sense of two of the stylized facts that we have
found. Output per capita is positively correlated with the savings rate and negatively correlated
with the population growth rate. We now have a theory that tells us that causality runs from

5
Which would be surprising as we already worked out the condition for kt+1 = 0:

10
Bilkent Economics, Ankara Econ 102, Spring 2025 G¸rkaynak

savings and population growth rates to output per capita. Good. We learned something.

Predictions of the Solow Model

The model presented in the previous section is a simple but a very powerful one to analyze cross-
country di§erences in income. A clear prediction of this model is that countries that save more and
that have lower population growth rates will have higher income per capita at steady state.
Can we have a similar prediction in terms of growth rates? SpeciÖcally, can we infer from the
model that poorer countries will grow faster compared to richer countries? To answer this question
we should delve deeper into the model and work out what it implies about the growth rates. The
main driver of aggregate output and output per-capita growth in this simple version of the Solow
model is the change in the capital stock. Hence to be able to understand the modelís predictions
about growth rates, Örst we need to understand the modelís prediction about growth rate of per-
capita capital. So letís go back to equation (18). Left hand side of that equation is the change in
per-capita capital stock, and the right hand side is the di§erence between the actual and break-even
levels of investment. Divide both sides by kt to have:
 
1 s
g(k) =  (n + )
(1 + n) kt1

where g(k) is kt+1 =kt = (kt+1  kt )=kt , which is the per capita capital growth rate. This equation
shows that lower capital stock implies higher growth rate since the Örst term inside the brackets
on the right hand side will be larger than the second term.6 The expression inside the bracket on
the right hand side is a measure of how far the economy is from its steady state. Notice that if
kt = k ss , then this expression is equal to 0. The farther below kt is compared to k ss , the higher the
growth rate of per capita capital. Since output per-capita growth depends on the growth rate of
the capital stock, the model will have the same predictions about the variable of interest, growth
rate of output per capita.
Assume that there are two countries with the same characteristics. That is, they have the
same saving rate s, population growth n, depreciation rate , and capital share . What generates
the cross-country di§erences in growth rates? According to the model (and the derivation above),
distance from the steady state level determines how fast the economy is growing. If an economy
is away from (and below) its steady state it grows faster compared to other countries with similar

6
The second term,(n + ), is a constant and the Örst term goes to inÖnity as kt goes to zero.

11
Bilkent Economics, Ankara Econ 102, Spring 2025 G¸rkaynak

parameters, and hence the same steady state, that are closer to that steady state.
Assuming that countries have exactly the same characteristics (or the same steady state) is very
restrictive. The natural question here is what happens if countries have di§erent steady states? It is
reasonable to assume that di§erent countries have di§erent steady states. For instance countries in
sub-Saharan Africa might have a lower steady state of per capita capital stock (hence lower steady
state of per capita income) than European countries due to di§erences in saving and population
growth rates. The Solow model predicts that di§erences in steady states do not a§ect how fast
the country will grow because at the steady state all growth rates are zero. Away from the steady
state, what matters for growth is the countryís distance from its own steady state. This is called
ìconditional convergenceî. Poorer countries are not predicted to grow faster, only countries that
are low income compared to their steady states are.
Another very important question has to do with growth at the steady state. This version of the
Solow model cannot generate long run growth: Once the country is at its steady state, output per
capita stops growing. This is counterfactual since we see even the highest income countries growing.
To generate that result, we need to have a mechanism that would generate growth at the steady
state. An extension of this model, which incorporates exogenous technological change, predicts per
capita output growth even at the steady state. There are also models that endogenize (explain
within the model) steady state growth. These are called endogenous growth models. These will be
covered in more detail in more advanced courses.
Do note a very important insight Solow had: while saving and investing more will produce
more output per capita at steady state, it will not bring perpetual growth in income per capita.
ìGrowthî is change. Income per capita will not change based on constant investment. And we
cannot increase the savings/investment rate forever. So if there is to be permanent growth, it will
have to come from a source that is not exhaustible. That is why technological advancement is
so important. If we are away from our steady state, we will grow towards that level even in the
absence of teachnological growth. But near steady state we will stagnate unless the steady state
itself is growing (hence there is no ìsteady stateî, the term for something resembling a steady state
with growth is a ìbalanced growth pathî, more on that in future classes).

12