View original document

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

Is an Infrastructure Crisis L o w erin g
the Nation’s Productivity?
The Determ inants of Consum er
Installment Credit
M easuring Labor Market Dynamics:
Gross F low s of W o rk ers and Jobs

J & R X S K o t'


Federal R eserve B ank o f St. Louis
R eview
November/December 1993

In This Issue . . .
Is an In frastru ctu re Crisis L o w e rin g the Nation's Productivity?
John A. Tatom
Has the United States allowed its public infrastructure to decline? More
importantly, has such a decline lowered the nation’s productivity? John A.
Tatom describes and evaluates the currently popular view that the an­
swer to these questions is yes. Supporters of this view advocate sharp in­
creases in federal government spending on infrastructure, with the
expectation o f a boost in the productivity o f the nation’s business sector.
Tatom also reviews the criticisms of this view, especially the fact that,
when flaws in previous statistical studies are addressed, the perceived,
positive effects of the public capital stock on business productivity vanish.
Tatom summarizes the reasons for a decline in the growth of some com­
ponents of the nation’s infrastructure in the 1970s and early 1980s, as well
as for the reversal of these trends since 1984. He shows that a slowdown
of public capital formation also occurred in Europe and Japan, presumably
for many o f the same reasons. According to Tatom, the federal capital
stock per capita has been quite steady for more than 40 years, and there
has been no connection between swings in federal aid to state and local
governments and the latter’s capital formation. Thus, dramatic boosts in
federal spending in recent years have not raised overall public capital for­
mation, but this has not been an obstacle to advances in productivity.


The Determ inants o f C o n su m er Installm ent Credit
Sangkyun Park
The behavior of consumer credit has attracted considerable attention
during the last 10 years. After growing rapidly in the mid-1980s, consumer
installment credit declined in many quarters during 1991 and 1992.
Changes in consumption expenditures, however, may not fully explain
the wide fluctuations in the growth of consumer credit.
Sangkyun Park examines both short-term fluctuations and the long-term
trends of consumer installment credit in relation to economic and institu­
tional factors. Between 1970 and 1992, particularly significant factors ex­
plaining the growth of consumer installment credit were the emergence
of home equity lines of credit, the difference between the real after-tax
interest rate on consumer credit and the return on household assets, and
consumers’ confidence about the future. This finding suggests that the
Tax Reform Act of 1986, which contributed to the emergence o f home
equity lines of credit and raised the real after-tax interest rate on conven­
tional consumer credit, played a significant role in slowing down the
growth of consumer installment credit in the early 1990s.




M easurin g L a b o r M arket Dynam ics: G ross F lo w s of W o rk e rs and Jobs
Joseph A. Ritter
Gross flows—the creation and destruction of specific jobs or the move­
ment of workers into and out of employment—are the immediate out­
comes of labor market processes. Firms create and destroy jobs. Workers
enter and leave employment. Usually all such developments are con­
densed into a single number, the net change in employment data. In this
article Joseph A. Ritter investigates several measures o f gross flows,
which reveal some striking features of U.S. labor markets and suggest
new perspectives on how the economy operates.

All non-confidential data and programs for the
articles published in Review are now available
to our readers. This information can be ob­
tained from three sources:

1. FRED (F e d e ra l R e se rv e E conom ic Data),
a n electron ic b u lle tin b o a rd service.
You can access FRED by dialing 314-6211824 through your modem-equipped PC.
Parameters should be set to: no parity,
w ord length = 8 bits, 1 stop bit and the
fastest baud rate your modem supports,
up to 14,400 bps. Information will be in
directory 11 under file name ST. LOUIS
REVIEW DATA. For a free brochure on
FRED, please call 314-444-8809.


2. T h e F ed eral R e se rv e B an k o f St. L o u is
You can request data and programs on
either disk or hard copy by writing to:
Research and Public Information Division,
Federal Reserve Bank o f St. Louis, Post
Office Box 442, St. Louis, MO 63166.
Please include the author, title, issue date
and page numbers with your request.
3. In t e r -u n iv e r s it y C o n s o r t iu m f o r
P o lit ic a l a n d S o c ia l R e s e a r c h
(IC P S R ). M em ber institutions can r e ­
quest these data through the CDNet
O rder facility. Nonm em bers should
w rite to: ICPSR, Institute fo r Social
Research, P.O. Box 1248, Ann A rb or, MI
48106, or call 313-763-5010.


John A. Tatom
John A. Tatom is an assistant vice president at the Federal
Reserve Bank of St. Louis. Jonathan Ahlbrecht provided
research assistance.

Is an Infrastructure Crisis
Lowering the Nation}s

' l Ml' STATE OF THE NATION S public capital
stock and its importance to the nation’s overall
economic well-being have become the subject of
widespread speculation, investigation and con­
cern. This concern has been reinforced by a
decline in the rate of growth o f the public sec­
tor capital stock that began in the 1970s. This
decline, some analysts argue, caused stagnation
of U.S. productivity growth and a correspond­
ing decline of the nation's standard of living and
in its international competitiveness.1 These ana­
lysts conclude that increased federal infrastruc­
ture spending is an urgent national priority with
high expected returns. Their view is referred to
as the infrastructure deficit hypothesis below.
Candidates for the presidency in the 1992 elec­
tions lent credibility to this view and expressed
strong commitment to boosting infrastructure
spending.2 (The Clinton Administration's infra­
structure program, called “Rebuild America” and
announced in February 1993, is described in the
1See especially Aschauer (1989b,c) and Munnell (1990b).
Aschauer has referred to the infrastructure problem as the
nation’s third deficit (presumably along with federal budget
and trade deficits), hence the view supported by his work
is referred to as the infrastructure deficit here. Reich (1991)
provides a useful summary of the view that there is an
infrastructure deficit and that attention to it should be a
central national priority.
2A more traditional view of the role of public capital forma­
tion emphasizes its use a countercyclical tool for altering

shaded insert on page 14.) This article reviews
the claims made by proponents of the infra­
structure deficit view and the evidence against it.

