EC 502 Lecture 5: Solow Model in

Continuous Time
Dongling Su*

1 Continuous Time Basics

We have previously worked with discrete time using integer values for t = 0, 1, 2, ... However, this
lecture introduces a version of the Solow model in continuous time, with t ∈ [0, ∞). This will require
us to use different notation to express growth rates. In discrete time, variables evolved according
to rules like
Xt+1 − Xt
Xt+1 = (1 + g)Xt → g = ,
Xt
and we said that g was the growth rate of X from t to t + 1. Now rewrite this expression as
∆X 1
g= , (1)
∆t Xt
where ∆X is the change in X over the period ∆t in which the growth rate is computed. For
continuous time, we shrink the interval of time ∆t → 0 to consider the instantaneous rate of
change:
∆X dX(t)
lim = .
∆t→0 ∆t dt
When working in continuous time, we write X(t) rather than Xt to emphasize that X(t) is a function
of the continuous time index t. Time derivatives show up a lot, so we use dot notation Ẋ(t) instead
of dX(t)
dt . Now return to Equation (1) with the new notation and write, for small ∆t → 0:

Ẋ(t) = gX(t). (2)

This is a differential equation implicitly defining X(t). If X(0) is the initial condition for the variable
X(t), the solution to this equation is given by

X(t) = X(0)egt ,

since in that case Ẋ(t) = gX(0)egt = gX(t). Also, note that Equation (2) defining the instanta-
neous growth rate of X is usually written

Ẋ(t)
g= .
X(t)
* These notes borrow heavily from notes by Adam Guren, Stephen J. Terry, Simon Gilchrist, and Francois
Gourio.

1
Starting from that formulation, it is easy to see by applying the Chain Rule that

Ẋ(t) d log X(t)

g= = .
X(t) dt

So in continuous time, the instantaneous growth rate of X(t) is equal to the change in log X(t).
This relationship between growth rates and log differences is only approximate for discrete time,
Ẋ(t)
but it holds exactly in continuous time. For situations described by X(t) = g or X(t) = X(0)egt , we
say that “X grows at rate g.”

2 Four Key Equations Restated in Continuous Time

Equations (3) - (5) below are the core of the continuous time Solow model.

2.1 Output
Output Y (t) is determined according to the neoclassical production function

Y (t) = F (K(t), A(t)L(t)), (3)

with capital K(t), labor L(t), and technology A(t) at time t. The neoclassical production func-
tion satisfies all the assumptions covered in Lecture 2, including CRTS, productive inputs, and
diminishing marginal returns to each input.

2.2 Capital
The capital stock K(t) at any point in time evolves instantaneously according to investment I(t)
and depreciation δK(t), through
K̇(t) = I(t) − δK(t).
A constant fraction s ∈ (0, 1) of output is exogenously saved, so savings S(t) are given by S(t) =
sY (t) = I(t). Capital must satisfy

K̇(t) = sY (t) − δK(t). (4)

2.3 Labor and Technology

Labor L(t) grows exogenously at rate n, and technology grows exogenously at the rate g:

L̇(t) Ȧ(t)
= n, =g (5)
L(t) A(t)

2
3 Normalization of Capital
As in the discrete time case, normalize to obtain capital per efficiency unit of labor k(t) and output
per efficiency unit of labor y(t) linked by the intensive form f of the production function:

K(t) Y (t)
k(t) = , y(t) = , y(t) = f (k(t)),
A(t)L(t) A(t)L(t)

where f (k(t)) = F (k(t), 1). Rewrite equation (4) as

1
K̇(t) = sy(t) − δk(t) = sf (k(t)) − δk(t). (6)
A(t)L(t)
d f (x) f ′ (x)g(x)−f (x)g ′ (x) f ′ (x) f (x)g ′ (x)
Then, apply the Quotient Rule dx g(x) = g(x)2
= g(x) − g(x)2
to obtain

1 K(t)  
k̇(t) = K̇(t) − Ȧ(t)L(t) + A(t) L̇(t)
A(t)L(t) [A(t)L(t)]2
!
1 K(t) Ȧ(t) L̇(t)
= K̇(t) − + .
A(t)L(t) A(t)L(t) A(t) L(t)

Use equations (5) and (6), together with the final expression above, to obtain

k̇(t) = sf (k(t)) − (n + g + δ)k(t). (7)

The continuous time equation (7) is the equivalent of the discrete time equation (5) from the notes
for Lecture 3, but it is more parsimonious and convenient.

