Full text of Business Review (Federal Reserve Bank of Philadelphia) : Second Quarter 2008

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

Why Are Goods So Cheap
in Some Countries?*
BY GEORGE ALESSANDRIA AND JOSEPH KABOSKI

L

ooking around the world, we observe
substantial differences across countries in
prices for most goods. These price differences
also tend to be positively correlated with
income differences, so that citizens of high-income
countries tend to pay more for the same goods than
citizens in low-income countries. In this article, George
Alessandria and Joseph Kaboski summarize some of the
evidence related to the big price differences across
countries for a broad set of goods. They then discuss the
relationship between prices and income levels and some
possible explanations for that relationship.

Lovers of Big Macs will find China
to be a true paradise and Switzerland
quite the opposite, since the money
spent to buy one Big Mac in Switzerland will get you almost four Big Macs
in China.1 These big international
price differences haven’t led Swiss Big
Mac lovers to move to Beijing. In fact,

1

Big Mac™ is a registered trademark of the
McDonald’s Corporation.

George Alessandria
is a senior economist
in the Research
Department of
the Philadelphia
Fed. This article
is available free
of charge at www.
philadelphiafed.org/econ/br/.
www.philadelphiafed.org

despite a much higher price, based on
annual income data for 2005, the average Swiss citizen earned enough to eat
eight times as many Big Macs as the
average Chinese citizen.2
These differences in prices and
purchasing power extend beyond just
Switzerland and China and Big Macs.
In fact, when we look across the world,
we find substantial differences across
countries in prices for a broad range
of goods. These price differences also
tend to be positively correlated with
income differences so that citizens
of high-income countries tend to pay
more for the same goods than citizens
of low-income countries.

*The views expressed here are those of the authors and do not necessarily represent the views
of the Federal Reserve Bank of Philadelphia or
the Federal Reserve System.

In this article, we will summarize
some of the evidence of the big price
differences across countries for a broad
set of goods. We will then discuss
the relationship between prices and
income levels. Finally, we’ll discuss
some possible explanations for this
relationship.
MAKING INTERNATIONAL
PRICE COMPARISONS
Comparing prices across countries
can be difficult because prices are
typically quoted in different currencies.
For instance, to compare the yuan
price of a Big Mac in China with the
franc price in Switzerland, we need
to use the nominal exchange rate
between the yuan and the Swiss franc
to convert the prices into a common
currency. Movements in the nominal
exchange rate3 over time can thus lead
Swiss Big Macs to become relatively
more or less expensive compared with
Big Macs in China. We will ignore
the short- to medium-run fluctuations

2

Based on 2005 data on gross national income
taken from the World Development Indicators:
China $1,700 and Switzerland $54,930 (U.S.
dollars).

3

The nominal exchange rate is the value of one
country’s currency in terms of another country’s
currency.

Joseph Kaboski
is an assistant
professor in the
Department of
Economics at
the Ohio State
University.

Business Review Q2 2008 1

related to exchange rates and instead
focus on long-run differences in prices
across countries.
Even though we’ve already seen
otherwise, a natural expectation is
that the price of a Big Mac should be
the same everywhere; after all, it is the
same good.4 This idea is known as the
law of one price (LOP). More formally,
the LOP states that once prices are
converted to a common currency, the
same good should sell for the same
price everywhere, provided there are
no barriers to trade and markets are
competitive.
The basic idea behind the LOP
is that if prices differ across locations,
firms can make some profits by buying
in the low-price place and selling in
the high-price place. This activity,
which is called arbitrage, will continue
until prices are similar in the two locations.
While the LOP is described as a
“law,” it does not hold for all goods.
Gold and Big Macs provide evidence of
its respective successes and failures as a
description of world prices. The prices
of Big Macs across countries reported
in Table 1 provide a clear example of
its failure. When converted into U.S.
dollars, Big Macs sell for up to 65 percent more than in the U.S. and down
to 57 percent less than in the U.S. On
the other hand, from Table 2, which
reports the price of one troy ounce
of gold quoted on the same day at
nearly the same moment on different
exchanges throughout the world, we
see that the LOP seems to hold, since
the price of gold ranges in a 3 percent
band around the price in the U.S.
One important reason the LOP
does not hold is that there are barri-

4

There are some minor differences in the size
and condiments across countries. The biggest
difference is in India, where Hindu and Islamic
Sharia dietary laws prohibit eating beef, and
so the Big Mac is made with two all-chicken
patties.

2 Q2 2008 Business Review

TABLE 1
Some Big Mac Prices
Big Mac Price

Actual Exchange Rate

Country

in Local
Currency

Switzerland

SFr6.30

5.12

1.23

$3.10

3.10

1.00

Yuan10.50

1.31

8.00

U.S.
China

in U.S.
Dollars

1 USD=

Based on Big Mac prices and exchange rates as of March 25, 2006. The Big Mac index is published periodically by The Economist. Go to http://www.economist.com/markets/bigmac/ to find
more information about Big Mac prices and exchange rates across many countries.

TABLE 2
Gold Prices Around the World
Time
(Eastern Standard Time)

USD/Troy Ounce

United States

10:28

$625.01

Australia

10:28

$625.00

9:53

$617.71

10:28

$625.51

6:07

$634.89

10:21

$622.75

Luxembourg

5:58

$624.50

Hong Kong

0:51

$623.40

Exchange

Brazil
Switzerland
India
United Kingdom

Prices were downloaded from Bloomberg on November 3, 2006.

ers that make international trade, and
thus arbitrage, costly.5 These barriers can be man-made, such as tariffs,

5

For a detailed breakdown of the costs to
trading goods across countries, see the Business
Review article by Edith Ostapik and Kei-Mu Yi.

taxes, or trade restrictions, or physical,
such as distance, which incurs shipping
costs. The costs of these barriers differ
quite a bit across goods. For instance,
shipping costs primarily depend on
the distance, weight, and mode of
transportation. For goods such as gold,

www.philadelphiafed.org

which have a high value to weight ratio, shipping costs are fairly minor. For
Big Macs, which, based on U.S. prices,
are 1/1400 as valuable per ounce as
gold and don’t travel particularly well,
shipping costs are relatively large.6
However, even though it’s expensive to
ship a Big Mac, Big Mac prices might
be the same in different countries if
the inputs to producing it are very easy
to trade. This is essentially true for
the beef and special sauce, but it’s not
true for the workers who fry it up or
the building in which it is consumed.
For some goods, such as buildings or
haircuts, the shipping costs are so high
that they are almost never traded.
Economists call these goods nontraded
goods.
Another reason prices may differ
across countries is that the competitive
environments may differ. For instance,
in some countries, there may not be
many close substitutes for a Big Mac,
and so Big Macs might be relatively
expensive. However, in countries with
lots of low-cost alternatives, Big Macs
might cost relatively less. Or it might
be the case that people in some countries are just willing to pay more for
certain goods. Firms take advantage of
these differences in willingness to pay
for certain goods by charging different prices across countries. Charging
different people different prices for the
same good is known as price discrimination, and it is a common practice
in many industries.7 To make this
strategy effective, firms make arbitrage

6

A Big Mac weighs 7.5 oz. (www.mcdonalds.
com/app_controller.nutrition.index1.html)
and 1 oz. equals 0.9 troy ounces. So a Big Mac
weighs 6.75 troy ounces. Based on a U.S. price
of $3.10, a Big Mac costs $0.46 per troy ounce
compared with gold, which costs $625.01 per
troy ounce.

difficult by changing their product
slightly across countries. For instance,
film studios embed region codes on
their DVDs so that they work only on
DVD players in particular parts of the
world.8 Similarly, makers of cameras,
electronics, and cars often won’t honor
warranties of products purchased in a
different country.
A Broader Test: Comparing
the Price of a Basket of Goods. As
we have already discussed, prices of
individual goods may not be equated
across countries for many reasons. We

Charging different people different prices
for the same good is known as price
discrimination, and it is a common practice
in many industries.
would like to know if these deviations
from the LOP are systematic. One way
to do this is to see if these individual
price differences wash out when we
buy a broad basket of goods. But what
basket should we compare? In the U.S.
the consumer price index measures the
price of the basket of goods the typical
U.S. consumer purchases. Similarly,
many countries measure the price of a
basket of goods that their consumers
purchase.
There are two problems with
comparing these price indexes across
countries. First, they are indexes, so
their level is not meaningful, and
therefore, we can talk only about how
prices change over time relative to
one another. Second, countries do
not sample the same basket of goods,
so we are comparing the prices of different baskets of goods, making price

7

For example, by allowing children to fly
for half price, airlines are engaging in price
discrimination.

www.philadelphiafed.org

comparisons meaningless. Fortunately,
there is a way around these problems.
The International Comparison
Program (ICP) and the Penn World
Tables (PWT) collect data that allow
us to compare prices and income
across countries. The ICP is a series of
statistical surveys that collect prices
on a representative sample of approximately 3000 goods and services. These
surveys are conducted in many countries and are very careful to sample the
price of very similar goods. Surveys are
large projects involving each country’s

8

This also occurs with video games and video
consoles.

national statistical agency and are
coordinated by the World Bank and
the Organization for Economic Cooperation and Development (OECD).9
The last survey took place from 1993
to 1996 and involved 117 countries,
and it provides a useful starting point
for analyzing prices and income across
countries.10
Measuring Prices and Income.
Based on the prices collected in each
country, it is possible to come up
with a world price for each good as
a weighted average of all the prices
in the world. For each country, real

9

The World Bank is an international organization that provides financial and technical assistance to developing countries. The OECD is a
group of 30 countries committed to democracy
and market economies, and this organization
collects and publishes a range of economic and
social statistics.

10
For a brief overview of the ICP, refer to the
article by Sultan Ahmed. A new survey is
underway with almost 150 countries.

Business Review Q2 2008 3

income is then calculated as the value
of the goods purchased at world prices.
Because each country’s income is
measured using the same prices, these
measures of income are directly comparable across countries. The value
of each country’s purchases is then
calculated at its own prices; this is a
measure of its income at local prices.
The ratio of income at local prices to
income at world prices is a measure of
a country’s price level relative to world
prices. In this way the Penn World
Tables construct a measure of the price
level and purchasing power (real income in each country). The procedures
for measuring income and prices across
countries are quite similar to how the
Bureau of Economic Analysis measures
income and prices in the U.S. over
time.
Figure 1 presents a scatter plot
of the relative price of the common
basket of goods (on the y-axis) against
the relative income of each country
(on the x-axis). These data are from
the 1996 Penn World Tables, and
each point is relative to the U.S. and
measured in logarithms, which means
that the slope approximates the percentage change in the price level for a
given percentage change in per capita
GDP. There are obviously substantial
differences in price level and income
per capita. Turkmenistan has the lowest prices (-2.18, or 11 percent of the
U.S. level), while Switzerland has the
highest prices (0.53, or 170 percent of
the U.S. price level). Tanzania has
the lowest income per capita (-4.12,
or 1.6 percent of the U.S. level), and
Luxembourg has the highest income
per capita (0.18, or 120 percent of the
U.S. level).
From Figure 1 we see that there
is a positive relationship between
prices and income. As we saw with Big
Macs and Switzerland and China, the
countries with the highest income also
pay the highest prices for a broad range

4 Q2 2008 Business Review

of goods. A measure of the strength of
this relation can be found by estimating how much relative prices increase
with relative income. The results of
this estimate are reported in the lower
right corner of Figure 1. We find that a
doubling of income per capita is associated with a 43 percent increase in the
price level.
The differences in price levels and
income per capita are quite persistent
over time. For instance, of the 32
countries with price levels one-half of
those in the U.S. in 1996 for which we
also have data on price levels in 1985,
26 also had price levels less than half
of those in the U.S. in 1985.
EXPLAINING THE PRICEINCOME RELATIONSHIP
Economists tend to attribute
the price-income relationship seen in
Figure 1 to either differences in the
prices of tradables (those goods that
are either traded frequently or easy
to trade) or differences in the prices
of nontradables, those goods that are
both costly and infrequently traded
across countries.11 We will discuss
an explanation for the price-income
relationship based on deviations from
the LOP in tradables. (An alternative,
complementary explanation based on
deviations from the LOP in nontradables is presented in Another Theory to
Explain the High Prices in High-Income
Countries.)
Examining the role of prices for
tradables for the relationship seen in
Figure 1 requires a measure of the
price of tradable goods. Fortunately,
the ICP contains prices for over 3000
goods, so we can compare the price of
a basket of those goods that are traded
frequently across countries. Examples
of the types of goods classified as

tradable are machinery and equipment, tobacco, alcohol, and personal
transportation equipment.
Figure 2 shows the relationship between the price of a basket of tradable
goods and income per capita. Similar
to what we saw in Figure 1, there is a
positive relationship between the price
of this tradable basket and income. In
the lower right-hand corner of Figure
2, we estimate that a doubling of income per capita is associated with a 26
percent increase in the price of tradable goods. Comparing our measures
of price differences of tradable goods
to the measure of price differences for
all goods, we find that differences in
prices for tradables account for about
60 percent of the aggregate priceincome relationship.12
One possible explanation for the
positive relationship between prices for
tradables and income is that prices for
tradable goods include some nontradable inputs, which are cheaper in
low-income countries. For instance,
the price of a car includes the cost
of transporting it from the factory to
the dealership as well as the costs the
car dealership incurs in selling the
car. The costs in getting products to
the consumer, essentially retail and
wholesale distribution, are mostly
nontradable and contribute to price
differences across countries. If retail
and wholesale distribution services are
cheaper in low-income countries,13 that

12
In our paper, we show that for certain commonly used price indexes, the contribution
of differences in the prices of tradables to the
relationship between price levels and income
can be measured by comparing the coefficient
from the regression of prices for tradables on
income to the coefficient from the regression of
price levels on income.
13

11

Much of this section is based on our working
paper.