Infrastructure refers to the relatively large
physical capital facilities and organizational,
knowledge and technological frameworks that
are fundamental to the organization of commu­
nities and their economic development. It in­
cludes legal, educational and public health
systems; water treatment and distribution sys­
tems; garbage and sewage collection, treatment
and disposal; public safety systems, such as fire
and police protection; communications systems,
public utilities and transportation systems. The
federal government’s principal involvement in
infrastructure formation involves the military,
legislative and judicial functions. The compo­
nents of infrastructure in these areas largely
are not physical capital, nor is the largest physiaggregate demand, output and employment. Recent poli­
cies adopted in Japan and proposed in the European Com­
munity (EC) take this approach. See IMF (1993, p. 34) and
EC (1993) for discussions of the specific Japanese
proposals of August 1992 and April 1993, and the June
1993 EC summit proposals, respectively. The discussion in
each case emphasizes the conventional effects on
aggregate demand and employment that some analyses
suggest can arise from countercyclical fiscal stimulus



cal component of public sector infrastructure,
national defense, generally included in the dis­
cussion of the infrastructure deficit hypothesis.
In this article, public infrastructure—or the pub­
lic capital stock—is defined narrowly to include
only tangible, nonmilitary public capital goods.3
The key word in describing infrastructure
above is system. Infrastructure typically re­
quires relatively large initial capital outlays to
provide services potentially to all persons in a
geographic area; its incremental services are
relatively cheaply provided to any new house­
hold. In many cases, physical units of infra­
structure capital come in relatively large and
"lumpy” units, such as highways, plants and
Table 1 provides a detailed breakdown of the
components of public capital at the end of 1992,
measured in current prices. This measure of
the capital stock, which is net of depreciation, is
an estimate of the replacement cost of capital at
current prices; it is prepared by the Bureau of
Economic Analysis of the U.S. Department of
Commerce. The constant-dollar (1987 prices) net
stock is used below to compare trends in the
volume, or quantity, of public capital. The col­
lection o f public sector physical plant and equip­
ment includes a broad range of capital goods,
some of which, especially when held under pri­
vate ownership, are not commonly thought of
as infrastructure, and some which are not related
to any special function of government.

Public vs. Private Infrastructure
Much of the infrastructure in a highly deve­
loped market economy such as the United
States’ is privately provided and managed. Some
examples include most electric and gas utilities;
communications firms, such as telephone, radio,
television and cable services; private educational
institutions; and private providers of transporta­
tion services. Similarly, local governments have
recently begun to privatize infrastructure by
selling o ff public assets, contracting for the capi­
tal services (for example, private prisons or
police), or mandating that private developers
This choice follows the practice of other researchers who
refer to this measure of public capital as infrastructure.
Other definitions, however, have been used. For example, a
narrower definition can be found in the U.S. Congressional
Budget Office (CBO, 1992), which defines infrastructure to
include highways, mass transit, rail, aviation, water trans­
portation, water supply and waste water treatment. The
study indicates that some of the excluded areas could also
be considered to be infrastructure.


provide infrastructure capital to secure develop­
ment approval. Significant private sector initia­
tives in areas like telecommunications, transpor­
tation and waste processing dominate U.S. in­
frastructure developments. Recently, mergers
and joint ventures by cable companies, the Baby
Bells, entertainment and other information and
communication firms have accelerated the de­
velopment of private communications infrastruc­
ture, especially the so-called information super­
Other nations lead the way in privatizing pub­
lic capital facilities, especially telecommunica­
tions and transportation, to promote efficiency
and improve the quality and quantity of capital
services, or simply to raise government revenue.
While relatively more of these countries' infras­
tructure has been provided by the public sector
in the past, many are currently privatizing in­
frastructure and shifting to private sector provi­
sion. Prominent examples include the Mexican
and other telephone companies, the earlier
privatization of Mexico's airlines, the privatiza­
tion of infrastructure activities in Eastern Euro­
pean countries, and recent proposals to privatize
railroads, airlines and communications systems
in western Europe.4 The Japanese telephone
company, Nippon Telephone and Telegraph
(NTT), was privatized in 1985 and Japan recent­
ly began to privatize its railways.

According to the infrastructure deficit hypoth­
esis, a decline in public capital formation began
in the early 1970s and has reduced productivity
in the nation’s business sector. In addition, the
hypothesis continues, the decline has reduced
the rate of return to private capital formation,
so that private capital formation has been
reduced as well. This, in turn, has further lo­
wered private sector productivity. Since produc­
tivity is the principal determinant of the nation’s
income per capita, the decline has caused, ac4For example, the EC leaders agreed on June 16, 1993, to
end state-owned telephone monopolies in six EC nations
by 1998, while four other countries (Greece, Portugal,
Spain and Ireland) were given an additional five years, and
Luxembourg and, perhaps, Belgium were given two addi­
tional years. See Wall Street Journal (1993) and Reuters
Limited (1993).


Table 1
The Composition of the Net Stock of Public Capital: 1992
(end of year, billions of dollars)

Nonmilitary structures
Highways and streets
Educational buildings
Other buildings
Hospital buildings
Water supply facilities
Sewer systems structures
Conservation & development
Industrial buildings
Other structures
Nonmilitary equipment

State and

















Nonmilitary structures and equipment





Military equipment
Military structures


Total (including military)





•Shares of the nonmilitary total are given in parentheses. Components may not add to total due to rounding.
SOURCES: Unpublished data provided by the National Income and Wealth Division (BE-54), Bureau of Economic Analysis,
U.S. Department of Commerce.

cording to this view, a decline in the nation’s
real income and international competitiveness.5
The growth of the nation’s stock of public sec­
tor capital slowed sharply in the late 1960s and
in the early 1970s. The infrastructure deficit
view focuses on this decline in the trend growth
of public capital.6 Figure 1 shows the net stock
of nonmilitary fixed public sector capital meas­
ured in 1987 prices. While the capital stock has
climbed steadily, its growth rate slowed in the
early 1970s. Proponents of the infrastructure
deficit view often measure the public capital
stock relative to private sector employment and
5An earlier debate over the role of infrastructure focused on
many of the same issues as the recent discussion of pub­
lic capital formation. This earlier literature concerned the
mainstream view of the indispensable role of the railroad in
19th-century economic growth. Fogel (1964) relied heavily
on the role of substitutability of competing transportation
modes in his seminal work rejecting this hypothesis. In his
analysis of the railroad’s contribution, he also addressed
the role of substitutability in both private and public provi­
sion and in financing.
T h e re is a built-in bias in the United States favoring ex­
cessive investment in public physical capital facilities,
however, which must be kept in mind in discussions of the
public capital stock. This bias occurs because the public
sector is not taxed on the return or benefits from public

show that such capital per worker fell begin­
ning in the mid-1970s. In their view, public cap­
ital yields services in private sector production
so that, like private capital, its contribution is
best assessed by measuring its quantity relative
to private sector employment. Public capital is
aimed at providing services to all residents, how­
ever, especially children and the aged. Thus, a
broader assessment is afforded by its availability
per person.
On a per capita basis, the public capital stock
(Figure 2), including the federal as well as state
and local governments, nearly doubled between
capital formation, while the private sector is taxed on the
benefits from private sector capital formation. As a result,
capital formation is restricted in the taxed private sector
and the cost of capital to the public sector is lowered,
boosting public capital. In addition, public capital formation
is often financed by the sale of tax-free bonds, which me­
ans that these projects face a taxpayer-subsidized, lower
cost of capital than private capital formation. If government
decisionmakers look only at their direct cost of finance
rather than the social opportunity cost (the before-tax
return on private capital), they will tend to invest in projects
with benefits that are worth less than their cost to taxpay­
ers. In this sense, the nation would “ overinvest” in public



Figure 1
Net Stock of Nonmilitary Public Capital
Billions of dollars (1987 prices)

NOTE: End-of-year data

Figure 2
Real Nonmilitary Government Capital Stock per Person
Thousands of dollars (1987 prices) per person

NOTE: End-of-year data.