• The first right hand side term sf (k(t)) is actual investment per effective unit of labor.

• The second right hand side term −(n + g + δ)k(t) is breakeven investment representing
the combined effects of depreciation δ, population growth n, and technology growth g. To
prevent k(t) from falling with k̇(t) < 0, investment must be greater than or equal to the loss
in normalized capital units k(t) from each of these three forces.

4 Steady State and Dynamics of the Model

The top panel of Figure 1 plots the first and second terms on the right hand side of equation (7)
as a function of k(t). The bottom panel plots k̇(t) as a function of k(t). The level of capital per
efficiency unit of labor such that k(t) does not change with k̇(t) = 0 is the steady state k ∗ . The
steady state k ∗ therefore satisfies

sf (k ∗ ) = (n + g + δ)k ∗ .

At the point k ∗ , new investment perfectly offsets depreciation, population growth, and technolog-
ical advancement to result in no change in k(t). The continuous time version displays stable
convergence of k(t) towards k ∗ for any starting point k(t) > 0. In other words

lim k(t) = k ∗ .
t→∞

3
 Investment and Breakeven

(n + g + δ)k(t)

sf(k(t))

k*

Rate of Change of Capital

0
dk dt

k*
k(t): Capital per Efficiency Unit
Figure 1: Evolution of Capital in the Solow Model

4.1 Transition Dynamics

We call the movement of k(t) as it approaches k ∗ transition dynamics. There are two cases:

• For starting points k(t) < k ∗ lower than steady state, sf (k(t)) > (n + g + δ)k(t). Investment
is higher than breakeven levels, so k(t) increases over time and k̇(t) > 0.
• For starting points k(t) > k ∗ higher than steady state, sf (k(t)) < (n + g + δ)k(t). Investment
is lower than breakeven levels, so k(t) decreases over time and k̇(t) < 0.

4
 k*

Rate of Change of Capital

0
dk dt

k*
k(t): Capital per Efficiency Unit
Figure 2: Convergence in the Solow Model

Intuitively, the high marginal product of capital for low levels of k(t) guarantees that the investment
force is stronger for low levels of capital. Diminishing returns to capital imply that for very high
levels of k(t), the extra savings generated by new units of capital is smaller than the breakeven
investment needed to avoid a fall in k(t). The direction of convergence towards k ∗ , which is
documented in Figure 2, implies that k ∗ is a stable steady state.

4.2 Long Run Behavior

Given initial conditions K(0), L(0), and A(0), the long run behavior of all other variables in the
economy can be characterized using the steady state k ∗ . First, let y ∗ = f (k ∗ ) be the steady state
level of output per efficiency unit of labor. Then, note that

• Capital: K(t) = k(t)A(t)L(t) → k ∗ A(0)L(0)e(g+n)t grows at rate g + n in the steady state.

• Output: Y (t) = f (k(t))A(t)L(t) = y(t)A(t)L(t) → y ∗ A(0)L(0)e(g+n)t grows at rate g + n in

the steady state.

• Investment: I(t) = sy(t)A(t)L(t) → sy ∗ A(0)L(0)e(g+n)t grows at rate g + n in the steady

state.

• Consumption: C(t) = (1 − s)y(t)A(t)L(t) → (1 − s)y ∗ A(0)L(0)e(g+n)t grows at rate g + n

in the steady state.
K(t)
• Capital per Capita: L(t) = k(t)A(t) → k ∗ A(0)egt grows at rate g in the steady state.

Y (t)
• Output per Capita: L(t) = y(t)A(t) → y ∗ A(0)egt grows at rate g in the steady state.

5
In the long run, increases in living standards YL(t)
(t)
are only generated by increases in technology,
and living standards grow at the same rate g as technology. This economy also exhibits balanced
growth, with the same growth in the long run for capital per capita and output per capita. We also
k∗
have K(t)
Y (t) → y ∗ converges to a constant in the long term.