Obviously, wholesale and retail distribution
also includes some tradable inputs, such as
trucks, airplanes, and fuel, which would tend to
make their prices similar across countries.

www.philadelphiafed.org

FIGURE 1
Prices and GDP Per Capita
Log Price Level (relative to U.S.)
1
Switzerland

0.5
Luxembourg

0
United
States

-0.5
Tanzania

-1
-1.5
-2
Turkmenistan

y = 0.43x - 0.02
2
R = 0.49

-2.5
-4.5

-4

-3.5

-3

-2.5

-2

-1.5

-1

-0.5

0

0.5

Log GDP Per Capita (relative to U.S.)

FIGURE 2
Price of Tradables and GDP Per Capita
Log Price of Tradables (relative to U.S.)

may explain why prices for tradables
are lower in low-income countries.
To isolate the source of differences
in prices for tradables, we must compare the price of goods before these
retail and wholesale distribution services are added. One way of doing this
is to measure the price of goods as they
leave the U.S. and are being shipped to
different destination markets.
Measuring U.S. Export Prices
at the Border. Destination-specific
export prices can be constructed using
data collected from shippers’ export
declaration forms. These are forms
filed with Customs for every shipment
of goods that leaves the U.S.14 For each
good, there are data, by destination
country, on the average price of all
shipments in each year from 1989 to
2000. These prices are measured at the
U.S. border or the shipping dock before
any taxes or nontradable services are
added. Goods are classified according to the Harmonized Commodity
Description and Coding System (HS).
This is a system of names and numbers
for classifying traded products.15 The
data cover 10,471 goods.
We focus on shipments to OECD
countries plus some low-income
countries for which we also have wage
data. The complete list of countries

1
Switzerland

0.5

Luxembourg

0

United
States

Tanzania

-0.5
-1
Turkmenistan

-1.5
-2

y = 0.26x + 0.20
R2 = 0.34

-2.5
-4.5

-4

-3.5

-3

-2.5

-2

-1.5

-1

Log GDP Per Capita (relative to U.S.)

www.philadelphiafed.org

-0.5

0

0.5

14

These forms aren’t necessary for shipments
with values below $2000. These small-value
shipments account for a very small share of U.S.
exports.

15
The HS system is an international classification system based on broad six-digit categories.
Many countries classify traded goods in more
detail. For example, the U.S. defines products
using 10-digit HS codes. Export codes (which
the U.S. calls Schedule B) are administered by
the U.S. Census Bureau. In this system, 10-digit
goods can be incorporated into nine-digit
goods, nine digits into eight digits, and so on.
The U.S. Census Bureau’s website offers this
example: Concentrated frozen apple juice is
assigned a 10-digit number, but this product can
be included in the broader six-digit category described as apple juice, which, in turn, can be incorporated into the broader four-digit category,
fruit juices and vegetable juices, and so on.

Business Review Q2 2008 5

Another Theory to Explain the High Prices in High-Income Countries
Second, it assumes that across countries, there are much
larger differences in the productivity of workers producing
tradable goods than nontradable goods. Since the LOP
holds for tradable goods, the cost of producing tradables
must be the same everywhere. This means that international wage differences are determined by differences in
labor productivity in traded goods and are quite large.
With large wage differences across countries and relatively
small differences in labor productivity in nontradables,
prices for nontradables will differ substantially across
countries and will be higher in high-wage/high-income
countries.
A simple two-country, two-goods example might
help to explain how the theory works. Suppose the two
countries, call them Richland and Poorland, can make
cars, which can be freely traded, and haircuts, which are
impossible to trade. The table below describes the productivity of workers in each country. Starting with case
1, we see that in Poorland, one worker can produce either
one car or one haircut per day, while the typical worker
in Richland is more productive and can produce either
four cars or two haircuts per day. To keep things simple,
suppose that workers in both countries get paid in dollars
and that the daily wage in Poorland is $1.
Given that a worker in Poorland earns $1 per day and
can produce one car per day, the price of a car must be $1
everywhere, since cars can be freely traded. Now, since
Richland workers can produce four cars a day, they will
earn $4 per day. With these wages, the price of haircuts

number of explanations of the aggregate
price-income relationship attribute it
to deviations from the law of one price
(LOP) in nontraded goods. Recall that
nontraded goods are those goods that
have high international shipping costs
and thus are infrequently traded across countries; such
goods include haircuts, restaurant meals, housing, and
medical services.
To get an idea of how the price of nontraded goods
differs with income, we plot in the Figure the relative
price of nontraded goods to traded goods against real income per capita. By looking at how the ratio of nontraded
to traded prices differs with income, we can isolate anything that affects nontradables separately from tradables.
As we saw with the relationship between aggregate price
levels and incomes, we find that the ratio of nontradable prices to tradable prices also rises with income. In
fact, a doubling of income is associated with a 34 percent
increase in the relative price of nontradables.
There are a variety of competing explanations of this
observation. The most common is known as the BalassaSamuelson theory.a It contains two main elements. First,
the theory assumes that the LOP holds in tradables.

A

a

For alternative explanations of this observation, see the work by William Baumol and William Bowen; Irving Kravis and Robert Lipsey;
and Jagdish Bhagwati.

TABLE
A Two-Country Example of the Balassa-Samuelson Model
# of units produced
per worker
Case 1

Prices

Price
Level

Real
Wage

Cars

Haircuts

Wages

Cars

Haircuts

Poorland

1

1

1

1

1

2

0.50

Richland

4

2

4

1

2

3

1.33

Poorland

1

0.5

1

1

2

3

0.33

Richland

4

2

4

1

2

3

1.33

Case 2

6 Q2 2008 Business Review

www.philadelphiafed.org

Another Theory ... continued
will be $1 in Poorland, since a worker earning $1 can
able goods compared to nontradable goods across rich
give one per day, while in Richland a haircut will cost $2,
and poor countries. While we don’t have good measures
since it takes a worker earning $4 a half a day.
of these productivity differences across countries, we do
To see how prices vary with real income, we must dehave good measures from the U.S. Using data on secfine the bundle of consumption goods. Let’s suppose that
toral labor productivity growth in the U.S. from a paper
the typical basket of goods is composed of one car and
by Dale Jorgenson and Kevin Stiroh, we find that labor
one haircut. Given the prices for individual goods, this
productivity in the nontradables sector has grown by
basket will cost $2 in Poorland and $3 in Richland. We
about two-thirds as much as labor productivity in the
can use these prices to get a measure of real wages in each
tradables sector. Finally, for nontradable goods to explain
country as the wage divided by the price. So notice that
the aggregate price-income relationship, the nontradables
real income is 50 cents in Poorland and $1.33 in Richprice-income relationship must be much stronger than the
land. Clearly, then, the higher price country, Richland,
aggregate price-income relationship.b Comparing Figure
also has a higher real income, as in the data.
1 in the text and the figure in this box, we see that this is
To see how prices and income depend on productivnot the case.
ity in each sector, let’s look at case 2 in the table. In this
b
case, workers in Poorland are one-quarter as productive
Notice from our first case that while the price of haircuts is twice as
high in Richland, the price level is only 50 percent higher. The weaker
as workers in Richland for both goods. A Poorland worker
relationship between aggregate prices and income is due to nontradables’
still gets a daily wage of $1 and produces a car a day, so
accounting for only a part of the final basket of goods. In their paper,
the price of a car is $1 and the price is $1 everywhere,
Alan Stockman and Linda Tesar measure the size of the tradables
sector in OECD economies and find that it accounts for about one-half
since cars are freely traded. The price of a haircut will
of the economy. This implies that the relationship between prices for
be $2, since it now takes two days to produce a haircut.
nontradables and income per capita needs to be twice the relationship
between the full basket of goods and income to explain the data.
In this case, the price level in Poorland will rise to $3,
and the real wage will fall to 33 cents,
while it is the same as case 1 in Richland. The price level is now the same
FIGURE
across the two countries, and there is
no positive relationship between prices
Relative Price of Nontradables and GDP
and income. Thus, to get a positive rePer Capita
lationship between prices and income,
Log PNT/PT (relative to U.S.)
it is necessary for low-income countries
1
to be relatively productive in producing
Switzerland
0.5
nontradables compared to high-income
United
States
countries.
0
There are three reasons to quesLuxembourg
-0.5
tion the Balassa-Samuelson theory as a
complete explanation of the aggregate
-1
price-income relationship. First, as
Tanzania
-1.5
we have seen, there are large deviaTurkmenistan
tions from the LOP in tradable goods.
y = 0.34x - 0.24
-2
R = 0.36
Second, the Balassa-Samuelson theory
-2.5
requires relatively large differences
-1
-4.5
-4
-3.5
-3
-2.5
-2
-1.5
-0.5
0
0.5
Log
GDP
Per
Capita
(relative
to
U.S.)
in the efficiency of producing trad2

www.philadelphiafed.org

Business Review Q2 2008 7

can be found at the bottom of Table 3.
Overall, there are almost 1.2 million
observations, where an observation is
a particular good sold to a country in
a particular year, accounting for about
75 percent of the value of U.S. trade in
goods over the period.
We can use these data to ask
whether, on average, goods being
shipped to markets with relatively high
income tend to be sold for relatively
high prices (a description of the empirical specification can be found in
the footnotes to Table 3).16 The results
of our analysis using these data on
export prices confirm what we found
using retail prices for tradables from
the Penn World Tables: Prices for
tradables increase as income per capita
increases. Moreover, in export prices,
this effect is about two-thirds as strong
as that for retail prices for tradables in
the Penn World Tables. This finding
suggests that differences in the factory
prices of tradables account for about
40 percent of the differences in retail
price levels across countries, while
wholesale and retail margins account
for about 20 percent.17
For the most part, then, the
evidence points to retail prices for
tradables being higher in high-income
countries because exporters sell these
goods at higher prices in these countries.
Digging a Little Deeper into Export Prices. Even though we perform
the analysis using data that have been

16

Here we are looking at prices for individual
goods rather than baskets of goods. Since not
all goods are exported to all countries, we cannot construct a representative basket as in the
previous analysis.

17

Recall that differences in the prices of tradables account for 60 percent of the difference in
price levels. Since differences in export prices
account for two-thirds of the differences in
prices for tradables, we can conclude that differences in export prices account for 40 percent of
the difference in price levels.

8 Q2 2008 Business Review

broken down into subcategories, one
might suspect that the price differences uncovered may be related to differences in the quality of the products
being sold. For instance, it could be
that a 10-digit category contains different quality goods, say, a high-quality
11-digit good and a low-quality 11-digit
good, and that high-income countries
purchase relatively more of the highquality good. While this idea can’t be
directly tested for goods classified at
the 10- and 11-digit levels, we can see

wages on export prices, in the last two
columns of Table 3, we find that wages
explain export prices by destination.
This leads us to conclude that high
prices are associated with high wages.18
Since this is a big data set, we can
dig a little deeper. We next examine
the relationship between export prices
and destination characteristics for
different types of goods. This analysis
can be found in the bottom seven
rows of Table 3. We find that export
prices increase more with wages for

For the most part, the evidence points to
retail prices for tradables being higher in highincome countries because exporters sell these
goods at higher prices in these countries.
if this is happening at broader levels
of classification. For instance, we can
compare the price-income relationship
on 10-digit goods to the same goods
classified at the nine-digit level. If rich
countries purchase relatively more of
the high-quality, more expensive 10digit goods, we should find that the relationship between prices and income
is stronger at the nine-digit level. We
actually find the opposite and conclude that quality differences do not
explain the differences in export prices
by destination.
To get at the source of international price differences, we next
examine the association between
export prices and the real wage in the
destination market. Not surprisingly,
since high-income countries also tend
to have high wages, we find a strong
positive relationship with wages in
destination countries (see the column
labeled “Wages only”). However, wages
and income per capita are not perfectly
correlated, since there are differences
in labor force participation, hours
worked, capital income, and taxes
across countries. When we examine
the independent effect of income and

consumption goods than for capital
goods, industrial supplies, autos, and a
range of other products. We also find
that the price of medicinal products
tends to be most affected by the wage
and income in the destination market.
Finally, notice that when we control
for wages and income per capita in
the final two columns, for each type
of good we find that wages are always
positively associated with prices, while
income per capita may have a negative
or positive association with prices.
The analysis of export prices tells
us three things. First, high prices for
tradables are largely due to exporters’
charging high prices as goods leave
the country. Second, export prices are
more strongly related to wages in the
destination market than income per
capita. Third, this effect is stronger for
consumer goods than industrial supplies or capital goods.