1950 and 1971, but then showed little change
for the next 20 years, averaging about $7,600
per person, measured in 1987 prices. Such
slowing is the basis for the view that the nation
has been deficient in providing for its infras­
tructure.7 Of course, a slowing in the growth of
the capital stock may reflect a decline in de­
mand and not a deficient supply, but this is the
standard used in recent discussions o f the na­
tion’s public capital stock. Similarly, a rise in the
public capital stock may exceed or fall short of
demand growth so that, again, the change in
the stock is not strong evidence about the
optimality of the nation's public capital.
Advocates of the infrastructure deficit view
suggest that the problem arose at the federal
level and that it requires a federal solution, but
these conclusions are not supported by the com­
position of public capital and its trends. Table 1
shows that about 86 percent of the nation’s
public capital stock is held by state and local
governments; it is these governmental units that
make decisions to augment it.8 Figure 2 also
shows that federal nonmilitary capital has been
about $1,100 per person (1987 prices) since
1950. There has been essentially no upward
(or downward) trend in this level, either before
or after the early 1970’s.
Thus, it is difficult to see that federal govern­
ment holdings of capital have played a role in
the slowing in overall public capital formation.
Finally, Figure 2 suggests that public capital for­
mation resumed its growth beginning in 1984.
Thus, if the previous slowing was a problem, it
appears to have ended almost a decade ago.
7Seely (1993, pp. 35-7) suggests that a comparison with the
earlier infrastructure trend, especially in the 1960s and ear­
ly 1970s, overstates the decline. In particular, he refers to
this earlier period as the “ golden age of infrastructure de­
velopment in the United States.” He cites the beginning of
the Department of Transportation (1968), the Environmental
Protection Agency (1969), and three new laws—the Solid
Waste Disposal Act of 1965, the Urban Mass Transit Act of
1970 and the 1972 Water Pollution Control Act as the
sources of this boom.
8Seely (1993) explains that this division of responsibility
dates back to early debates over the role of the federal
government in capital formation, especially for roads. Con­
gress severely limited this role from the outset. The first
major debate involved the Gallatin Plan, a proposal for fed­
eral road development, which was proposed in 1808 and
ultimately rejected by Congress. In contrast, the role of the
state governments dates back to early road and canal
projects, especially New York Governor DeWitt Clinton’s
construction of the Erie Canal, “ Clinton’s Ditch,” which set
off a wave of development of state-owned or financed

An International Comparison
Is it really true that the U.S. has fallen be­
hind? A common criticism of U.S. infrastructure
policy is that foreign countries have more in­
frastructure and a faster pace of infrastructure
formation than the United States.9 Ford and
Poret (1991) have examined the public capitalprivate productivity link for 11 major industrial
countries and find that the evidence o f a link is
not robust. Also, criticisms of U.S. infrastruc­
ture development ignore the relatively large
U.S. private sector holdings of capital that,
abroad, would be held by the public sector; this
is especially the case in transportation, commu­
nications, and electric and gas utilities.1 It is
not possible to construct a comprehensive and
exactly comparable measure of infrastructure.
Analysts would differ over the types of capital,
and the various sectors to be included or ex­
cluded in constructing such a measure; more­
over, it is not possible to obtain comparable
detail or component measures for comparable
Table 2 provides some insight into these is­
sues. The size of public capital formation relative
to private capital formation in eight countries,
including the United States, is shown for vari­
ous five-year periods since I960, where data are
available.1 The table shows that public capital
formation has generally slowed when measured
relative to private capital formation or GDP in
all o f the countries except France. In the latest
period, public capital formation continued to
decline in all countries except the United States
and France. Thus, the decline in U.S. public
10Ford and Poret also note the distinction between public
capital and a broader measure of infrastructure by measur­
ing infrastructure narrowly as the public capital stock of
“ producers of government services,” and more broadly by
including public and private capital stocks in electricity, gas
and water, and structures in transportation and communica­
tion. They do find some support for a link between private
productivity and the broader measure in a cross-country
comparison, but they reject the public capital linkusing the
narrower measure, and find that the significant results with
the broader measure are not robust.
11The capital formation data are from the Organization for
Economic Co-operation and Development (OECD), except
that this source contained no Japanese data; these were
obtained from a national source. The countries shown are
the only major industrial countries for which OECD reports
data for most of the period. For example, data for Italy, the
only G-7 country omitted in Table 2, are reported only for

9See Aschauer (1989a), for example.



Table 2
Public Capital Formation as a Percent of Business Capital Formation



















Share of Public Capital Formation in Gross Domestic Product











NOTE: The statistics for Germany refer to western Germany. Business excludes private, non-profit institutions serving
households. The gross domestic product data are from the International Monetary Fund’s International Financial
SOURCE: Organization for Economic Co-operation and Development, National Accounts, Volume 2, and Bank of Japan,
Comparative Economic and Financial Statistics: Japan and Other Major Countries.

capital formation is not unique and, as a share
of GDP, it is small compared with the ex­
perience abroad.
Table 2 also shows that relatively more o f in­
vestment (and economic activity) is determined
by the public sector in Europe and Japan. In
most periods shown, public capital is a larger
percentage of private capital formation in all of
the foreign countries (except the United King­
dom) than it is in the United States. Public capi­
tal formation has generally been a smaller share
of GDP in the United States, according to these
data, than in any of the other countries except
the United Kingdom.1 While data from the 1960s
are only available in a few countries, the decline
in public capital formation as a share of GDP
has not been smaller abroad than in the United
States. Thus, it is hard to attribute any purported
decline in U.S. competitiveness to a relative
decline in the pace of public capital formation.
12While France shows relatively stronger public capital for­
mation, the French government has announced an exten­
sive privatization program for 21 companies, including
some of the largest in France. The program began in fall
1993. See Belsie (1993) and Riding (1993).