5 Describing the Steady State & Parameter Changes

Let’s take a look at the equation defining the steady state

sf (k ∗ ) = (n + g + δ)k ∗ .

With Cobb-Douglas production, y = f (k) = k α , so k ∗ and the steady state level of output per
efficiency unit of labor y ∗ satisfy
  1   α
∗ s 1−α
∗ s 1−α
k = , y = .
n+g+δ n+g+δ

5.1 An Increase in the Savings Rate

To understand how a change in the savings rate affects the steady state, take logs to get
α
log y ∗ = [log s − log(n + g + δ)] .
1−α
∗
Therefore, ∂∂log y α
log s = 1−α > 0 is the elasticity of output with respect to savings in the long run. If
α = 31 , then 1−α
α
= 0.5. So a 1% increase in the savings rate leads to a 0.5% increase in output.
Consider an economy that starts in the steady state with k(t) = k ∗ for all t ≤ t̃. But then, at
y∗
time t̃, the savings rate increases from s to s̃ > s. The elasticity ∂∂log
log s is positive, so this change
in the savings rate must lead to a long run increase in output per efficiency unit of labor y ∗ → ỹ
where ỹ > y ∗ . Intuitively, the investment rate term sf (k(t)) in the top panel of Figure 1 shifts up,
and k̇(t) shifts up in the bottom panel. But this change implies that the level of output per capita,
or living standards, will be higher in the long run since

Y (t)
= y(t)A(t) → ỹA(t) > y ∗ A(t).
L(t)

In the short run, the growth of output per capita will increase to accommodate the higher level
of output per capita. However, the long run growth of YL(t) (t)
is still fixed at the growth rate g of
technology. The savings rate increase has level effects but does not have growth effects. The
pattern of output per capita over time will look something like the line plotted in Figure 3

5.2 Consumption and Savings

GDP per capita is a convenient measure of living standards in cross-country data, but economists
often prefer to define human welfare in terms of consumption. In this model, consumption satisfies

C(t) = Y (t) − I(t) = (1 − s)Y (t),

6
 Log GDP per Capita

log(Y(t) L(t))

~
t
Time t

Figure 3: An Increased Savings Rate at Time t̃

since households are able to consume any output which is not saved. Normalizing by effective
C(t)
units of labor c(t) = A(t)L(t) we have long run convergence to a steady state level c∗

c(t) = (1 − s)y(t) → c∗ = (1 − s)y ∗ . (8)

Equation (8) allows us to describe the impact of a change in the savings rate on consumption in
the long run. When s increases, there are two forces at work:
• Negative Direct Effect: A higher savings rate reduces consumption because households
save more and “eat less” as a fraction of output. This effect is embedded in the term 1 − s.
∗
• Positive Indirect Effect: A higher savings rate increases the level of output, since ∂y
∂s > 0.
Even if consumers are eating less as a fraction of output, the amount of output increases,
embedded in the term y ∗ above.
In general, the impact of increased savings on consumption in the long run is ambiguous. Figure
4 plots c∗ as a function of s, and the function displays a hump-shaped or non-monotonic pattern.
Note that
∂c∗ ∂y ∗
= −y ∗ + (1 − s) .
∂s ∂s
With Cobb-Douglas production, we have
 α
∂y ∗ α y∗

∗ s 1−α
y = → = .
n+g+δ ∂s 1−α s
Combining these formulas, write
∂c∗
 
α−s
= y∗.
∂s (1 − α)s

7
 Long Run Consumption

Consumption per Efficiency Unit of Labor c*

α
Savings Rate s

Figure 4: Consumption and Savings

So if s < α, increased savings lead to higher long run consumption levels. Intuitively, this is
because the marginal product of capital is high for low levels of saving and hence capital. Extra
units of capital increase output enough to compensate for the reduced consumption per unit of
output. When s > α, the marginal product of capital is lower and the impact of extra savings
and hence capital on output is smaller. The direct effect of savings dominates, and consumption
decreases with a rise in s. The intermediate level of savings s = α exactly balances these two
forces and results in the maximum level of consumption in the long run.