18

These results hold even after controlling for
a wide range of factors such as trade costs, the
share of intra-firm trade, and the level of intellectual property protection.

www.philadelphiafed.org

TABLE 3
Export Prices and GDP Per Capita and Wages*
(t-statistics in parenthesis)
Number
of Obs.

Fraction of
Total Value
Exported

GDP Per
Capita Only

Wages Only

Both Together
GDP Per
Wages
Capita

1,177,803

0.751

0.170
(64.3)

0.162
(86.2)

0.012
(3.25)

0.156
(57.5)

Consumption Goods

228,074

0.085

0.236
(39.6)

0.218
(51.9)

0.036
(4.3)

0.200
(33.8)

Food/Feed/Beverages

109,646

0.078

0.156
(31.7)

0.091
(26.5)

0.128
(18.2)

0.027
(5.6)

Capital Goods

322,105

0.248

0.087
(14.5)

0.146
(33.5)

-0.126
(14.4)

0.213
(33.4)

Industrial Supplies

484,661

0.247

0.201
(52.9)

0.168
(63.4)

0.063
(11.4)

0.136
(34.9)

Autos

25,694

0.082

0.158
(8.84)

0.113
(8.9)

0.090
(3.5)

0.066
(3.6)

Agricultural Goods

61,991

0.044

0.140
(21.1)

0.077
(16.7)

0.128
(13.0)

0.012
(1.7)

Medicine

15,859

0.014

0.187
(6.4)

0.282
(13.2)

-0.201
(4.7)

0.390
(12.4)

All Goods**

* Income per capita and wages are measured in real terms using price deflators in the Penn World Tables.
The table reports the relationship between export prices and the characteristics of the export destination from a regression of export prices on the
characteristics of the export destination. The regression takes the form: pijt = ait + b1*yjt + b2*wjt+ eijt, where pijt measures the logarithm of the
price of good i sold to country j at time t. In country j at time t, income per capita is measured as yjt, and the hourly manufacturing wage is measured as
wjt. The term eijt accounts for errors. The term ait is a dummy variable that accounts for good-specific attributes, such as marginal cost. To explore the
relationship between destination prices and just income per capita (or wages), we can run the regression constraining b2=0 (b1=0). The equation can be
estimated by ordinary least squares. We construct White robust standard errors that allow for heteroskedasticity in eijt and also allow for country-year
clustering.
** Countries include Australia, Austria, Belgium-Luxembourg, Brazil, Canada, Denmark, Finland, France, Germany, Greece, Hong Kong, Ireland,
Israel, Italy, Japan, South Korea, Mexico, the Netherlands, New Zealand, Norway, Portugal, Singapore, Spain, Sri Lanka, Sweden, Switzerland, and the
total U.S. exports of these goods.

Explaining the Export PriceIncome Relationship. The export
data confirm that exporters ship goods
at lower prices to low-income locations. As we have already discussed,
this type of price discrimination by
destination market is possible only if
trade barriers make it difficult for other

www.philadelphiafed.org

firms to arbitrage these destinationspecific prices away. Given that we see
these price differences, it must be the
case that arbitrage is limited, so prices
are determined by either differences in
competitive environments or consumers’ tastes for particular goods, or a
combination of the two.

In a recent paper, we develop a
theory of price discrimination that
can generate a positive relationship
between export prices and wages.
It builds on our second and third
findings from studying export prices:
Wages seem to matter most for export
prices, and export prices increase with

Business Review Q2 2008 9

wages more for consumer goods. The
key idea of the theory is that searching to find goods at a really low price
doesn’t pay. We mean this literally.
Since searching takes time away from
working, the higher one’s wage, the
more costly it is to search repeatedly
to find a good at a lower price. For this
reason, high-wage individuals will be
willing to accept higher prices than
low-wage individuals and will pay
more, on average.
To be a little clearer, our theory
assumes that consumers do not know
where to buy goods at the lowest price
and must spend some time searching
for goods. This is a theory of a cost
that limits arbitrage, the time it takes
to search, and is consistent with everyone’s experience of finding the same
good selling for different prices in different stores. It is also consistent with
consumers’ trade-off of paying a higher
price at a local store to save time
rather than traveling to a store farther
away that sells goods at lower prices.
As individuals search, they find goods
at some price and must decide whether
to accept a store’s price or continue
to search. Because search takes time
away from work, consumers consider
the forgone labor income of continuing to search, so the consumer’s wage
determines which prices the consumer
will accept. Individuals with higher
wages have a higher opportunity cost
of time and therefore are willing to
accept higher prices rather than search
repeatedly.
Firms, knowing consumers’ purchasing behavior, will charge higher
prices in markets where it is more
costly for the average consumer to
search repeatedly. This implies that
prices are higher in high-wage locations. Now, as long as the time it takes
to shop is not so different between
high- and low-income countries, lowincome countries will have a comparative advantage in search, so prices will

10 Q2 2008 Business Review

be lower in low-income countries. This
is a natural extension of the BalassaSamuelson mechanism described in
the box on page 6.
The theory developed here also
tells us something about the source of
income differences across countries.
In this model, countries with more
productive workers will earn higher
wages and be willing to pay higher
prices for all goods, both tradables
and nontradables. In contrast, in the
Balassa-Samuelson theory, for prices to
rise with wages, high-wage countries
must be relatively more productive at
producing tradables than nontradables.
Thus, the Balassa-Samuelson theory
requires that cross-country productivity differences in tradables be much
larger than productivity differences in
nontradables. The Balassa-Samuelson
theory suggests that countries mainly
become richer by becoming better at
producing tradables, while our theory
suggests a more balanced approach to
growth in which workers in a country
become better at producing everything.
Evidence on Shopping Time,
Prices, and Income Per Capita. The
theory we have described also implies
a relationship between wages, shopping
time per purchase, and prices. There
is some evidence that these variables

are related based on time-use surveys,
which are studies in which respondents
are asked to track their every activity in small time increments over the
course of a day or week. Examples of
activities tracked are sleeping, eating,
working, commuting to work, shopping, traveling to shopping, and listening to the radio.
Two recent papers use time-use
survey data to confirm a positive
relationship between wages and
prices paid and a negative relationship
between wages and time spent shopping predicted by our theory. Using
time-use data from the U.S., economists Mark Aguiar and Erik Hurst
find that when people retire, and the
opportunity cost of their time declines,
they spend more time shopping per
purchase and tend to pay less per unit
purchased. Likewise, using time-use
data from Argentina, David McKenzie
and Ernesto Schargrodsky find that
higher income individuals spend less
time shopping per purchase and pay
higher prices, on average (Table 4). In
fact, shopping time per expenditure of
people in the lowest income quartile is
about 80 percent higher than that of
people in the highest income quartile.
Moreover, after the economic crisis in
2001, which lowered all Argentineans’

TABLE 4
Shopping Frequency Per Real Expenditure by
Income in Argentina
Households by Income Quartile
All

Lowest

2nd

3rd

Highest

Pre-Crisis

0.24

0.29

0.26

0.21

0.16

Post-Crisis

0.28*

0.35*

0.31*

0.26*

0.20*

* Statistically different from 2001 mean at 1 percent level.
Shopping frequency measures time spent shopping.
From Table 3: McKenzie and Schargrodsky (2005)

www.philadelphiafed.org

real income, shopping time increased
by about 25 percent per expenditure
across all income levels.
We can also compare results of
time-use surveys in different countries
to get an idea of how shopping time
differs by income per capita. Figure 3
presents a scatter plot of time spent
per purchase against income per capita
based on data collected from countries
that participated in the European Harmonized Time Use Survey. The data
show that shopping time per purchase
tends to fall with income per capita, so
that in low-income countries people
tend to search more intensively than
in high-income countries. As we have
already seen, prices and wages tend
to rise with income per capita. Thus,
both the within-country evidence and
the cross-country evidence are consistent with the model we have described.
SUMMARY
There are large differences in
prices across countries that are related
to income per capita. On average, the
cost of a basket of goods tends to be
relatively high in high-income countries. These price differences exist both
for goods that are easily and frequently
traded and those goods that are not
traded. Moreover, these price differences show up at the dock, so that
export prices to high-income countries
tend to be higher than export prices to
low-income countries.
Understanding the determinants
of the price-income relationship sheds
light on the source of the large differ-

www.philadelphiafed.org

FIGURE 3
Time Spent Shopping Per Purchase
Log Shopping Time Per Expenditure (relative to UK)
1
Poland

0.5

Estonia

Latvia
Lithuania

Hungary

France
Germany
UK
Italy

0

Belgium

Slovenia
Spain

Finland

Sweden

-0.5
Norway

y = 0.80x - 0.11
R2 = 0.90

-1
-1.25

-1

-0.75

-0.5

-0.25

0

0.25

Log Real GDP Per Capita (relative to UK)

Data: European Harmonized Time Use Survey and Eurostat (2002)

ences in income and well-being across
countries. Traditional models of these
price differences have focused on differences in prices of nontradable goods
and thus attributed income differences
largely to differences in productivity in
the tradables sector. The evidence presented here that price differences are
quite large in the tradables sector as
well suggests a more balanced view of
productivity differences across sectors
and countries. The large price differences in tradable goods suggest that
policymakers should target improving

efficiency across the entire economy
and not just in the tradables sector.
The discussion has purposely
avoided nominal exchange rates.
However, a good theory of price levels
across countries is also useful as a
long-run theory of nominal exchange
rates. It provides a natural benchmark
for determining whether a currency
is overvalued or undervalued. For
countries that actively manage their
exchange rate, this may be a useful
guide in determining an appropriate
target level. BR

Business Review Q2 2008 11

REFERENCES

Aguiar, Mark, and Erik Hurst. “Lifecycle
Prices and Production,” mimeo, University
of Chicago (2005).

Bhagwati, Jagdish. “Why Are Services
Cheaper in Poor Countries?,” Economic
Journal, 94 (1984), pp. 279-86.

Ahmed, Sultan. “Historical Overview of
the International Comparison Program,”
available at http://siteresources.worldbank.
org/ICPINT/Resources/historical_
overview-Icp_SA.doc.

Jorgenson, Dale W., and Kevin J. Stiroh.
“U.S. Economic Growth at the Industry
Level,” American Economic Review, 90
(December 2000), pp. 161-67.

Alessandria, George, and Joseph Kaboski.
“Violating Purchasing Power Parity,”
Philadelphia Federal Reserve Bank
Working Paper 04-19 (2004).
Balassa, Bela. “The Purchasing Power
Parity Doctrine: A Reappraisal,” Journal of
Political Economy, 72 (1964), pp. 244-67.
Baumol, William, and William Bowen.
Performing Arts: The Economic Dilemma.
New York: The Twentieth Century Fund
(1966).

12 Q2 2008 Business Review

Kravis, Irving, and Robert Lipsey. Toward
an Explanation of National Price Levels.
Princeton Studies in International
Finance, No 52, Princeton University
(1983).
McKenzie, David, and Ernesto
Schargrodsky. “Buying Less, but Shopping
More: Changes in Consumption Patterns
During a Crisis,” BREAD Working Paper
092 (February 2005).

Ostapik, Edith, and Kei-Mu Yi.
“International Trade: Why We Don’t
Have More of It,” Federal Reserve Bank
of Philadelphia Business Review (Third
Quarter 2007).
Samuelson, Paul A. “Theoretical Notes on
Trade Problems,” Review of Economics and
Statistics, 46 (1964), pp. 145-54.
Stockman, Alan, and Linda Tesar. “Tastes
and Technology in a Two-Country
Model of the Business Cycle: Explaining
International Comovements,” American
Economic Review, 85(March 1995), pp.
168-85.

www.philadelphiafed.org

Liquidity Crises*
BY RONEL ELUL

F

inancial markets have experienced several
episodes of “liquidity crises” over the past 20
years. One prominent example is the collapse
of the Long Term Capital Management hedge
fund in 1998. The recent market disruption brought
about by the downturn in subprime mortgages also shares
many features with liquidity crises. What is liquidity?
Why does it sometimes seem that the market’s supply
of it is insufficient? Can anything be done about it? In
this article, Ronel Elul outlines some theories of market
liquidity provision, how it breaks down in times of crisis,
and some possible government responses.
Over the past 20 years, financial
markets have experienced several
episodes of “liquidity crises.” Among
these are the 1998 collapse of the Long
Term Capital Management hedge fund
and the disruption in financial markets
that began in the summer of 2007,
sparked by the downturn in subprime
mortgage markets.
In many of these cases, the
market’s supply of liquidity seemed
to be insufficient, and moreover,
liquidity does not always appear to
be allocated to those who need it

most. Lack of liquidity also sometimes
forces “fire sales,” actions that, in turn,
push down asset prices, thus making
liquidity problems worse. Economists
have sought to understand the nature
of market liquidity provision, how it
breaks down in times of crisis, and possible government responses.1
ANATOMY OF A
LIQUIDITY CRISIS
What Is Liquidity? One author
has pointed out that “liquidity, like
pornography, is easily recognized but

not so easily defined.”2 For understanding liquidity crises, however, it
may be useful to think of liquidity
as the ease of selling an asset at its
“true,” or fundamental, value. This
fundamental value may be defined as
the present value of the asset’s future
cash flows. Alternatively, liquidity can
be viewed as the extent to which it is
possible for the holder of an asset to
borrow against these future cash flows.
The Collapse of LTCM. The
events of the summer and fall of 1998
provide an illustration of many of the
main features of liquidity crises. These
events revolve around the collapse of
the Long Term Capital Management
(LTCM) hedge fund.3
During the summer of 1998,
LTCM took large losses on many of
its trades; these losses were intensified
when Salomon Smith Barney’s arbitrage group, which had positions very
similar to LTCM’s, was broken up and
its positions liquidated. But LTCM’s
position became much more precarious
on August 17, 1998, when the Russian
government devalued the ruble and
declared a moratorium on repaying 281
billion rubles ($13.5 billion) of its Treasury debt. The fact that the IMF had
allowed a major economy to default
shocked the markets.4

2

See the book by Maureen O’Hara.