Two of the largest components of the nation’s
public capital stock (see Table 1) are highways
and streets, and educational buildings. While
the first is closely related to the notion of infra­
structure, it is only one part of a much larger
(and generally private) set o f capital goods that
includes automobiles, trucks, buses and trailers,
which are involved in the provision of transpor­
tation services. Educational buildings (unlike the
educational system, its organization, content,
processes and outcomes) are less related to the
infrastructure concept and also have a large pri­
vate sector counterpart. These two components—
highways and streets, and educational buildings
—account for most of the slowing in the growth
of U.S. public capital. Figure 3 shows these two
components of public capital per person and to­
tal public capital per capita excluding these two


Figure 3
State and Local Net Capital Stock per Person
Thousands of dollars (1987 prices) per person

NOTE: End-of-year data

components, called “ other” in the figure. Exclud­
ing these two components, there was little or
no slowing in public capital formation during
the 1970s and early 1980s.
Tatom (1991b) argues that there are three
principal reasons for the slowing in highway
and street, educational building, and total public
capital formation. First, the post-World W ar II
baby boom and associated temporary surge in
population growth played a major role in the
subsequent decline in growth o f the educational
buildings stock and in the earlier growth of
highways and streets, especially surrounding
cities.1 The interstate highway system, began in
the mid-1950s and largely completed by 1975,

also contributed to a temporary surge in public
capital formation and, subsequently, gave rise to
part of the apparent slowing. Second, changes
in the cost of driving played an important role
in accounting for the decline in road capital for­
mation. Following sharp increases in the price
of oil and gasoline in 1973-74 and again in
1979-80, the growth of passenger-miles driven
per person dropped very sharply. Reductions in
the growth of highway and street use reduced
the growth of this form of capital.
The third factor influencing roads, highways,
educational buildings and other public capital
formation is the price of such capital goods.
From the early 1950s to the early 1960s, the

13The decline in the growth of the stock of educational build­
ings reflects the fact that the share of the school-age popu­
lation (ages 5 to 24) rose from about 31 percent in 1949 to
about 37.5 percent in 1971, then fell steadily to less than 29
percent in 1990. Not surprisingly, public educational build­
ings per person peaked in 1974. The subsequent decline in
public educational buildings per person was smaller than
the decline in its private educational counterpart. The slow­
ing in highway and street capital formation was similarly
not unusual when compared with private capital formation
associated with road transportation; see Tatom (1993).



prices of public capital goods fell relative to the
prices of private capital goods. Consistent with
the law of demand, the quantity of public capi­
tal grew much faster than that of private capital
goods over the period. Since then, and especial­
ly since the late 1960s, the relative price of
public capital goods has climbed sharply; not
surprisingly, the demand for, and quantity of,
public capital has declined relative to private

growth rate of the total public capital stock ac­
celerated to a 1.7 percent rate and that o f the
per capita stock rose to a 0.7 percent rate. If
there was an infrastructure deficit as measured
by declining public capital per person, it ended
in 1983.

Tw o of the three factors depressing public
capital formation began to reverse in the 1980s.
The relative price of gasoline generally declined
after 1980, falling dramatically in 1986, boosting
road travel and the demand for highway and
street capital. The share of the school-age popu­
lation has nearly bottomed out as well. For ex­
ample, the population ages 5 to 19 rose from
52.4 million in 1986 to 53.2 million in 1991
(the latest year available), after declining from a
peak of 60.3 million in 1971.1 Not surprisingly,
school districts have responded to the recent
baby "boomlet” by building new schools. From
1983 to 1992, the stock of state and local high­
ways and streets and educational buildings per
person declined at a 0.2 percent rate, much
slower than their 0.9 percent rate of decline
from 1975 to 1983.

While the closer look above at the composi­
tion of public capital and its trends shows that
federal government capital formation plays a
minor role in public capital acquisition and the
overall trend, the federal government does play
a role in financing some state and local govern­
ment capital formation. This financing role
could account for some of the earlier slowing in
public capital formation. Tatom (1993) shows,
however, that changes in federal financing do
not account for the past state and local slow­
ing. The argument there uses data on overall
federal grants-in-aid to state and local govern­
ments and the latter’s public capital formation
relative to GDP. The total grant is the relevant
gauge of federal assistance for infrastructure
development because of the fungibility of funds
within state and local government budgets.1
These data are plotted in Figure 4. Most of the
post-1968 decline in the share of public invest­
ment in GDP occurred before the share of fed­
eral grants to state and local governments in
GDP peaked in 1978.1 The share of public in­
vestment showed a further slight decline from
1978 to its 1984 level of about 2 percent of
GDP, but it rose to about 2.2 percent of GDP in
1985 through 1991, despite a further decline in
federal grants until 1989.

These two changes have resulted in a resump­
tion of growth in the overall stock of public
capital per person since 1983. During the 1975-83
period, the total stock rose at only a 1.1 percent
rate, so the per capita total public stock rose at
only a 0.1 percent rate. The only period of
decline in the per capita total occurred from
the end of 1980 to the end of 1983, when it fell
at a 0.1 percent rate. From 1983 to 1992, the

14ln Tatom (1991b), the relative price of public capital to pri­
vate capital is measured by the implicit price deflators for
public nonmilitary investment and for private nonresidential
fixed investment. It declined about 13 percent from 1954 to
1960, then rose about 36 percent from 1960 to 1989. As a
result of the latter movement, in part, the quantity of
constant-dollar net stock of nonmilitary public capital fell
from about 58 percent of the constant-dollar net stock of
nonresidential private capital in 1964 to about 42 percent in
15See Council of Economic Advisers (1993), p. 381. The num­
ber of persons ages 20 to 24 continued to decline after
1986, falling 1.6 million from 1986 to 1991, so that the total
for both groups may not have reached its trough in 1991.
16Federal grants to state and local governments include fund­
ing for other programs besides capital outlays. Indeed,
such grants are nearly four times as large as federal grants
for state and local government spending on infrastructure


Did R educed Federal Funding Play
a Role?

capital. About 60 percent of federal infrastructure spending
occurs through such grants. Also, see Moore (1992) for
more detailed analysis of the growth of federal spending
for infrastructure since 1989.
17The share of federal grants for major public physical capi­
tal investment in GDP rose from about 0.2 percent in fiscal
1948-56 to 0.8 or 0.9 percent in 1976-80, then declined to
0.5 percent in 1987-92, according to the U.S. Office of
Management and Budget (1993, p. 376). This pattern es­
sentially mirrors, at a lower level, the pattern of total federal
grants to state and local governments.


Figure 4
Federal Aid to Governments and Public Investment
Percent of GDP

Since 1989, federal grants to state and local
governments have ballooned, rising from 2.2
percent of GDP in the federal government’s
1989 fiscal year to 2.8 percent in the 1992 fiscal
year. These expenditures rose from $116 billion
in fiscal 1989 to $167.8 billion in fiscal 1992 (a
44.7 percent increase); had their share in GDP
not risen, such spending would have climbed to
only $130.7 billion. Despite this $37.1 billion ex­
tra boost in funds available to state and local
governments, there has been no change in pub­
lic sector investment as a share of GDP.
Table 3 shows federal outlays aimed directly
for public capital in selected years; it also shows
GDP in each year to facilitate assessments of the
growth of nominal expenditures. The table
shows that federal outlays aimed at state and
local capital formation are much larger than
18The shaded insert to this article discusses the Administra­
tion’s perspective on the role of public capital formation.
19Bartlett (1992) describes several programs to boost state
and local government employment that failed because of

direct federal outlays. The table also shows that
recent growth has outstripped overall growth in
the nation’s GDP. From 1989 to 1993, federal
outlays grew 47.4 percent, more than twice as
much as the 21.6 percent rise in GDP over the
same period. Over the same period, grants to
state and local governments for major public
capital projects rose 39.6 percent, still nearly
twice the percentage increase in GDP. Despite
this surge, public investment as a percent of
GDP changed little after 1989.1 The overall
boost in federal government funding was appar­
ently offset by reduced funding at the state and
local levels.
Such a substitution in funding is not unusual.
Earlier attempts by the federal government to
boost specific state and local spending compo­
nents have met with such substitutions.1 Thus,
such substitution effects. The CBO (1986) provides strong
evidence of such substitutability. Also, see CBO (1988),
Gramlich (1978) and Jondrow and Levy (1984) for discus­
sions of this phenomenon.



