EC 502 Lecture 4: Implications of the Solow Model

Dongling Su*

1 Convergence
1.1 Solow Theory
The central transition equation for kt in the Solow model is
s 1−δ
kt+1 = f (kt ) + kt ,
(1 + g)(1 + n) (1 + g)(1 + n)

implying that the growth rate of kt from t to t+ is

 
kt+1 − kt s f (kt ) 1−δ
= + −1 .
kt (1 + g)(1 + n) kt (1 + g)(1 + n)

Since f (k)
k is decreasing in k, e.g. f (k)/k = k
α−1 for Cobb-Douglas production, the growth of

kt is higher when the initial level of kt is lower. This growth pattern for kt implies that over time,
countries in the Solow model converge.

• Poor countries grow faster until they reach the steady state. Intuitively, poor countries grow
faster because their marginal product of capital is higher, so new units of capital lead to
larger increases in output than in rich countries.

• Once a country reaches the steady state, output per capita grows at the rate g due to
technological improvements rather than capital accumulation.

• So the Solow theory predicts a negative correlation in the data between the level of living
standards and the growth of living standards.

Crucial Note: Convergence to the same living standards or output per capita, is not guaranteed
by the Solow model if countries have different steady states! In other words, if the characteristics
of an economy that determine their steady state k ∗ like production structures, savings rates, and
population growth rates are the same, then the two countries’ living standards should converge.
But countries that differ along these dimensions will converge to different steady states and hence
different levels of living standards. Practically, the Solow model’s convergence predictions depend
upon the following questions: “Do a given pair of countries differ just because of the current capital
stock, i.e. an initial condition? Or are there fundamental differences in their economic structures?”
* These notes borrow heavily from notes by Adam Guren, Simon Gilchrist, Francois Gourio, and Stephen
Terry.

1
1.2 Data
In the data, not all nations which were poor in 1960 had high subsequent growth, as shown in Fig-
ure 1. In other words, nations do not appear to unconditionally converge. However, restricted to
the set of developed nations (Figure 2) or to homogeneous economies such as US states (Figure
3), growth rates and income levels are negatively correlated. This is consistent with conditional
convergence, as predicted by the Solow model among economies with similar structures and
hence steady states.

2 Growth Accounting
2.1 From Production Functions to Growth Rates
Using the neoclassical production function embedded in the Solow model, we can decompose
growth in GDP into contributions from technology growth, growth in labor, and capital accumula-
tion. Let’s start with the Cobb-Douglas production function Yt = Ktα (At Lt )1−α . Note that

log Yt = α log Kt + (1 − α) log At + (1 − α) log Lt

log Yt−1 = α log Kt−1 + (1 − α) log At−1 + (1 − α) log Lt−1
→ ∆ log Yt = α∆ log Kt + (1 − α)∆ log At + (1 − α)∆ log Lt

Now recall that for small changes, the log difference is approximately equal to the percent change:
Xt − Xt−1
%∆Xt = ≈ ∆ log Xt = log Xt − log Xt−1 .
Xt−1
This approximation formula follows from a first-order Taylor expansion of the natural logarithm.
This formula is also the reason why ∂∂ log
log Y
X can be interpreted as the elasticity of Y with respect to
X, as we previously showed. But then from the equation for the log difference of Yt above, note

gY ≈ αgK + (1 − α)gL + (1 − α)gA (1)

In words, the growth rate of output is a weighted sum of the growth rates of capital, technology,
and labor.

2.2 Measuring Production Function Elasticities

Equation (1) involves the parameter α. We previously demonstrated that with a Cobb-Douglas
production function, α is the elasticity of output with respect to capital. But to measure this in
the data, we need to start from the optimality conditions for profit maximization. Recall that these
first-order conditions reflect choices for capital and labor which solve the following problem

max F (Kt , At Lt ) − Rt Kt − Wt Lt
Kt ,Lt

where Rt is the rental rate of capital and Wt is the wage rate. The optimality conditions or first-order
conditions are given by
∂F
= Rt → M P Kt = Rt
∂Kt

2
 ∂F
= Wt → M P Lt = Wt
∂Lt
With Cobb-Douglas production functions, the capital optimality condition can be rewritten
M P Kt = Rt
Yt
α = Rt
Kt
Rt Kt
α= ,
Yt
and the labor optimality condition can be rewritten
M P Lt = Wt
Yt
(1 − α) = Wt
Lt
W t Lt
(1 − α) = .
Yt
The quantity RYt K
t
t
is known as the capital share and WYttLt is known as the labor share. They
represent the share of output paid to the owners of capital and to laborers, respectively. The
formulas above demonstrate that with Cobb-Douglas production functions and profit maximization,
the capital elasticity α is equal to the capital share and the labor elasticity 1 − α is equal to the
labor share.1
In theory, we can measure α from either the capital share or the labor share in the data.
However, in practice it is easier to measure compensation than the return on capital, so economists
typically use the following formula:
Total Labor Compenstationt
α = 1 − Labor Sharet = 1 − .
GDPt
In US data, the labor share has historically been around 23 , as shown in Figure 4. This would imply
a value α ≈ 13 . The labor share has historically been fairly stable over time in the data, which is an
argument in favor of the Cobb-Douglas production function.