1

Ronel Elul is a
senior economist
in the Research
Department of
the Philadelphia
Fed. This article
is available free
of charge
at www.philadelphiafed.org/econ/br/.
www.philadelphiafed.org

I use the term “government intervention”
broadly. In principle, this might include fiscal
policy or central bank monetary policy. In this
paper, I will focus on monetary policy.

3

Much of this account is drawn from Roger
Lowenstein’s book.

4

*The views expressed here are those of the author and do not necessarily represent the views
of the Federal Reserve Bank of Philadelphia or
the Federal Reserve System.

In addition, a further surprise occurred when
Russian banks and securities firms exercised
force majeure clauses and refused to honor the
derivatives contracts they had sold to foreign
customers. These clauses, which are common
in many contracts, are intended to free a party
from liability when an extraordinary event
prevents him from fulfilling his obligation.

Business Review Q2 2008 13

LTCM had indeed invested in
Russian bonds and lost money following this default. However, the
resulting flight to quality had an even
bigger effect on the value of LTCM’s
portfolio. Investors who had become
nervous as a result of these events
pulled out of risky assets and rushed
to assets considered safe. For example,
the yield on the 30-year U.S. Treasury
bond (a safe security) fell to its lowest
level up to that time. Many of LTCM’s
strategies had involved betting that the
spread between safe and risky assets
would actually decline; thus, the flight
to quality caused it to lose substantially
more. Finally, in addition to a flight to
quality in security markets, there was
a broad-based drying up of liquidity as
banks chose to preserve their liquidity
and cut back on lending.5
As a result of declines in prices on
the risky assets in its portfolio, LTCM
breached collateral agreements with
its lenders and was forced to sell assets
to meet these margin calls.6 These
asset sales had ramifications for other
markets and other hedge funds. Mark
Mitchell, Lasse Pedersen, and Todd
Pulvino recount an example: “When
LTCM incurred large losses on macroeconomic bets, the firm was forced
to liquidate large convertible bond
positions.7 These sales led to depressed
valuations of convertible bonds despite

the fact there was little change in overall fundamentals.8 As a result, other
hedge funds incurred large losses and
were also forced to sell their convertible bond holdings.” The authors show
further that prices of convertible bonds
fell far below their “fair” value, as calculated by mathematical models.9
Because of concerns that the
forced liquidation of LTCM’s huge

The events of the
summer and fall of
1998 provide an
illustration of many of
the main features of
liquidity crises.
portfolio would cause further upheaval
in financial markets, the Federal
Reserve helped coordinate a privatesector bailout of the fund in September
1998.10 The Fed also cut its fed funds
rate target by 75 basis points during
the fall of 1998, in part because of
concerns that financial market turmoil

7
A convertible bond is a type of bond that can
be converted into shares of stock in the issuing
company, usually at some pre-announced
ratio. Hedge funds are significant traders of
convertible bonds, as part of a popular strategy
known as convertible arbitrage.
8

5
See, for example, the September 1998 Federal
Reserve Senior Loan Officer Opinion Survey:
http://www.federalreserve.gov/boarddocs/
snloansurvey/199810/default.htm.
6

Like most hedge funds, LTCM had borrowed
heavily to finance its portfolio and borrowing
allowed it to generate higher returns per dollar
of outside investment. However, LTCM’s lenders
required that the value of these assets, which
served as collateral to secure its loans, stay
above a certain minimum margin requirement.
When the prices of its assets fell, these collateral
agreements were breached, and lenders issued
margin calls. This required LTCM to come up
with additional cash or securities in order to
avoid the forced liquidation of its portfolio.

14 Q2 2008 Business Review

In this example, prices are below their fair
value because of binding collateral constraints.
However, another reason that prices may fall in
a crisis is that one side of the market has more
information than the other, and thus, asset
sales may be interpreted as negative information
about fundamentals. For a similar model
motivated by the 1987 stock market crash,
see the paper by Gerard Gennotte and Hayne
Leland.

9
The fair value of the convertible bond is
calculated by using an option valuation model;
these models are extensions of the well-known
Black-Scholes pricing formula.
10
Although it will not be discussed here,
the economic rationale behind the Fed’s
coordinating role is also of interest; see the
paper by Stephen Morris and Hyun Shin.

might spill over to the real side of the
economy.
From this account we can identify
several key features of liquidity crises.
The apparent trigger for the crisis was
an unexpected event that called longstanding models into question. Lenders
responded by cutting back on providing liquidity. The effect of the crisis
was to push prices below their fundamental, or fair, value. More precisely,
the prices of risky assets fell, while
those of assets perceived to be safe
rose; that is, there was a flight to quality. There was commonality of illiquidity — problems spilled over from one
market to another. A liquidity spiral was
created: These falling prices caused
margin requirements to be breached,
thus leading to asset sales, which then
led to further drops in prices and thus
to further losses, and so on. Government intervention played a role in
resolving the crisis.
The Current Financial Market
Turmoil. Many of these features are
also present in the disruption in financial markets that began in the summer
of 2007, sparked by the downturn in
subprime mortgage markets. Not surprisingly, the sharp increase in default
rates on mortgages called into question
models of subprime mortgage credit
quality (as well as lenders’ underwriting standards). There was also a flight
to quality — for example, the premium
paid by high-quality (AAA-rated)
corporate borrowers over U.S. Treasury
bonds nearly doubled in the summer
of 2007 (Figure 1). In this case, market
participants suddenly demanded much
more compensation to bear even a
small amount of risk. The cutback in
the provision of private-sector liquidity
was even more dramatic than in the
case of LTCM. This may be seen most
strikingly in the interbank market;
the London InterBank Offered Rate
(LIBOR) that banks charge one another in the London interbank market

www.philadelphiafed.org

FIGURE 1
Five-Year AAA Corporate Bond Rate —
Five-Year Treasury Bond Rate
Percent

2007 - 2008

shot up relative to the baseline U.S.
Treasury bill rate as banks sought to
conserve their scarce liquidity. The
problems were particularly pronounced
in term (that is, not overnight) interbank markets (Figure 2). These events
were widely understood in the popular
press as reflecting liquidity hoarding.
Finally, the Federal Reserve and other
central banks intervened in several
different ways. (See Federal Reserve Responses to Recent Problems in Interbank
Markets.)
PRIVATE MARKETS MAY
PROVIDE TOO LITTLE
LIQUIDITY
One central feature of these episodes is the inadequacy of the private
market’s provision of liquidity. In
studying this issue, Bengt Holmstrom
and Jean Tirole explore various means
by which firms may obtain liquidity
and show that the private market may

www.philadelphiafed.org

not always be able to provide adequate
liquidity on its own. They then consider possible government responses.
They consider firms that have
long-lived projects, for example, manufacturing plants that can produce a
good for several years before becoming
obsolete. These firms may experience
a “liquidity shock,” a sudden need for
funds to keep the project going. This
could be due, for example, to an unanticipated, temporary shortfall in sales,
so that internal funds that were previously used to keep projects going are
no longer available. But if these funds
are not available, the firms’ assets
must be liquidated immediately, at a
loss. How can the firms obtain enough
liquidity to continue their projects?
These projects are still profitable,
so one might think that a firm that has
been hit by a shock could simply borrow against its future project returns.
But lenders may be unwilling to offer

sufficient funds to the firm because the
greater the firm’s required debt payments, the smaller the firm’s own share
of the returns from the project. This
means that the firm has less incentive
to exert enough effort to ensure that
the project succeeds.
To guard against this risk of illiquidity, a firm might hold cash or other
safe assets, such as Treasury securities,
that can be sold in case it experiences
a shock. Since these assets are safe,
the firm can always sell them to raise
funds. But Holmstrom and Tirole also
show that this is not generally an ideal
way to allocate scarce liquidity because
lucky firms that do not experience a
shock will be left with assets they do
not need, while unlucky firms have no
way to gain access to those assets.
What is needed instead is some
way for firms to obtain insurance
against unexpected liquidity needs.
This can be facilitated through a
financial intermediary that can offer
lines of credit to firms, which they
draw upon only if they experience a
shock. In effect, the financial intermediary takes stakes in all of the firms’
future returns and lends only to those
firms that have been hit by a shock.
When liquidity shocks are idiosyncratic — that is, the shock hits only a
few firms at once — Holmstrom and
Tirole show that this is indeed the best
way to provide liquidity to the private
sector.
However, in a liquidity crisis, in
which the liquidity shock is an aggregate one (that is, it hits many firms at
once — for example, a recession that
hits all firms’ sales), the private market
is not able to meet each firm’s liquidity
needs. The reason is that firms’ aggregate demand for liquidity will exceed
the private sector’s ability to meet this
need. In this case, there is scope for
the government to provide liquidity in
times of crisis. The government is able
to commit to providing liquidity when

Business Review Q2 2008 15

FIGURE 2
Three-Month LIBOR — Three-Month
T-Bill Rate
Percent

Jan

Feb Mar

Apr

May

Jun

Jul

Aug

Sep Oct

Nov

Dec

Jan

Feb Mar

Apr

2007 - 2008

the private market can’t, either by taxing consumers or by printing money.
Holmstrom and Tirole suggest that this intervention may take
many forms. For example, it could
take the form of government securities that pay off only in the event of a
particular aggregate liquidity shock.
Sundaresan and Wang document this
in connection with the run-up to Y2K.
They show that privately supplied
liquidity dried up as the millennium
approached. In response, the Federal
Reserve intervened by issuing options
on the fed funds rate.11 Alternatively,

11
These options, which were sold to Treasury
bond dealers, each gave the holder the right
to borrow $50 million from the Fed at a
pre-specified interest rate, on a specific date
between December 23, 1999, and January 12,
2000. The Fed also responded in other ways,
for example, by creating a “century date change
special liquidity facility” for banks.

16 Q2 2008 Business Review

Holmstrom and Tirole suggest that
monetary policy could serve this role
by easing financing conditions in times
of crisis.
VERY UNLIKELY
CONTINGENCIES CAN
AFFECT THE AVAILABILITY
OF LIQUIDITY
While Holmstrom and Tirole
focus on the lack of sufficient liquidity in the private sector as a rationale
for government intervention, another
feature of some liquidity crises is that
what liquidity is available is not efficiently allocated. That is, liquidity is
not allocated to those who need it the
most. The reason is that the liquidity
crisis may make market participants
overly concerned about extremely
unlikely risks and lead them to hoard
liquidity so as to insure against these
risks.
Ricardo Caballero and Arvind

Krishnamurthy study this phenomenon and show how government
intervention may be useful in resolving
it. They focus on liquidity crises that
begin with unexpected events that call
widely held beliefs and models into
question. We have seen that this may
characterize both the 1998 LTCM collapse as well as the recent disruption in
financial markets that began with the
downturn in subprime mortgage markets. Having scrapped old models, but
without well-articulated new models to
take their place, investors may tie up
so much capital in response to concerns about extreme — but unlikely
— events that they are unwilling to
provide financing to meet more moderate — but likelier — liquidity needs.
Consider the example of corporations that deposit funds in a bank and,
in return, have access to lines of credit
that they can draw on should they experience a liquidity shock. In Caballero
and Krishnamurthy’s model, a sudden
liquidity shock hits some firms in the
economy and generates a need for borrowing. But those firms not affected by
this first shock grow concerned that
they may be hit by a second shock,
even though this second shock is very
unlikely. The unaffected firms react
by preemptively drawing down their
own lines of credit.12 That is, they
hoard liquidity. The result is that there
is much less available for those firms
that actually need liquidity because
they have been hit by the first shock.
Reports in the popular press
during the recent financial market
disruption frequently refer to liquidity
hoarding motivated by uncertainty. For
example, in explaining elevated inter-

12

Firms may act preemptively because they are
concerned that when the second shock hits,
their credit quality will deteriorate so much that
they will violate the covenants in their lines of
credit and, thus, will be unable to borrow any
further.

www.philadelphiafed.org

est rates in the interbank market, the
Wall Street Journal quoted one banker
as saying that “[banks and investors]
are still fearful of each other and everybody is worried about counterparty
risk and so people are hoarding their
balance sheets.”13 This article also suggested that government intervention
might reassure market participants and
so reduce the impetus to hoard liquidity. We will see that, in Caballero and
Krishnamurthy’s model, government
intervention can play such a role.
But why would banks hoard
capital in response to an unforeseen
shock? Caballero and Krishnamurthy
assume that market participants are
uncertainty averse. That is, when evaluating outcomes about which they are
uncertain, they use the most pessimistic
probability assessments. In particular,
each participant overweights the probability that he will be among those hit
by the second shock. (See Uncertainty
Aversion.) This creates a desire to
hoard liquidity against this unlikely
shock.
Caballero and Krishnamurthy
then discuss how government intervention might remedy this market
failure. Their prescription is for the
government to act as a lender of last
resort. More precisely, by committing
to provide liquidity in the event that
the second shock occurs, the government thereby frees the private market
to insure itself against the first, more
likely, shock. Indeed, Caballero and
Krishnamurthy quote former Fed
Chairman Alan Greenspan to this
effect: “‘... [p]olicy practitioners operating under a risk-management paradigm
may, at times, be led to undertake
actions intended to provide insurance
against especially adverse outcomes.’”