Table 3
Federal Outlays for Major Public Physical Capital: Selected Years
(billions of dollars)

Grants to state
and local








*GDP and budget data here are for fiscal years.
SOURCE: Budget data for 1960-90 are from The Budget for Fiscal Year 1991.
Budget data for 1991 are from The Budget for Fiscal Year 1993.
Budget data for 1992-94 are from The Budget for Fiscal Year 1994.

it is difficult to ensure that targeted assistance
programs, such as those for public capital for­
mation, will result in net increases in spending.
New federal funding for state and local govern­
ment capital spending may well finance projects
that would have been done in any case; in turn,
the savings these federal grants generate are
used to fund more pressing current expenses
instead o f new capital formation.

The principal damage attributed to the infra­
structure deficit, according to proponents of
this view, is that it led to the stagnation of pri­
vate sector productivity beginning at nearly the
same time as the slowing in public capital for­
mation. Statistical estimates by Aschauer (1989c)
and Munnell (1990b) indicate that the public
capital stock has an unusually large effect on
private sector output, given the use of fixed
amounts of private sector resources.2 Criticisms
20Ratner (1983) had obtained this result earlier, although his
estimate was somewhat smaller. Holtz-Eakin (1988) also ob­
tained similar results, although later (1992 and 1993) he
presented more detailed analysis for cross-section and
time-series results that rejected the hypothesis.


of these estimates have arisen for several rea­
sons outlined below.

The Benefits o f Public Capital
Are N ot Necessarily Reflected in
Business Output
Both Aaron (1990) and Musgrave (1990) criti­
cize the Aschauer (1990) discussion for ignoring
the fact that most of the services o f public capi­
tal have no effect on measured national output,
not to mention measured business sector output
or, even more to the point, business sector
productivity. Similarly, Aaron insists that "the
argument that public sector investments con­
tribute massively to measured national output is
not strengthened by arguing that such invest­
ments contribute to items that do not appear in
measured output” (p. 59). Even in the case of
investments in airports or highways to reduce
congestion costs, there are other benefits to the
public besides increased efficiency of work and,
therefore, greater business output. Time savings
due to reduced congestion could result in in­
creased work time and business output, but this


would not necessarily boost their ratio, or busi­
ness sector productivity.2

Public and Private Capital Are
Proponents of the link between private produc­
tivity and public capital tend to ignore substitut­
ability in public and private capital services.2
Increased highway stocks, for example, could
raise the rate of return to trucking firms, but
these gains come, in part, at the expense of low­
er social and business returns to public and pri­
vate capital in water, rail and air transport.
Public projects involving locks and dams, air­
ports or roads produce services that are likely
to be substitutes for each other and for private
capital services as well. The presence of such a
substitutability relation reduces the expected
returns from public capital formation and leads
to offsetting reductions in the other components
of the public capital stock and in the private
stock. More importantly, these substitutions
offset, in part, any gain in private output
directly associated with a rise in one component
of public capital.
Over the period 1929 to 1991, the growth
rates (continuously compounded) of the end-ofyear public and business sector capital (meas­
ured by the constant-dollar net stock of fixed
nonresidential private capital) stocks have a con­
temporaneous correlation coefficient of -0.287,
which is statistically significant at a 95 percent
confidence level. Even within the government
total there is no evidence of complementarity,
21Tatom (1993) suggests that reduced congestion and travel
cost could lower real wages, leading to a lower level of
marginal and average productivity of labor. The latter effect
requires that the marginal and average productivity of labor
are proportional and that employment rises. Output, em­
ployment and the typical consumer’s standard of living all
rise in this analysis, but private sector productivity does
22Aschauer (1989b) claims that public and private capital are
complements, although he finds evidence that public capi­
tal formation crowds out private capital formation, dollar for
dollar. Aschauer’s estimates using data on capital stocks,
however, indicates that the stocks are complements. The
apparent inconsistency in these results is not explained.
Eisner (1991) tests the relationship of private capital invest­
ment to public capital investment using Munnell’s data
(1990a, 1990b) in an accelerator model. He finds evidence
that public and private capital are not related, although a
negative public capital formation coefficient (-0.07) found
in the time-series data for 48 states suggest a substitute
relationship and this effect is only marginally not statistical­
ly significant (t = -1.75) at a 5 percent level. Erenburg
(1993) reports results supporting the complementarity
hypothesis, although the statistical significance of the
result is not indicated. The Erenberg result arises for pri­
vate equipment, but not for private structures.

as the growth rates for the federal stock and
the state and local stock have a correlation
coefficient of -0.143; this negative relationship is
not statistically significant at the 95 percent
level, however.

Estimates o f the Size o f the Public
Capital Stock Effect
The Aschauer/Munnell estimates have been
widely criticized as being implausible because of
their sheer magnitude. Aaron (1990) focuses on
their implausible real rate of return estimates
for components of the capital stock. The real
rate of return (ignoring nonmarket benefits) to
private (public) capital equals what is called the
marginal private sector product, or the contri­
bution to private sector output of an additional
unit of private (public) capital. Aaron points out
that Aschauer's estimates imply a real rate of
return to some components of public capital
that is about five times that of private capital
(146 percent vs. 21 to 29 percent). Such esti­
mates imply that moving a dollar of private in­
vestment spending to such public investment
would boost output by more than $1.15 per
More recently, Aschauer (1993) has scaled
back his estimate of the effect of public capital
on private sector output. He now assumes that
this marginal product of public capital is the
same as that of the marginal product of private
capital.2 As a result, he has also scaled back his
earlier claim that the slowdown in public capital
formation accounted for all of the slowing in
23lf government decisionmakers maximize the value of the
nation’s resources, the opportunity cost of public capital
would be the private rate of return on private capital adjust­
ed upward up to reflect capital income taxation. Then the
marginal benefit of public capital would be equal to the
marginal product of private capital services. In practice,
however, the cost of capital used in public decisions tends
to be lower than it is for the private sector; thus, it is more
likely that the marginal product of public capital is lower
than that of private capital. See the discussion of the the­
ory and practice of cost-benefit analysis in Musgrave and
Musgrave (1989). Moreover, even if it were the case that
the public sector decisionmaker equates the public cost of
capital to the marginal benefit of public capital, the latter is
composed of marginal nonmarket benefits plus any margi­
nal private sector product of public capital. Thus, to the ex­
tent that government capital yields direct services to
consumers, the marginal private sector product of public
capital will be less than the marginal private sector product
of private capital. When public decisionmakers pursue pri­
vate benefits of their own, there is an additional incentive
to “ overinvest” in public capital by using a lower cost of
public capital than otherwise.