2.3 Growth Accounting Applications

Equation (1) is a powerful decomposition, and we can measure output growth, capital growth,
and labor growth in the data. However, technology growth gA is more difficult to measure. But
armed with an estimate of α from the labor share, technology growth is the only unknown quantity
in the growth decomposition. So we can compute all of the components of the Equation (1) to
decompose growth over some time period as follows:

gY = αgK + (1 − α)g + (1 − α)gA

|{z} |{z} | {z L} | {z }
GDP Growth Capital Contribution Labor Contribution Technology Contribution

The resulting measure of technology’s contribution to growth is known as Total Factor Pro-
ductivity (TFP) Growth or the Solow Residual. In the US data, a simple and very rough growth
accounting exercise results in the decomposition in Table 1.
1
Equality of input shares and input elasticities given profit maximization is a result which generalizes to
the Cobb-Douglas case with more than two inputs but does not necessarily hold for production functions
other than Cobb-Douglas.

3
 Table 1: Growth Accounting in the US
Percent GDP Growth Capital Contr. Labor Contr. Technology Contr.
1950-1975 3.57 1.35 0.73 1.50
1976-1990 3.56 1.06 1.38 1.12
1991-2011 2.42 0.81 0.41 1.19
Growth Fraction GDP Capital Labor Technology
1950-1975 1.00 0.38 0.20 0.42
1976-1990 1.00 0.30 0.39 0.32
1991-2011 1.00 0.34 0.17 0.49

Note: The top panel reports the individual components of Equation (1) in the US in three separate periods in percentages. GDP
growth is the mean annual percentage growth in real GDP over the indicated period. The capital contribution is equal to one minus
the labor share times the mean annual percentage growth of the real capital stock. The labor contribution is equal to the labor share
times the mean annual percentage growth in total hours worked. The technology contribution is equal to the residual, i.e. GDP growth
minus the capital contribution minus the labor contribution. In the bottom panel, each contribution is expressed relative to GDP growth
as a fraction. The data is from the Penn World Tables version 8.1, and the labor share is the mean value of the labor share in the US
from 1950-2011, 64.9%.

In the US in the postwar period 1950-1975, capital accumulation and TFP or technology
growth both contributed around 40% to overall growth. Starting in the 1970s, a productivity
slowdown occurred and technology’s contribution fell to around a third of overall growth.
Following the IT revolution in the 1990s, technology led overall growth with a contribution
of around half.
It is apparent from the table above that in the US the growth decomposition is not
always constant. Unsurprisingly, growth accounting doesn’t always yield the same results
across countries/regions either. In a paper called “The Tyranny of Numbers: Confronting
the Statistical Realities of the East Asian Growth Experience” in the Quarterly Journal of
Economics in 1995, Alwyn Young argued that the Asian Growth Miracles of Hong Kong,
Singapore, South Korea, and Taiwan were primarily accounted for by capital accumulation
rather than technological growth.

3 Factor Prices and the Solow Model

When we first introduced the Solow model, we did not discuss markets. In a “fuller”
version of the Solow model, there is an explicit labor market, a capital market, and an
investment or savings market, and all actors are assumed to behave optimally in these
markets. It turns out that, as we did above, you can ignore them if you only want to
describe quantities given exogenous savings behavior. However, our discussion of profit
maximization by firms gives us some insights into these markets already. The optimality
conditions for profit maximization are given by

M P Kt = Rt

M P L t = Wt .
In the labor market, the market-clearing wage is equal to the marginal product of labor of
firms. In the capital market, the rental rate of capital is equal to the marginal product of
capital.

4
3.1 The Rental Rate of Capital, Interest Rates, & the Savings Market
The labor market and the wage seem straightforward to understand. But what is the rental
rate of capital and how does it interact with capital and savings markets?
• Rt is the cost to rent one unit of capital for one period from the owners of capital.