13

See the article by Greg Ip and Joellen Perry.

www.philadelphiafed.org

LIQUIDITY AND ASSET PRICES
We have seen that one way for
firms to generate liquidity in times of
need is to sell assets. But the level of
liquidity can affect the value of these
assets. This can then result in a “spiral,” in which falling liquidity reduces
asset values, which, in turn, leads to
lower liquidity, and so on.14 We have
already discussed one example of this:
the convertible bond market during
the 1998 collapse of LTCM. These
forced sales, Mitchell, Pedersen, and

example, banks. So they themselves
also have a need for liquidity. Brunnermeier and Pedersen call this funding
liquidity.
In normal times, financiers themselves have adequate funding liquidity; therefore, they are able to provide
market liquidity to their customers and
thus assets are priced “fairly.” That is,
the price of an asset accurately reflects
its expected future cash flows. However, when funding liquidity is scarce,
there will also be insufficient market

Falling liquidity reduces asset values, which,
in turn, leads to lower liquidity.
Pulvino argue, were the result of binding capital constraints.
Markus Brunnermeier and Lasse
Pedersen develop a model that explains
these spirals, along with many other
features of liquidity crises. They focus
on a particular aspect of liquidity: the
need for immediacy. In their model a
customer may arrive with an immediate need to sell an asset today, but
no buyer may be available. So there
is a need for temporary liquidity to
bridge this gap (what they term market
liquidity). This need for immediacy is
provided by speculators (for example,
securities dealers or hedge funds). The
speculators serve a valuable economic
role: They buy the asset today and
then sell it at some later date when a
buyer arrives. The speculators require
funds in order to operate, and they
obtain these funds from financiers, for

14
The feedback between asset values and
financing conditions has also been explored by
macroeconomists seeking to explain the depth
and persistence of economic downturns. An
early and influential example is Irving Fisher’s
“debt deflation” theory of the Great Depression.

liquidity, and asset prices will need
to fall below this fair value to induce
speculators to buy.
But why would funding liquidity
be scarce? One reason is that speculators may have incurred losses on their
other activities (as LTCM did). In
addition, falling prices can themselves
negatively affect speculators’ funding
liquidity. The reason is that speculators
are limited in how much they can borrow by a collateral constraint.15 That

15

Brunnermeier and Pedersen model this as a
maximal value-at-risk (VaR) for the speculators.
Banks commonly use value-at-risk to measure
market risk, both for themselves and for their
counterparties. Indeed, the Basel II Accord —
an international agreement regarding how much
capital banks need to put aside to guard against
financial and operational risks — encourages
the use of VaR to determine the amount of
regulatory capital a bank must hold against its
market risk. In the Basel II framework, VaR is
calculated using a 10-day horizon, at a 1 percent
probability level. So if a bank’s market risk
model predicts that there is only a 1 percent
chance that the value of its portfolio will
decline by more than $1 million in the next 10
days, its VaR is $1 million. VaR thus depends
critically on the volatility of the value of the
assets. In Brunnermeier and Pedersen’s model
this goes up in a crisis because price declines
make the world appear to be more volatile.

Business Review Q2 2008 17

Federal Reserve Responses to Recent Problems in Financial Markets

A

s of March 2008, the Federal Reserve has
responded in several ways to the liquidity
problems associated with the recent
disruptions in financial markets. These
interventions provide examples of the
policy instruments available to central banks.
• Discount Window – The Fed took two broad classes
of actions to ease disruptions in financial markets
by making it less costly for depository institutions to
borrow directly from the Fed through the discount
window. The discount window offers some advantages
over private markets during episodes of tight credit.
First, the Fed accepts a wider variety of collateral
than do bank lenders (particularly during periods
of financial market turmoil). In addition, by lending
directly to depository institutions, the Fed can
supplement the interbank market at times when it is
not functioning well.a However, depository institutions
are often reluctant to borrow directly from the Fed
because of the perceived stigma it carries.
One step the Fed took was to narrow the spread
between the discount rate (which is the rate that
depository institutions must pay to borrow directly
from the Fed’s primary credit facility) and the federal
funds rate (the rate at which banks borrow and
lend among themselves, for one day at a time, on
an unsecured basis). It also extended the terms of
discount window loans; before the summer of 2007
they were overnight or very short-term loans. The
Fed did this in two stages: On September 18, 2007,
the Fed reduced the spread from 100 basis points
above the target fed funds rate to 50 basis points and
extended the maturity of discount window loans to up
to 30 days. On March 16, 2008, it lowered the spread

further, to 25 basis points, and also extended the
maximum maturity of discount window loans to 90
days.
• Term Auction Facility (TAF) – This was a new policy
tool announced on December 12, 2007. The Fed
undertook to make 28-day loans directly to depository
institutions at rates determined through competitive
auctions. From these institutions’ perspective, the TAF
has several potential advantages over the discount
window. One is that borrowing from the TAF may
carry less of a stigma for a depository institution than
accessing the discount window. In addition, depository
institutions were able to place bids below the discount
rate, so that they had the possibility of receiving
funding at lower rates. While the TAF was new for
the Fed, the European Central Bank regularly uses a
similar tool.
• Primary Dealer Credit Facility (PDCF) – On March
16, 2008, the Fed announced the PDCF, which is a
new, temporary discount window facility. The PDCF
provides overnight funding to primary dealersb at the
discount rate, in exchange for a specified range of
collateral, including investment-grade mortgage-backed
securities and asset-backed securities. This facility is
intended to improve the primary dealers’ ability to
provide financing to participants in securitization
markets and promote the orderly functioning of
financial markets more generally.
• 28-Day Single-Tranche Repurchase Agreements – On
March 7, 2008, the Federal Reserve announced that it
would initiate a series of term repurchase agreementsc
that are expected to cumulate to $100 billion. There

a Recall that there was a particularly large spread between LIBOR and Treasury rates, suggesting that there were indeed problems in the interbank
market.
b Primary government securities dealers (primary dealers) - of which there are 20 - are banks or securities broker-dealers who may trade directly with
the Fed. They are active participants in the Fed’s open market operations as well as in U.S. Treasury auctions. Several of these are investment banks,
and many others are subsidiaries of commercial banks. The current list of primary dealers may be found at http://www.newyorkfed.org/markets/
pridealers_current.html.
c A repurchase agreement (or “repo”) is a collateralized borrowing agreement structured as a sale of the collateral (in this case by dealers to the Fed),

along with an agreement to buy it back at a higher price in the future (in this case, 28 days later). This higher price implicitly determines an interest
rate, known as the “repo rate.”
d Although these 28-day single tranche repos differ from the ones typically conducted by the Fed, in the past the Fed has occasionally conducted
either 28-day, or single-tranche, repos.

18 Q2 2008 Business Review

www.philadelphiafed.org

Federal Reserve Responses ... continued
are two main differences between these agreementsd
and typical Fed repurchase agreements. First, they
are 28-day repos; typically the term is shorter. In
addition, they are “single-tranche”: dealers may submit
any of the following types of collateral — Treasuries,
agency debt, and agency mortgage-backed securities
— and pay the same repo rate regardless of its type.
By contrast, the repo rate typically differs by the type
of collateral, with those pledging Treasuries paying
the lowest rate and those pledging mortgage-backed
securities the highest. Since under the new program
market participants face the same rate regardless of
the collateral, they have an incentive to submit only
mortgage-backed securities.e
• Term Securities Lending Facility (TSLF) – On March
11, 2008, the Fed announced an expansion of its
securities lending program. Under the new program,
the Term Securities Lending Facility, the Federal
Reserve will lend up to $200 billion of Treasury
securities to primary dealers for a term of 28 days
(rather than overnight, as in the existing program) by
a pledge of other securities, including federal agency

debt and both agency and AAA-rated nonagency
mortgage-backed securities. The TSLF is intended to
promote liquidity in the financing markets for Treasury
and other collateral and thus to foster the orderly
functioning of financial markets more generally.
• Cooperation with Other Central Banks – Other central
banks also undertook to increase their liquidity
provision through similar means. In addition, the Fed
entered into “reciprocal currency arrangements,” in
which it lent dollars to the European Central Bank and
the Swiss National Bank, which, in turn, offered dollar
loans to their member banks. This was the first time
since September 11, 2001, that the Fed had entered
into such arrangements with central banks in Europe.
This period was also characterized by a slowing economy
and by concerns that continued financial market turmoil
could slow the real economy further. In response, the
Fed reduced its target for the fed funds rate. Between
September 2007 and March 2008, the Fed cuts its target
from 5.25 percent to 2.25 percent

e That this was in fact the case can be seen from the New York Fed’s announcements of the results of its single-tranche repos on March 7, 11, 18,

and 25, 2008.

is, the speculators must post securities to back any loan they take. (See
Margins and Liquidity.) This collateral
constraint becomes tighter during a
crisis for two reasons. First, the value
of the collateral is lower in the crisis,
because asset prices have fallen. In
addition, falling prices make the
world appear more volatile, which also
leads the bank to tighten its collateral
constraint. This results in a liquidity
spiral, as described above.
In addition to liquidity spirals,
Brunnermeier and Pedersen are able
to explain other characteristic fea-

www.philadelphiafed.org

tures of liquidity crises. The first is
the flight to quality: a rush to buy safe
assets, which is reflected in a relative
appreciation in their price. This is an
outcome of Brunnermeier and Pedersen’s model because when liquidity is
scarce, market participants prefer to
conserve it by investing in less risky
assets. This causes the price of riskier
assets to fall more than those of safe
ones (and hence causes their yield
to rise relative to that of safe assets).
Another feature that Brunnermeier
and Pedersen are able to explain is the
commonality of liquidity across securi-

ties and markets, that is, the tendency
for liquidity problems to spread from
one market to another (as was the
case for the convertible bond market
in 1998, discussed earlier). This occurs
because speculators provide liquidity
in many markets simultaneously, and
so a deterioration in their financing
position (even if initially caused by a
shock to a single market) will affect all
of these markets.
Brunnermeier and Pedersen
also discuss the possible regulatory
responses to a crisis. First, they argue
that if the regulator “knows” that this

Business Review Q2 2008 19

Uncertainty Aversion

R

icardo Caballero and Arvind Krishnamurthy use uncertainty aversion in
their model to explain the hoarding of
liquidity. There is evidence that decisions
are indeed characterized by aversion to
uncertainty; that is, individuals seek to
avoid ambiguous situations in which probabilities are not
known. A classic example is the Ellsberg paradox.
In the Ellsberg paradox, there is an urn containing 90 balls: 30 red balls and 60 black and yellow balls in
unknown proportion. A ball is drawn, and an individual
is then asked whether he prefers to bet that this ball is
red (gamble A) or black (gamble B). Most people choose
gamble A. That is, they prefer to bet on a red ball, which
they know will occur one-third of the time, against the
chance of a black ball, whose probability lies somewhere
between 0 and two-thirds. Note that gamble B reflects
uncertainty, in that the precise probability that a ball is
black is unknown.
Then the ball is replaced and another is drawn.
The (same) individual is now asked whether he prefers
to bet that the ball is either red or yellow (gamble C) or
black or yellow (D). Most people choose gamble D in this
case. That is, they prefer to bet that a black or yellow ball
is drawn, which they know will happen two-thirds of the
time, against the chance of a red or yellow ball (the latter

probability lies somewhere between one-third and 1).
But there is a certain inconsistency in these two
choices because the second gamble is really equivalent to
the first, with the addition of the possibility of yellow balls
to each option. So if someone prefers gamble A over B, he
should actually prefer gamble C over D. For example, if
one believes there are more red balls than black balls (and
hence prefers gamble A over B), one also believes that
there are more red or yellow balls than black or yellow
balls, and so should prefer gamble C over D.
Ellsberg explained these seemingly contradictory
choices as reflecting individuals’ dislike for uncertainty,
that is, for unknown probabilities. They prefer A over B
because they know that one-third of the balls are red;
conversely, they prefer D over B because they know that
two-thirds of the balls are either black or yellow.
Since this evidence is inconsistent with the canonical economic model of expected utility maximization,*
it has led to the development of alternative models of
decision-making under uncertainty. One of these, developed by Itzhak Gilboa and David Schmeidler, models
individuals faced with unknown probabilities as pessimistic – that is, as making decisions under the “worst case”
probabilities. This is the approach used by Caballero and
Krishnamurthy.