The Shifting Policy Perspective on Public
Capital Formation
Aaron (1990) has pointed out that one of
the peculiarities of the infrastructure debate
is how readily the evidence supporting the in­
frastructure view was accepted, despite the
implausibility o f the estimates. He argues that
such evidence is welcome relief to those ana­
lysts who are "sick and tired—with good rea­
son, in my view—of continuous and unsup­
ported allegations that everything the govern­
ment does is wasteful or harmful” (p. 62). He
also points out how welcome the kind of re­
sults in Aschauer and Munnell are to groups
who stand to gain from expanded infrastruc­
ture spending.
The two leading candidates for president in
the 1992 election proposed increased federal
infrastructure spending as critical priorities.
For example, the Clinton campaign planned
to boost federal infrastructure spending by
$20 billion each year from 1993 to 1997 as
part of a $50 billion per year program to “put
America back to work—the most dramatic eco­
nomic growth program since World W ar II.”
The centerpiece of this growth program, called
“ Rebuild America,” included the extra $20 bil­
lion per year to rebuild America and develop
the world's best communication, transporta­
tion and environmental systems.1 Despite the
public attention to the problem of the na­
tion's physical infrastructure, the proposed
program represents a smaller overall boost in
spending than the rhetoric might suggest.
More importantly, the plan also contains a
major shift toward technology developments
(instead of physical capital formation) that are
traditionally the province of the private
The table shows the Clinton Administra­
tion's path o f planned new spending for Re­
build America from 1994 to 1998.2 Note that
the proposed additional spending in each year

1See Clinton and Gore (1992).
2The figures in the table are from Clinton (1993). Data
from the Administration’s budget released on April 8,
1993, differ slightly. Aschauer apparently agrees with
the shift in emphasis or broadened view of infrastruc­
ture described here. According to the Washington Post
(1992), "Aschauer stresses that not all public investment


does not reach $20 billion until 1998, and
only then by including tax incentives for pri­
vate capital. More importantly, most of the
nearly $70 billion of increased "investment”
in 1994-98 are not for public capital and in­
frastructure. Only about $27 billion of this to­
tal is for traditional public infrastructure
spending in transportation, environmental,
rural and community development (which
also include funds for some housing stock re­
habilitation programs). While this total ex­
cludes public housing and spending for new
technology development (including short-haul
aircraft research, high-performance comput­
ing, information highways, alternative fuels,
vehicles and research on natural gas, and fu­
sion energy), it includes items that would
result in little new physical capital formation,
such as spending for smart cars/smart high­
way research, alcohol-related safety pro­
grams, and community development banks.
Proposed spending from 1994-98 also shows
a significant change in the composition of
federal programs. The current administration
plans to devote an increasing share o f spend­
ing to innovative activities. The emphasis in
infrastructure spending has shifted to infor­
mation networks and the technologies of the
future, which are largely private sector activi­
ties in the United States. The technologies for
fast computing, information highways and
fiber-optic networks already exist and innova­
tions are being implemented in the U.S. pri­
vate sector (for example, through the use of
existing super computers, the Internet sys­
tem, private phone systems, and satellite com­
munication centers, respectively). Expansion
of fiber-optic networks and the introduction
of information highway systems by the Baby
Bells represent continuing efforts to extend
the use of the processes. There is even a pri-

will produce a big long-term return. For example, he op­
poses pork barrel highway projects, but strongly sup­
ports government investment on advanced technologies,
such as ‘intelligent highways.’ ”


The Clinton Budget Plan: Rebuild America Public Capital Formation
(billions of dollars)1


Rebuild America:
total increase in
public investment


Total, excluding
energy, housing
and technology2


$1.2 ($1.3)
4.1 (4.4)
6.0 (6.8)
7.4 (8.6)
8.2 (9.9)
26.9 (31.1)

$0.7 ($1.9)
2.3 (3.8)
3.4 (5.2)
4.2 (6.2)
4.5 (6.7)
15.1 (23.7)

1Numbers in parentheses are total investment outlays including tax incentives,
in clu d e s transportation, environmental, rural and community development.
NOTE: Numbers do not add to total due to rounding.
SOURCE: Clinton, A Vision of Change for America.

vately owned and funded high-speed rail sys­
tem in the planning stage in Texas.
One indicator of the shift to competition
with private sector activities is indicated in
the table. Revitalizing technology, about onethird of Rebuild America, includes substantial
sums for research and development in main­
stream private sector activities. Most o f the
energy outlays and a substantial fraction of
transportation, environmental and community
development outlays are for similar projects.
Ironically, as Butler (1992) shows, there is no
evidence that total R&D, or public R&D, in­
vestments slowed in the late 1970s and 1980s,
especially to the extent o f the slowing in physi­
cal public capital.
The CBO (1991) discusses the federal role
in spending on research and development,
pointing out that returns to academic and
basic research (about one-fourth of federal
R&D spending) has significant positive effects
on private productivity but that (except in
health and agriculture) there is no consistent
evidence o f significant returns from federally
applied R&D. Similarly, Griliches (1988) re­

jects a role for R&D movements in explaining
the slowdown in private productivity growth.
CBO (1991) refers to the fact that production
function estimates find no consistent, positive
effects of federal contract R&D expenditures
on productivity as a "major puzzle."
The CBO argues that public R&D could
crowd out private R&D, but that this inter­
action probably has only small overall impor­
tance.3 In research on new technology, how­
ever, government funding can crowd out
private R&D because the returns to new tech­
nology are not as easily captured by private
entrepreneurs when the technology is funded
by the government. In addition, government
investment crowds out private activity by
competing for scarce specialized research
resources. The potential for the public sector
to further technological innovation is open to
serious doubt.4 While public sector R&D in­
vestments may be productive, the evidence
for a larger or smaller effect of public physi­
cal capital on business output does not have
any bearing on the purported effects on pri­
vate output or productivity of public sector
spending on research and development.