• When a firm rents a one unit of capital, they pay the capital owner the rental rate
Rt . But the capital depreciates at rate δ, so the owner only receives 1 − δ units of
capital back. So the net return to capital owners is given by rt = Rt − δ. We call rt
the interest rate.
In the full model, the savings market is the market where individuals invest in capital with
their own savings and therefore become the owners of capital. They receive interest on
their savings, which is linked to the capital market through the formulas above.

3.2 Open Economies & Capital Flows

So far all of our analysis has been limited to the closed-economy case. But what if the
economy is open?

• We saw above that the optimality condition for capital is given by M P Kt = Rt , where
rt = Rt − δ is the interest rate.

• Assume that developed economies are in steady state. Imagine that a small open
economy with a low capital stock – which takes the world interest rate as given –
opens up to the rest of the world and allows capital and savings to flow inwards.
Initially, firms in that country will borrow to use capital and savings from abroad
since the local interest rate is much higher than the world interest rate, a fact which
follows from a low level of capital and a high marginal product of capital relative to
rest of the world.

• In other words, small economies that open up should receive a large and sudden
infusion of capital from the rest of the world. They will accumulate capital quickly to
the point where the domestic marginal product of capital equals the world interest
rate plus depreciation r + δ

In the data, capital does not seem to flow systematically to poor countries, which
seems a bit puzzling. Several explanations have been suggested:

• Adjustment costs on capital limiting the quick buildup of capital in poor countries.

• Mistrust of developing country firms by developed country lenders.

• Lower technology levels in poor countries. If At is lower in a poor country, then

M P Kt = r + δ may hold even for small levels of capital. But then another puzzle
suggests itself: why aren’t poor countries adopting better technology?

5
 Figure 1: No Unconditional Convergence
KOR
●

Real GDP per Capita Growth, 1960−2011, %

6
TWN
●

EGY ROU
● THA ●
CHN
●
● MYS JPN

4
TUN
●ARG ●
● PRT
GRC ESP
● ●
BRA ● ● ITA
IDN ● IRN AUT
BEL
DEUNLD
MAR
IND ● TUR
●
●
●
●
● PER
DOM CHL ●● ●
FRASWE
ECU ● ●● ISR
PAK ● ● ●CAN
●GBR AUS
● ● ● USA
●
● PHLSYR LKA COL MEX CHE
MOZ MLI
2 ● ●
ETH BOL
●
● ●
●
●
●
●BFANPL CMR ● GTM● ZAF
MWI● ●
TZABEN ●
NGA ●
●
● UGA HND
● VEN
RWA
●
● TCD
KEN
BGD ●
● ● ZMB
● ●
BDI ●
CIV
●
● ● GHA
● ●
MDG SEN
0

●
●

GIN
NER ●
●
−2

COD
●

6 7 8 9
GDP per Capita, 1960, Log

Note: The data is from the Penn World Tables version 8.1 and includes a total of 71 nations/regions. Real GDP per capita is the natural
logarithm of the ratio of expenditures-based GDP in 1960, measured at PPP-adjusted constant 2005 US dollars, to the population in
2011. The growth rate is the annualized percentage change in the same GDP per capita measure from 1960 to 2011.

Figure 2: Conditional Convergence among Developed Nations/Regions

KOR
●
Real GDP per Capita Growth, 1960−2011, %
6

TWN
●
5

MYS
JPN
4

●
●
ARG
● PRT
● GRC ESP
● ●
ITA
IRN ● AUT
BEL
3

● ● NLD
DEU
● ● ●
TUR
CHL
●● FRASWE
ISR ● ●
● CAN
GBR AUS
● ● ●
USA
● CHE
2

7.0 7.5 8.0 8.5 9.0 9.5

GDP per Capita, 1960, Log

Note: The data is from the Penn World Tables version 8.1 and includes a total of 23 nations/regions. Real GDP per capita is the natural
logarithm of the ratio of expenditures-based GDP in 1960, measured at PPP-adjusted constant 2005 US dollars, to the population in
2011. The growth rate is the annualized percentage change in the same GDP per capita measure from 1960 to 2011. The sample
consists of the top third of nations in 2011 as measured by GDP per capita.

6
 Figure 3: Conditional Convergence among US States

Note: This figure is from Ganong and Shoag (2015) “Why Has Regional Income Convergence in the US Declined?,” with underlying
data from the Bureau of Economic Analysis.

7
 Figure 4: Stable Labor Share in the US
100
80
US Labor Share, % of GDP
60
40
20
0

1950 1960 1970 1980 1990 2000 2010

Year

Note: The data is from the Penn World Tables version 8.1. The labor share is total labor compensation as a percent of US GDP.