* In particular, the “sure-thing principle.”

is, in fact, a temporary liquidity shock,
he should try to convince banks to
lend to speculators. If the regulator
is correct, banks’ profits in times of
crisis will actually be higher. However,
attributing superior information to the
regulator is, of course, a very strong
assumption. In addition, Brunnermeier
and Pedersen suggest that the regulator
can help stabilize prices by providing
liquidity directly to the speculators (or
to the banks that finance them).

20 Q2 2008 Business Review

CONCLUSION
Financial intermediaries serve to
allocate the private market’s wealth
so as to meet firms’ and investors’
liquidity needs. But “liquidity crises,”
in which the market fails to function
properly, are a recurrent feature of
financial markets.
During these episodes the demand
for liquidity may be so great that it
cannot be met by the private sector
alone. This creates a role for govern-

ment intervention; the government
can provide liquidity in these circumstances through its ability to raise
funds by taxing consumers.
In addition, these episodes may
also be characterized by a misallocation of private liquidity. Market participants may hoard liquidity because they
become concerned about extremely
unlikely events. In this case there may
be a further role for the government in
insuring against these extreme events.

www.philadelphiafed.org

Margins and Liquidity

I

n Markus Brunnermeier and Lasse Pedersen’s model, margin requirements play
a key role. They may exacerbate liquidity
crises and thus facilitate a liquidity spiral.
Likewise, Mark Mitchell, Pedersen, and
Todd Pulvino document the role that
capital constraints have played in many real-life liquidity
crises (for example, in the convertible bond market in the
1998 LTCM crisis). Given that they may exacerbate crises, why do margin requirements and capital constraints
exist? And are they optimally determined by private
markets?
On the most basic level, requiring borrowers to post
margin, or collateral, facilitates lending by increasing the
likelihood that lenders will be repaid.a Viral Acharya
and S. Viswanathan show how this can help to provide
liquidity. Consider firms facing liquidity shocks. If they
are able to borrow enough, they can meet their liquidity
needs without selling any assets. However, firms may
be limited in what they can borrow because lenders
are concerned that the firm might divert funds to risky
projects. In general, borrowers have a tendency to prefer
riskier projects than their lenders because lenders bear the
brunt of any losses when the firm defaults. By pledging
their assets as collateral, firms are able to reassure lenders
that they will indeed invest efficiently; so firms are able to
borrow more and thus meet larger liquidity needs.

While collateral does facilitate borrowing, if the
liquidity shock is large, not all firms will be able to borrow
enough to survive, and some will need to be liquidated
early. In cases where there are many more sellers than
buyers, this will, in turn, lower asset prices. These lower
prices then lead to tighter collateral constraints and thus
to a liquidity spiral, as described earlier.
Although Acharya and Viswanathan show that the
ability to post collateral is valuable to society, does it
follow that the private market sets collateral requirements
optimally? In a different model, John Geanakoplos and
Felix Kubler present an example suggesting that this is
not always the case.b They show that margins may be too
loose in good times and too tight in crises because lenders
do not take into account the effect that the margins they
set have on market prices. So there may also be a role
for government regulation of margin requirements (such
a view was also recently expressed by former Treasury
Secretary Robert Rubin). Indeed, Geanakoplos and
Kubler suggest that margins may be too loose in good
times, since lenders do not realize that their lending
increases the risk of a future crisis; conversely, margins
may be too tight in crises, since increasing lending would
raise prices and thereby ease the crisis. The extent to
which this intuition can be generalized is unclear, and
further research in this area would be valuable.

a For more on the economic role of collateral, see the article by Yaron Leitner.
b See also the presentation by John Geanakoplos at the Philadelphia Fed Policy Forum.

This frees up the private markets’
liquidity and allows it to be used more
effectively.
Another feature of liquidity crises
is the interaction between liquidity
and asset prices. A lack of liquidity
can lower asset prices below their fair,

www.philadelphiafed.org

or fundamental, value. Since liquidity
is often obtained by using these assets
as collateral for loans, this in turn
can lead to lower liquidity provision.
The outcome is a “liquidity spiral,”
in which the resulting illiquidity can
lower asset prices even further, and so

on. Further research is needed on this
role of collateral in providing liquidity
and, in particular, on the question of
whether government intervention can
improve on the private market’s use of
collateral. BR

Business Review Q2 2008 21

REFERENCES

Acharya, Viral V., and S. Viswanathan.
“Moral Hazard, Collateral and Liquidity,”
manuscript, December 2007.
Brunnermeier, Markus K., and Lasse Heje
Pedersen. “Market Liquidity and Funding
Liquidity,” NBER Working Paper 12939
(2007); Review of Financial Studies (forthcoming).
Caballero, Ricardo J., and Arvind Krishnamurthy. “Collective Risk Management
in a Flight to Quality Episode,” Journal of
Finance (forthcoming).
Fisher, Irving. “The Debt-Deflation Theory
of Great Depressions,” Econometrica 1
(1933), pp. 337–57.
Gilboa, Itzhak, and David Schmeidler.
“Maxmin Expected Utility with Nonunique Priors,” Journal of Mathematical
Economics, 18 (1989), pp. 141-53.
Geanakoplos, John, and Felix Kubler.
“Leverage, Incomplete Markets and Pareto
Improving Regulation,” mimeo, Yale University (November 2005).

22 Q2 2008 Business Review

Geanakoplos, John. “The Leverage Cycle,”
presentation at the Philadelphia Fed Policy
Forum, November 30, 2007. http://www.
philadelphiafed.org/econ/conf/forum2007/
presentations/Geanakoplos_leverage_
cycle_6_11-20-2007.pdf
Gennotte, Gerard, and Hayne Leland.
“Market Liquidity, Hedging, and Crashes,”
American Economic Review, 80:5 (December 1990), pp. 999-1021.
Holmstrom, Bengt, and Jean Tirole.
“Private and Public Supply of Liquidity,”
Journal of Political Economy, 106:1 (February 1998), pp. 1-40.
Ip, Greg, and Joellen Perry. “Central Banks
Launch Effort to Free Up Credit,” Wall
Street Journal, December 13, 2007.
Leitner, Yaron. “Using Collateral to Secure
Loans,” Federal Reserve Bank of Philadelphia Business Review (Second Quarter
2006).

Lowenstein, Roger. When Genius Failed:
The Rise and Fall of Long-Term Capital
Management. New York: Random House,
2000.
O’Hara, Maureen. Market Microstructure
Theory. Blackwell, 1995.
Morris, Stephen, and Hyun Shin. “Coordination Risk and the Price of Debt,”
European Economic Review, 48 (2004), pp.
133–53.
Mitchell, Mark, Lasse Pedersen, and Todd
Pulvino. “Slow Moving Capital,” American
Economic Review, Papers and Proceedings,
97:2 (2007), pp. 215-20.
Sundaresan, Suresh, and Zhenyu Wang.
“Y2K Options and the Liquidity Premium
in Treasury Bond Markets,” Federal
Reserve Bank of New York, Staff Report
No. 206.

www.philadelphiafed.org

The Evolution of the
World Income Distribution*

T

by KEITH SILL

here is tremendous disparity in the levels
of individuals’ incomes across countries.
However, this disparity in per capita income
has not always existed. In this article, Keith
Sill investigates some facts about the evolution of per
capita income across countries and reviews a simple
model that broadly captures the observed evolution
of the world income distribution since 1800. He then
discusses what predictions can be made about future
cross-country distributions of income and some policy
prescriptions that follow from our understanding of the
past and our predictions about the future.

There is tremendous disparity
in the levels of individuals’ incomes
across countries. Those fortunate
enough to live in the richest countries
have an average income that is about
30 times greater than the average
income of residents of the world’s poorest countries. Such a large disparity
in income across countries implies
large differences in living standards

Keith Sill is an
economic advisor
and economist
in the Research
Department of
the Philadelphia
Fed. This article
is available free of
charge at www.philadelphiafed.org/econ/br/.
www.philadelphiafed.org

and well-being. A significant share
of the world’s population has a living
standard well below that of the average
U.S. citizen. Indeed, inhabitants of the
world’s poorest countries face daily
hardships and deprivations that are so
foreign to the citizens of rich countries
as to be hard to believe.
However, this large difference in
per capita income across countries
has not always existed. It wasn’t until
the early 19th century that countries
began to experience significantly different growth rates in income as some
countries were quicker to begin the
process of industrialization. Conse-

*The views expressed here are those of the author and do not necessarily represent the views
of the Federal Reserve Bank of Philadelphia or
the Federal Reserve System.

quently, before the late 1800s, there
was relatively little income disparity
across countries, at least by today’s
standards. But it doesn’t take long for
small differences in income growth
rates to lead to wide divergence in per
capita income levels. From the late
1800s until about the 1960s, there was
a steady and rapid increase in inequality. Since then, the cross-country
dispersion in per capita income has
become somewhat more stable, while,
at the same time, world poverty has
been decreasing as countries with large
populations, like China and India,
begin to industrialize.
We’ll investigate some facts about
the evolution of per capita income
across countries and review a simple
model that broadly captures the observed evolution of the world income
distribution since 1800. Given our
analysis of what happened in the past,
we’ll discuss what predictions can be
made about future cross-country distributions of income. We’ll also discuss
some policy prescriptions that follow
from our understanding of the past and
our predictions about the future.
EVOLUTION OF COUNTRY PER
CAPITA INCOMES BEFORE 1800
Before 1800 and the onset of the
Industrial Revolution, the distribution
of world income looked very different
than it does today. While cross-country data on incomes and population
prior to 1800 are incomplete and challenging to piece together, the available information suggests that there
was little, if any, growth in per capita
incomes in any of the world’s economies. Before the Industrial Revolu-

Business Review Q2 2008 23

tion, economies were agricultural, and
living standards were similar across
countries and over time. People born
before 1800 could expect to be about
as well off as their parents, grandparents, and great-grandparents. In addition, they could expect their children
to be about as well off as they were.
Moving to a different country wouldn’t
have improved living standards much
either — the agricultural technology
across countries was about the same.
This stands in stark contrast to today’s
world, in which living standards have
increased rather consistently over time
(at least in the developed countries)
and vary greatly between poor and rich
countries.
We will measure the standard of
living, or economic well-being, of the
typical resident of a country using
real gross domestic product (GDP)
per capita, which is real GDP divided
by a measure of the population. Real
GDP is all of the goods and services
produced domestically by residents of a
country. Higher real GDP means that
a country produces more goods and
services for its residents to consume
and invest in. By itself, real GDP is not
a particularly good measure of how
rich a country is because a country
with a large population is likely to
produce more than a country with a
small population. When we divide a
country’s real GDP by its population,
we get a measure of goods and services
produced per person: Rich countries
will produce more per person than
poor countries.
However, real GDP per capita
is not an all-inclusive measure of a
country’s well-being. Factors that affect
well-being include leisure time, income
sharing within households, environmental quality, and health. These
factors may be imperfectly correlated
with output per person, and there is
some evidence from survey data that
the correlation between output per

24 Q2 2008 Business Review

capita and happiness is weak across
OECD countries.1 Despite these
potential problems, we will treat real
GDP per capita as a useful summary
measure of well-being for purposes
of cross-country comparison. After
all, it seems implausible to argue that
in some broad sense Africa’s poorest
residents are as well-off as the residents
of the U.S.
Unfortunately, official statistics
on GDP for most countries start after
World War II. So how do we measure
the world income distribution far back

that average living standards before
1800 were probably similar to the
living standards of today’s poorest
countries, which are agricultural societies that do not have much capital or
technology to work with.
Figure 1 plots some of the per
capita income data from Maddison’s
study for several regions of the world
from 1 AD to 1820.2 The figure shows
that even the fastest growing regions of
the world, which are denoted Western
Europe and Western Offshoots (Australia, Canada, New Zealand, and the

We could hypothesize that average living
standards before 1800 were probably similar
to the living standards of today’s poorest
countries, which are agricultural societies that
do not have much capital or technology to
work with.
in history? Recent work by Angus
Maddison, the eminent economic
historian, pieces together various bits
of evidence to develop measures of real
GDP per capita for several regions of
the world going back to 1 AD. Going
that far back in time means that there
is considerable uncertainty about
particular measures of living standards,
since the recorded evidence on how
people lived is sparse. However, over
that time span, virtually all societies
were traditional agricultural societies,
and agricultural technology seems not
to have varied greatly across countries.
Furthermore, we could hypothesize

The Organization for Economic Cooperation
and Development is a group of 30 countries that
share a commitment to democratic government
and the market economy. The working paper
by Romina Boarini, Asa Johansson, and Marco
Mira D’Ercole provides an overview of the
literature on wealth and happiness.