3Butler (1992) reviews trends in R&D expenditures in the
United States, Japan and Germany.
4See Cato Institute (1993), Gilder (1993) and Rodgers
(1993) for discussions of the private sector’s dominance
in invention and innovation of new high-technology



private sector productivity growth after 1970.
Now, "a non-negligible portion, perhaps around
10 percent, of the productivity slump can be ex­
plained by the lower rate of public capital ac­
cumulation” (1993, p. 13). The most recent
estimate is based on an assumption, however, not
on statistical evidence. Nevertheless, Aschauer
suggests that there is "a strong causal relation­
ship between public capital investment and
productivity and output” and that "The timeseries results suggest that, at the aggregate lev­
el, there is underprovision of public capital”
(1993, p. 22).
Munnell (1992) has also agreed with critics
who contend that "the numbers emerging from
the aggregate time-series studies are not credi­
ble.” Nevertheless, she concludes that an “evenhanded reading of the evidence—including the
growing body of cross-sectional results—
suggests that public infrastructure is a produc­
tive input which may have large payoffs.”2 Eis­
ner (1991) uses Munnell's (1990a,b) data to show
that the time-series evidence for the 48 states in
Munnell’s sample rejects the infrastructure
productivity hypothesis. The evidence for the
variation in gross state product (not private sec­
tor output) across states shows that states with
larger output have more public capital. Eisner
indicates that this is consistent with an alterna­
tive hypothesis that richer states buy more pub­
lic capital, as well as with the hypothesis in
question—that states with more public capital
produce more output.2

The Spurious Regression P roblem
The time-series estimates that show a positive
and statistically significant effect of the public
24Munnell also cites Peterson’s (1990) argument that a rela­
tively high voter-passage rate for infrastructure-related bond
issues indicates an undersupply of public capital. Of
course, voter revelations of expected net benefits from pub­
lic spending could reflect either direct, nonmarket benefits,
contributions to business output, or some combination of
the two. Peterson found that about 80 percent of bond
proposals were accepted from 1984 to 1989 by margins ex­
ceeding 66 percent on average. He argues that a “ median
voter” model would suggest approval rates and margins
closer to 50 percent if the voters consider the stock of pub­
lic capital to be about the amount desired. Tatom (1993)
cites a recent approval rate higher than 50 percent (62 per­
cent) as evidence for the reverse, that the public does not
accept unanimously or indiscriminately capital projects
offered by government officials. Recent polling results on
infrastructure demand are also reported that suggest that
such spending is a low priority compared with private capi­
tal formation or reducing the federal deficit, among other
priorities. Therefore, it appears that voters get about what
they want, just as Tiebout (1956) argued.


capital stock on private sector output do so be­
cause of a statistical fallacy called “spurious
regression.” For example, if two wholly unrelated
measures have similar time trends, then they can
exhibit an apparent, statistically significant rela­
tionship between them when no economic rela­
tionship, in fact, exists. In the infrastructure
case, the spurious regression problem can be
observed in the relationship of private productiv­
ity—business output per hour—and the stock of
infrastructure per hour. Both showed relatively
strong upward trends from the late 1940s to
the early 1970s and then each trend declined
sharply (see Figure 5).
Since the early 1980s, the evidence on the lev­
els of private output and public capital per hour
is considerably weaker. In particular, private
productivity accelerated sharply, rising at a 1.7
percent annual rate from 1982 to 1988; mean­
while, the growth of the stock of public capital
per hour actually slowed further, falling at a 1.5
percent annual rate from 1982 to 1988, down
from a 0.5 percent rate of increase from 1971
to 1982.2 The public capital stock per hour
then began to rise, growing at a 1.7 percent
rate to 1991, while private productivity growth
slowed to a 0.3 percent rate. Thus, the two
measures w ere negatively related from 1982
to 1991. In 1992 both measures accelerated.
The spurious regression problem in Figure 5
is easily illustrated using simple correlations.
The level o f business output per hour and of
public capital per hour are strongly and posi­
tively correlated from 1947 to 1992; the cor­
relation coefficient is 0.95, consistent with a
strong, but potentially spurious, relationship.
The correlation between the growth rates of
25Others have found conflicting evidence using crosssectional data. For example, Holtz-Eakin (1992) provides
cross-sectional evidence that rejects the infrastructure
productivity hypothesis. More importantly, he explains that
the existence of a cross-sectional effect does not support
an aggregate national effect because the former would im­
ply an interregional substitution of output, as private
resources migrate from other regions. Eberts (1986),
(1990a), (1990b) and Garcia-Mila and McGuire (1992) find
evidence for a cross-sectional effect of public capital on
output, although the latter study is supportive only for high­
ways and education, and the effects are relatively small.
26While the capital stock per person rebounded after 1983,
the capital stock per hour did not begin to rise until later.
This reflects the faster growth of the labor force than of the
population until the end of 1988.


Figure 5
Business Sector Output per Hour and Public
Capital Stock per Hour*

Capital Stock

* Measured in dollars per hour (1987 prices)

the two series, however, is not statistically sig­
nificant. The correlation coefficient for the
growth rates (1948 to 1992) of 0.15 is well
below the critical value o f 0.29 at a 95 per­
cent confidence level. Thus, simple correlation
analysis rejects the hypothesis that a contem­
poraneous rise or fall in the amount of public
capital per hour raises or lowers business sector
Studies o f the link between public capital and
private sector output do not use such a limited
two-variable comparison. Instead, they attempt
to control for other factors that determine ag­
gregate private sector production, such as pri­
vate sector hours (h) and the flow of private
capital services (k ). The basic statistical model
used by Aschauer, Munnell and others is a
production function estimate of the form:

(1) In Q / k t = In A + a In ( h / k j
+ 6 In (KG/KJ + rt + t
where Q, is business sector output in period t,
and KG and K are the public and private capital
stocks, respectively, and £ is a normally and in­
dependently distributed random disturbance
term. The scale parameter A , the rate of disem­
bodied technological change, r, and the output
elasticities, a and 6, are estimated using ordi­
nary least squares regression.2 The critical
parameter for the public capital-private produc­
tivity hypothesis is d, which is hypothesized to
be positive.
Estimation of such an equation requires that
all of the variables entering equation (1) must
have certain statistical properties for the esti­

27The specific derivation of (1) can be found, for example, in
Tatom (1991a), in which the effect of energy price shocks,
operating through energy use, capital obsolescence and
pure technology alterations is also included, as is a quad­
ratic time-trend specification.