1

United States), had per capita incomes
that increased only by a factor of two
to three over a span of 1800 years.
This amounts to minuscule growth
of only about 0.04 percent per year.
By 1820, the richest region (Western Europe) had per capita income
about three times that of the poorest
region (Africa). But this is nothing
like the 30-fold difference we see
today between the richest and poorest countries. The story of economic
growth before 1800 appears to be one
of stagnation in living standards.
The near-zero growth of per capita
incomes between 1 AD and 1820 does
not mean that there was no technological progress during that time. Productivity-improving inventions such as the
stirrup, the heavy plow, and the threefield system of crop rotation were being
adopted. However, societies responded
to technological advance by increas-

www.philadelphiafed.org

FIGURE 1
Per Capita Income 1 AD to 1820 AD
Per Capita Income $1990

Western Europe

Latin America

Western Offshoots

Asia

Eastern Europe

Africa

Year

Source: Maddison (2006) at http://www.ggdc.net/maddison

ing their populations rather than by
accumulating more capital per worker
and thus increasing output per worker.
In other words, population grew at the
same rate as output, so that output
per worker, or living standards, stayed
nearly constant.3
Why was there no growth in
per capita income in the pre-1800
period? Before 1800, land was a very
important factor of production, since

To get a common unit of measurement across
countries, Maddison expresses GDP in 1990
Geary-Khamis dollars. Geary-Khamis is an aggregation method in which international prices
and purchasing power parity are estimated
simultaneously. See the OECD’s website http://
stats.oecd.org/glossary/detail.asp?ID=5528 and
its references for a more detailed description.
3
The arguments in this section are akin to those
in Lucas (2003). For a different point of view,
Esther Boserup, who argues that
rising population density drove technological
progress.
2

www.philadelphiafed.org

economies were largely agrarian.
However, the quantity of land available for production was to some extent
fixed. If the return to adding a worker
to a plot of land diminishes as more
workers are added, output per worker
declines with population growth.
However, technological advance is an
offsetting factor that makes workers
more productive. If these two opposing forces of population growth and
technological advance approximately
balance each other, output per capita
remains roughly constant over time,
even though technological progress
occurs.4 It is not a coincidence that

The paper by Gary Hansen and Edward
Prescott presents a unified theory of growth
that makes an effort to account for the pre-1800
and post-1800 growth experience. Their theory
assumes that population increases with consumption at low standard-of-living levels.

4

these forces roughly cancel each other.
Thomas Malthus famously argued that
the stability of living standards at a
low level of subsistence resulted from
the tendency of population to grow
whenever living standards rose above a
subsistence level. The benefits afforded
by steadily improving technology were
absorbed in supporting a steadily rising
population.
POST-1800 INCOME
DISTRIBUTION: THE
INDUSTRIAL REVOLUTION
AND THE ERA OF MODERN
ECONOMIC GROWTH
While the pre-1800 era is marked
by relatively little, if any, growth in per
capita incomes across countries, the
post-1800 era is marked by a dramatic
surge in growth for some countries
but not others. We see this in Figure
2, which plots income per capita for
the same regions as in Figure 1 but
includes the period 1820 to 2000.
Some regions, such as the West, experience sustained increases in their per
capita incomes. Other regions, such
as Africa, show virtually no increase
in per capita income over this time
span. A consequence of this disparity
in growth is that the levels of income
per capita vary greatly across regions.
Indeed, by 2000, the countries labeled
Western Offshoots (Canada, the U.S.,
Australia, and New Zealand) had per
capita incomes that averaged about
15 times higher than the per capita
income of Africa. The U.S. has a per
capita income that is about 30 times
higher than that of the poorest countries in Africa.
The central question of economic
development is why some countries
make the transition to modern growth
while others stagnate. Many factors
appear to be at work, and the particular set of circumstances, policies, and
institutional structure that coincides
with the transition to modern growth

Business Review Q2 2008 25

FIGURE 2
Per Capita Income 1 AD to 2000 AD
Per Capita Income $1990

Western Europe

Latin America

Western Offshoots

Asia

Eastern Europe

Africa

Year

Source: Maddison (2006) at http://www.ggdc.net/maddison

varies from country to country. Developing well-defined property rights
seems to be an important factor as
does developing human capital so that
people can implement and take advantage of new technologies.
What we see happening as
countries make the transition from
stagnation to economic growth is that
countries industrialize, technological
change allows for increasing productivity, and capital accumulation begins.
That is, residents invest in buildings
and machines that help produce
more future output. Countries that
enter this modern growth phase shift
from primarily agricultural output to
industrial output that is not as land
intensive. That is, countries undergo
an industrial revolution. A second key
observation is that fertility changes,
so that improvements in technology

26 Q2 2008 Business Review

no longer result in matching increases
in population. Parents have fewer
offspring and invest more resources
in their care and development. This
demographic transition is a defining
feature of the transition to sustained
growth for economies.5
Great Britain was the first country
to enter the industrialization process. The exact date of the Industrial
Revolution’s beginning is difficult to
determine, but many historians place
it in the latter half of the 1700s. From
Great Britain, it spread to Western Europe and the United States. The onset
of the Industrial Revolution led to divergence in per capita incomes across
countries, since not all countries began

the industrialization process at the
same time. Consequently, the world
income distribution started to become more unequal as some countries
started to grow (at different dates) and
others that had not started to industrialize continued to stagnate.
Income inequality across countries
largely increased from about 1820 until
the latter half of the 20th century as
more and more countries began to industrialize. Since 1960, there are good
cross-country data that can be used
to examine the change in the world
income distribution more carefully.
Alan Heston, Robert Summers, and
Bettina Aten have developed these
data in the Penn World Tables. In
these tables, these researchers provide
national income accounts converted to
international prices for 188 countries
from 1950 to 2004, assuming purchasing power parity holds.6 Their data set
allows a direct comparison of income
across countries because incomes are
measured in common units.
Figure 3 plots the distribution of
income per capita across countries
using the Penn World Table Mark 6.2.7
The distribution is approximated by a
histogram, which shows what proportion of countries fall into a particular
category of per capita income levels. In
Figure 3, income levels are measured
relative to the U.S., so the scale runs
from zero to one.8 Histograms are

Purchasing power parity is a relative price
concept that measures the number of units of
country A’s currency that are needed in country
B to purchase the same quantity of a good or
service as one unit of country A’s currency will
purchase in country A. See the OECD’s glossary of statistical terms at http://stats.oecd.org/
glossary/index.htm.

6

The data are available at http://pwt.econ.
upenn.edu/.

7

Some countries have higher per capita income
than the U.S. We placed all countries with per
capita incomes greater than or equal to the U.S.
in the same data category as the U.S.

8

The article by Aubhik Khan discusses the relationship between the demographic transition
and industrialization in more detail.

5

www.philadelphiafed.org

plotted for three years: 1970, 1985,
and 2000. All countries that have data
available for each of the three years
are included in the charts. Perhaps
the most striking observation from
the charts is that so many countries
have per capita incomes far below that
of the U.S. In fact, about 75 percent
of countries in the 2000 sample have
per capita incomes that are less than
half that of the U.S. Furthermore, the
distribution has been fairly stable over
the 30-year span.
Although there wasn’t a large
change in the number of poor countries between 1970 and 2000, that
doesn’t mean that worldwide poverty
has not decreased. First, the histograms in Figure 3 assume that each
resident of a country gets the same
share of real GDP. However, we know
that there is income inequality within
countries and that this inequality
can change through time. Any such
change in within-country inequality
is not captured by the histograms in
Figure 3. Nevertheless, the difference
in living standards across countries
tends to exceed the within-country
differences by most measures. For
example, in 1988 the ratio of the per
capita income of a country in the
richest 10 percent of countries to
that of a country in the poorest 10
percent was 20. But for the U.S., the
ratio of the permanent income of an
individual in the richest 10 percent to
that of an individual in the poorest 10
percent was less than four.9 Second,
the histograms in Figure 3 give each
country equal weight, despite the fact
that some countries have much larger
populations than others. So if China,
with its massive population, jumps to
a new income category in the figure,

FIGURE 3
World Income Distribution Per Capita
Relative to U.S.
Fraction

1970

0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
0-.1

.1-.2

.2-.3

.3-.4

.4-.5

.5-.6

.6-.7

.7-.8

.8-.9

.9-1.0

.6-.7

.7-.8

.8-.9

.9-1.0

.6-.7

.7-.8

.8-.9

.9-1.0

Per Capita Income

Fraction

1985

0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
0-.1

.1-.2

.2-.3

.3-.4

.4-.5

.5-.6

Per Capita Income

Fraction

2000

0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05

9
Permanent income is a measure of a wage
earner’s long-term income that ignores shortterm fluctuations in earnings.

www.philadelphiafed.org

0
0-.1

.1-.2

.2-.3

.3-.4

.4-.5

.5-.6

Per Capita Income

Business Review Q2 2008 27

it has the same effect as if Bermuda
jumped income categories.10
We can partially correct our histograms for country size by multiplying
income per capita in each country by
that country’s share of world population (note, though, that this still
does not correct for within-country
inequality). The weighted country
distributions are shown in Figure 4,
and now the picture looks somewhat
different. We see that in 1970, ignoring
within-country inequality, nearly 60
percent of the world population had a
per capita income that was 15 percent
or less than that of the U.S. By 2000
there had been a significant shift, in
that the 15 percent of U.S. income or
less category shrank at the expense of
the 15 to 40 percent category. Thus,
many fewer people were in the lowest
income category compared to 1970.
What’s happening is that high-population countries like China and India
are starting to experience sustained
increases in per capita incomes. This
evidence suggests that world poverty
has been decreasing even though the
distribution of country per capita incomes has remained relatively stable.
A SIMPLE MODEL OF
THE WORLD INCOME
DISTRIBUTION
The evidence presented so far
suggests three phases for the dynamics of the world income distribution.
The first phase, the pre-1800 period,
is characterized by economic stagnation because per capita incomes were
not growing. The second phase, from
1800 to the latter half of the 20th
century, is characterized by increasing
inequality as industrialization diffused
through the world’s economies. The
third phase, from about 1960 to now,
was one in which the world income

FIGURE 4
Population-Weighted World Income
Distribution Per Capita Relative to U.S.
Fraction

1970

0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
0-.1

.1-.2

.2-.3

.3-.4

.4-.5

.5-.6

.6-.7

.7-.8

.8-.9

.9-1.0

.6-.7

.7-.8

.8-.9

.9-1.0

.6-.7

.7-.8

.8-.9

.9-1.0

Per Capita Income

Fraction

1985

0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
0-.1

.1-.2

.2-.3

.3-.4

.4-.5

.5-.6

Per Capita Income

Fraction

2000

0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0.0

See the article by Xavier Sala-i-Martin for a
thorough discussion of this issue.

10

28 Q2 2008 Business Review

0-.1

.1-.2

.2-.3

.3-.4

.4-.5

.5-.6

Per Capita Income

www.philadelphiafed.org

distribution was somewhat more stable.
How should we think about these
phases and the recent stability of the
distribution? Is the recent period a
pause before a return to a further
increase in inequality? Is it the shape
of things to come, in that things have
settled down and we should expect the
future to look much like the recent
past? Or is the most recent phase
merely a transition period that will
usher in a period of declining income
inequality across countries?
The last conjecture is nicely
advocated in a recent paper by Nobel
laureate Robert Lucas titled “Some
Macroeconomics for the 21st Century.”
We are going to explore Lucas’s reasoning further and, as a byproduct, see
that the conjecture about a “pause before a return to increasing inequality”
and the conjecture about the “shape
of things to come” are somewhat at
odds with the long-run world-growth
mechanism. However, none of the
views can really be proven or disproven
based on the existing set of facts, since
the truth will be revealed only as the
future unfolds.
Lucas uses a simple mechanical
model to articulate the view that the
last 40 years represent a transition
period for the world income distribution. The transition can be thought of
as moving from the pre-industrial era
to the post-industrial era. The post-industrial era is one in which industrialization has largely diffused throughout
the world’s economies and ever fewer
countries remain stagnant. The model
is mechanical in the sense that while
it is based in part on an economic
interpretation of the facts, there are no
choices, policies, or meaningful actions
at the model’s core.
The model has several key elements. Consider first the concept of
the technology frontier. The frontier
represents the best, most recent
technology that countries can use for

www.philadelphiafed.org

transforming labor and capital inputs
into output. These new technologies
represent things like information technology, genetics, advances in medical
care, improved organizational methods, and a host of other things that
make labor and capital more productive. If we examine the data on per
capita incomes for countries over the
past 200 years, we could reasonably argue that, on average, the richest countries, those at the technology frontier,
increase their real incomes at a pace
of about 2 percent per year. If we take
this as giving us some information
about the pace of technological advance, we can say that the technology
frontier grows about 2 percent per year.
This represents the maximum long-run
growth a country can achieve. The
model takes this as given and does not
describe an economic mechanism for
why the technology frontier grows at 2
percent or how to make it grow faster.
For simplicity, assume that all
countries have the same population
and that the population does not
grow.11 Also, think of these countries
as starting out at some time before the
Industrial Revolution, so that there is
no growth in real income per capita.
The model begins at a time when the
world consists of a bunch of poor, stagnant economies that have the same
incomes and the same populations.
How does growth occur and industrialization diffuse? Lucas uses a racetrack
metaphor to describe how the model
works.
Imagine all countries lined up in
a row like horses at the starting gate
of a racetrack. But instead of all gates
opening at the same time, they open

randomly. When the gate opens, the
corresponding country starts to grow.
Thus, some economies start to grow at
time 1, others at time 2, others at time
3, etc. In any year after the starting
period (taken to be 1800), the world
economy is composed of countries that
have already been growing, those that
just started growing, and those still
waiting to start growing. The Lucas
model does not posit any economic
reasons for a country to make the
transition to modern growth and so
does not offer policy advice to kick
start economies into the growth phase
of development. Rather, it models the
random process by which gates are
opened using observations on world
income growth since 1800.12
The model makes two other key
assumptions based on observations
of historical growth rates. First, when
a country begins to grow, it does not
necessarily grow at 2 percent, the rate
of growth of the technology frontier.
Rather, it grows at 2 percent plus a
growth rate determined by the income
gap between itself and the richest
country. The later a country starts
to grow, the larger is the income gap
between itself and the leader (which
we will take to be the U.S.) and the
faster it grows. As a country closes the
income gap with the leader, its growth
rate begins to slow toward 2 percent.
In the long run, a country that is
growing does so at 2 percent, the rate
of growth of the technology frontier.13
This modeling behavior is based on
observations from countries such as
South Korea, Japan, and China, which
experienced very high growth rates
as they began the industrialization

The model assumes that all countries are the
same size and so makes the same prediction
for inequality whether or not countries are
weighted by population. In the model, country
size does not matter for growth or development
or for the inequality consequences of development of large-population economies.

12
Thus, the model does not explain why the
Industrial Revolution started in England and
then quickly spread to continental Europe and
the United States.

11

See Lucas’s 2000 paper for a detailed description of the model and how it is calibrated.

13

Business Review Q2 2008 29

process. This can happen because
these later entrants to the growth
club do not have to reinvent all of the
advanced technologies they see in
countries like the U.S. To some extent
they can import advanced technologies
and try to copy existing best practices
and methods. This allows them to
industrialize more rapidly than if they
had to develop all the new technologies
themselves.
Another key assumption is that
the probability that a stagnant country
begins to grow in any given period is
positively related to the level of income
in the rest of the world. The more
countries that are growing in the rest
of the world, the higher the probability
that a stagnant country will start to
grow. This is a model of spillovers: The
more technology the world acquires
— the more prevalent or diffuse it is —
the easier it is for a stagnant country to
begin taking advantage of it and start
the industrialization process.
Figure 5 shows some results
from simulating Lucas’s basic model,
reproducing some figures from his
article. Panel A shows how the model
behaves for a sample of four countries
that start growing at different times:
1800, 1850, 1900, and 1950. The first
country that begins to grow (shown
in blue) – the leader country – grows
at a rate of 2 percent. The countries
that begin to grow later (shown in light
blue, black, and grey) do so at progressively faster rates initially because their
income gap with the leader is wider,
the longer they remain in stagnation.
Over time, the income per capita in
these countries just about catches up
with the leader’s per capita income.
Panel B plots the model’s prediction
about the fraction of countries that
have entered the modern growth phase
for the period 1800 to 2000.14 By 2000,

FIGURE 5
Results from Simulating Lucas Model*
Panel A
Income Paths, Selected Economies
Per Capita Income x $100

35
30
25
20
15
10
5
0
1800

1831

1862

1921

1890

1951

1982

Panel B
Fraction of Economies Growing by Year
Fraction

1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
1800

1831

1862

1890

1921

1951

1982

Panel C
World Growth Rate and Income Variability
Percent

3.5
3

Annual Growth

2.5
2
1.5

Log Standard Deviation

1
0.5
0

The model calibration is the same as in
Lucas’s 2000 paper.

1800

14

30 Q2 2008 Business Review

*

1831

1862

1890

1921

1951

1982

2012

2043

2074

Figures recreated by author based on Lucas (2000)
www.philadelphiafed.org

about 90 percent of the world’s economies have entered the modern growth
phase. Panel C shows the implied
world annual average growth rate and
a measure of cross-country inequality. The annual average growth rate
is simply the average rate of growth of
all the economies in the model. The
income inequality measure plotted is
the cross-country standard deviation of
incomes (in logs).
The key figure is Panel C. It shows
that the growth of world average annual income peaks at a rate a bit above
3 percent around the 1970s and then
begins to decline. Average annual
growth can exceed 2 percent (the
growth rate of the technology frontier)
because countries that start to grow
later than 1800 get to grow faster than
the leader country in order to close
the income gap. The model is set up
such that long-run annual growth is
2 percent for all growing countries,
so eventually world average annual
growth will stabilize at 2 percent when
all countries are growing at the same
rate as the leader country. The income
inequality measure also peaks in
the 1970s. It starts at zero, when no
country has yet started to grow and all
have the same per capita income. It
then rises until the 1970s, after which
it declines; eventually it will return to
zero when all countries are growing at
a rate of 2 percent. Thus, the calibrated model predicts that the maximum
dispersion in per capita income across
countries occurred around 1970, and
inequality will subsequently decline
as more and more countries begin to
experience modern growth and fewer
are left in stagnation.
We see then how the model helps
us think about the plausibility of
the “pause before a return to further
inequality” story and the “shape of
things to come” story mentioned
earlier. The basic force in the model is
that industrialization diffuses through-

www.philadelphiafed.org

out the world economy and eventually
all countries enter the club of growing
economies. What happened in the
past guides the model’s predictions
about what will happen in the future.
Eventually all countries get in on the
act, and incomes become less and less
unequal across countries. In this sense,
the stability of the income distribution
over the past 40 years or so shouldn’t
be extrapolated into the future to

Each country
is unique and
the particular
combination of
factors that will push
an economy over the
threshold to sustained
growth varies from
country to country.
predict permanent large income differences across countries.
Of course, the basic model is very
simple and lacks many features that we
expect to influence the growth paths
of individual economies and the transition of stagnant economies to growth
economies. For example, no business
cycles are built into the model: Once
countries start to grow, they continue
to grow without experiencing recessions. There are no disasters like wars
that could have long-lasting effects on
growth paths. In addition, the model
does not incorporate the demographic
transition we spoke of earlier or capital
flows across countries. Despite these
omissions, the model is basically in
accord with the evidence on crosscountry income growth since 1800.
Importantly, it provides a plausible
guide for thinking about the future
of the world income distribution, one
that is optimistic about the prospects.
It may take a long time, but eventually

all countries move from pre-industrial
to post-industrial.
POLICY PRESCRIPTIONS
The Lucas model assumes that
poor countries will eventually develop the environment necessary for
sustained growth in per capita income
to begin. It does not offer prescriptions
for policymakers on how to make that
happen. What can policymakers in
poor countries do to push their countries into the sustained-growth phase?
This question has generated enormous
interest on the part of economists and
policymakers. Perhaps the best that
can be said is that each country is
unique and the particular combination
of factors that will push an economy
over the threshold to sustained growth
varies from country to country. However, rich countries do appear to share
some common factors that suggest
directions for policymakers in poor
countries.
Recent work by Stephen Parente
and Edward Prescott examines the
question of why some countries are
richer than others. They come to
the view that cross-country differences in incomes can be traced to the
differential knowledge that societies
apply to the production of goods and
services. It’s not that societies differ
fundamentally in the knowledge available to them; that knowledge is largely
available to all countries by observing
the methods and practices of advanced
countries, by trading with advanced
countries, or by licensing advanced
technologies. Rather, Parente and
Prescott conclude that poor countries
do not fully exploit the existing stock
of usable knowledge because poor
countries implement policies that constrain work practices and hinder firms’
ability to implement more advanced
production methods. These barriers
are often put in place to protect the
vested interests of entrenched groups.

Business Review Q2 2008 31

Parente and Prescott reach this
conclusion in part because they find
that differences in savings rates across
countries are unable to account for differences in international incomes. This
is so even when they define savings
broadly to include intangible capital
and human capital. Intangible capital
includes things like expenditures allocated to developing and launching new
products and expenditures on increasing the efficiency of existing practices.
Human capital includes the acquired
knowledge that individuals obtain
from education and on-the-job training. In Parente and Prescott’s reading,
differences in savings rates across
countries can account for only a small
portion of the international differences
in incomes we observe.
In a famous 1988 paper, Robert
Lucas laid out a theoretical case for the
role of human capital in the analysis of
growth and development. Subsequent
research, including that of Parente
and Prescott, failed to find a large
role for disparities in human capital
in accounting for income differences.
However, recent work by Rodolfo
Manuelli and Ananth Seshadri and by

Andres Erosa, Tatyana Koreshkova,
and Diego Restuccia re-evaluates the
role that differences in human capital
across countries play in accounting for
income differences. Manuelli and Seshadri argue that when one measures
human capital correctly, paying careful
attention to differences in the quality
of human capital across countries, it
turns out that differences in human
capital play a large role in accounting
for income disparities. Indeed, in some
versions of their model, the differences
in the stock of usable knowledge across
countries that was emphasized by
Parente and Prescott play little, if any,
role in accounting for differences in
income across countries; the difference
is largely accounted for by accumulated factors: labor and capital broadly
defined to include human capital. In
Manuelli and Seshadri’s model, policymakers would do well to find ways to
facilitate human capital accumulation
among the residents of poor countries.
Erosa, Koreshkova, and Restuccia
similarly emphasize the role of human capital in accounting for crosscountry income differences. But their
model differs from that of Manuelli

and Seshadri in important details,
such as allowing wage growth over
people’s lifetimes to be influenced by
investment in physical capital and by
technological progress. They argue
that technology is strongly amplified
by the stock of human capital. Their
finding of significant differences in the
average level of human capital across
countries then goes a long way toward
explaining cross-country differences in
per capita incomes.
The somewhat opposing findings of Parente and Prescott vs. those
of Manuelli and Seshadri and Erosa,
Koreshkova, and Restuccia highlight
some of the difficulties researchers face
in studying cross-country income differences. The data on growth and development experiences of countries do
not point to an obvious, unique recipe
for making poor countries rich. The
development process is the outcome of
a complex mixture of policies and institutions. Securing property rights and
creating a level playing field are likely
to help individuals and firms acquire
the capital and technology required to
push an economy onto a sustainable
growth path. Br

Heston, Alan, Robert Summers, and Bettina Aten. Penn World Table Version 6.2,
Center for International Comparisons of
Production, Income and Prices at the University of Pennsylvania (September 2006).

Lucas, R.E., Jr. "The Industrial Revolution: Past
and Future," Federal Reserve Bank of
Minneapolis Annual Report (2003).

REFERENCES
Boarini, Romina, Asa Johansson, and Marco Mira d’Ercole. “Alternative Measures
of Well-Being,” OECD Working Paper 33
(2006).
Boserup, Esther. Population and Technological Change: A Study of Long-Term Trends.
Chicago: University of Chicago Press,
1981.
Eros, Andres, Tatyana Koreshkova, and
Diego Restuccia, “How Important Is Human Capital? A Quantitative Theory
Assessment of World Income Inequality,”
manuscript, 2007.
Hansen, Gary D., and Edward C. Prescott.
“Malthus to Solow,” American Economic
Review, 92:4 (2002), pp. 1205-17.
32 Q2
Q2 2008
2008 Business
Business Review
Review ww.philadelphiafed.org
32

Khan, Auhbik. “The Industrial Revolution
and the Demographic Transition,” Federal
Reserve Bank of Philadelphia Business Review (First Quarter 2008).
Lucas, Robert E. Jr. “On the Mechanics of
Economic Development,” Journal of Monetary Economics, 22 (1988), pp. 3-42.
Lucas, Robert E. Jr. “Some Macroeconomics for the 21st Century,” Journal of Economic Perspectives (Winter 2000), pp. 159-68.

Maddison, Angus. Contours of the World
Economy, 1-2030 AD: Essays in Macroeconomic History. Oxford: Oxford University
Press, 2007.
Manuelli, Rodolfo, and Ananth Seshadri.
“Human Capital and the Wealth of Nations,” Working Paper, University of Wisconsin-Madison.
Parente, Stephen L., and Edward C.
Prescott. Barriers to Riches. Cambridge,
MA: MIT Press, 2002.
Sala-i-Martin, Xavier. “The Disturbing
‘Rise’ in Global Income Inequality,” NBER
Working Paper 8904 (2002).