mate to be a meaningful long-run relationship.
In particular, the estimation of the parameters
in equation (1) requires that the error term
have a distribution with a mean o f zero and
constant variance. Thus, the linear combination
of variables that equals t also must have such a
distribution. This requirement is satisfied if each
of the variables entering equation (1) is station­
ary, which means each must have a tendency to
revert to its own fixed mean. In this case, each
measure is said to be integrated of degree zero,
1(0). Alternatively, each series can be integrated
o f a common degree, typically one 1(1), meaning
that each must be stationary when differenced
once or, in general, differenced the number
times indicated by the degree of integration.2
In the latter case, a linear combination of the
variables entering equation (1) can be stationary
if the variables are cointegrated.
The spurious regression problem in estimates
of equation (1) arises from the fact that for the
post-World W ar II periods used in studies of the
public capital stock effect, the public capital
stock variable and the term involving it in equa­
tion (1) are integrated of order two, or 1(2),
which means that either measure must be
differenced twice to be stationary. The depen­
dent variable in the production function is in­
tegrated of order one and ln (h /k ) is also 1(2) for
the period studied in Tatom (1991a). Taken
together, these properties imply that a linear
combination of the levels of the variables, like
the linear combination equal to £ in equation
(1), cannot be stationary, as required by statisti­
28lf all the variables in an equation are integrated of the
same degree, they are potentially cointegrated, regardless
of the degree. Lynde and Richmond (1993) use the meas­
ure In KG in level estimates like equation (1) to test the
public capital stock effect for two periods, 1948-89 and
1958-89. In both periods In KG is l(2) according to tests by
the author, although Lynde and Richmond do not note this
problem. Moreover, the method they use to address poten­
tially spurious regressions, employing what are called
Phillips-Hansen estimators, does not remove the possibility
of spurious outcomes when the included variables are in­
tegrated of mixed order with one or more variables that are
l(2), or are integrated of even a higher order.
29A cointegration test which avoids this issue is used in Tatom
(1991a). In this test, the public capital stock has a negative
but statistically insignificant effect on business sector out­
put and productivity.
30Some of the studies that have noted this fragility include
Aaron (1991), Holtz-Eakin (1988, 1992, 1993), Hulten and
Schwab (1991), Jorgenson (1991), Rubin (1991) and Tatom
(1991a). Finn (1993) uses the same method as Lynde and
Richmond and tests various components of the public capi­
tal stock. Her evidence supports the view that only the
highway component of the public capital stock is statistical­
ly significant. Her preferred measures using highway and
street capital in place of public capital suffer from the


cal theory. An estimate of such an equation can
result in the appearance of statistically signifi­
cant relationships when, in fact, the variables
are not related.2
There are well-known statistical methods for
assessing whether the spurious regression
problem is present and for removing its in­
fluence on statistical results. In this case, simply
first-differencing the data and including a time
trend in the estimate eliminates the problem be­
cause the growth rates of the two 1(2) variables
are trend-stationary. First-differencing the data
means that the effect is estimated using data on
changes (growth rates) in private sector produc­
tivity and the public capital stock, along with
growth rates of other factors influencing the
level of private sector productivity. Firstdifferencing earlier production function esti­
mates that include the public capital stock yields
estimates of the public capital effect that are
not statistically significantly different from
zero.3 Both Munnell (1992) and Aschauer (1993)
assert that first-differencing is inappropriate; for
example, Munnell states that this operation "des­
troys any long-term relation in the data” (p. 193).
But first-differencing equation (1) simply results
in the continuous growth rates of each measure
replacing the level measure of each variable; the
required quadratic and linear-trend terms are
replaced with a linear-trend term only and a
first-order, moving-average error term is in­
troduced. In differencing equation (1), the
parameters are unaffected. Hence, if they are
viewed as the appropriate long-run parameters
same lack of stationarity for the growth rates of highway
capital or of the ratio of public highway to private capital.
Thus, her results are also spurious. This is not surprising
since the time-series plots of public capital—in total, per
capita, or per unit of private capital— mirror those of high­
way capital. The cointegration test like that reported in Tat­
om (1991a) for the first-difference version of an equation
like equation (1) rejects the statistical significance of the
highway stock.


in a levels estimate, they will remain so in first
differences. Munnell (1992) recognizes the
problems posed by nonstationarity and recom­
mends testing for cointegration; Tatom (1991a)
provides such a test and rejects the hypothesis.3

An Alternative View: R everse
There is an alternative view that suggests a
positive link between private productivity and
the stock of public capital per worker. Eisner
(1991) suggests the fact that regions with rela­
tively high productivity have relatively higher
infrastructure, and simply may reflect an effect
of income on the demand for and quantity of
public capital.
A statistical test of whether higher productivi­
ty causes more public sector capital formation,
or the reverse is true, employs "Granger causali­
ty.” In these tests, causality means a statistically
significant temporal relation in which changes
in one measure temporally are followed by
statistically significant movements in the other
measure. It is possible, in principle, for each
measure to "cause” the other, for neither to
cause the others, or for only one measure
to cause the other.
Tatom (1993) provides a test of Granger
causality for the productivity-public capital for­
mation link.3 The test uses annual data (1949 to
1991) for the public capital stock or public sec­
tor investment and for the private sector’s total
factor productivity, the latter being output per
31Both Munnell and Aschauer also criticize the Tatom (1991a)
results for including an energy-price effect. Aschauer (1993)
emphasizes that including an energy effect in a production
function for a value-added output measure is inappropriate
because energy is an intermediate input. But the insig­
nificance of the public capital stock holds independently of
whether the energy-price effect (or a quadratic trend) is in­
cluded. Furthermore, a public capital stock effect also
arises through intermediate input services to private firms.
McMillan and Smyth (1993) include an energy-price meas­
ure in a vector autoregression model. They also reject the
public capital-private productivity link posited by Aschauer
and Munnell. The relative price of energy is usually includ­
ed because it alters the quantity of energy, and it also af­
fects total factor productivity and capital obsolescence.
Lynde and Richmond (1993) estimate the cost “ dual” of
the production function, a method preferred (but not used)
by Munnell (1992). This method imposes long-run equilibri­
um efficiency conditions, however, which should not be ex­
pected to hold in the short run or in quarterly data. Berndt
and Hansson (1991) and Nadiri and Mamuneas (1991) have
also used the cost function approach to test the public
capital hypothesis in Sweden and the United States,
respectively. Both studies find that the public capital stock
has relatively small effects on a measure of private cost.

unit of a weighted-average bundle of both pri­
vate capital and labor resources. The results in­
dicate that neither the growth rate of the public
capital stock nor the level of public sector in­
vestment cause total factor productivity growth.
On the contrary, the growth of private sector
productivity causes both measures of public
capital formation.
One of the advantages of this approach is that
it explicitly looks for statistically significant rela­
tionships between public capital formation and
subsequent private sector productivity growth,
and the reverse, between productivity growth
and subsequent changes in public capital forma­
tion. The use of longer periods for observing
expected effects allows for lags in the effect of
one measure on the other. Nonetheless, this ap­
proach finds only the reverse relationship to be
statistically significant.

The role of public capital formation and of
the federal government in its provision have
been the subject of widespread discussion and
concern in recent years. This concern has been
prompted by the infrastructure deficit hypothe­
sis, which argues that there has been a sharp
decline in public capital formation and that this
decline lowered U.S. private sector productivity
This article questions the infrastructure
hypothesis. Trends in U.S. public capital forma­
tion indicate that the federal government’s role
32Duffy-Deno and Eberts (1991) examine causality in a simul­
taneous equations framework for 28 metropolitan areas
during the first half of the 1980s using personal income in­
stead of business output. Their results find causality from
public capital to personal income.



in public capital formation has been quite limit­
ed; only a small fraction of the nation’s public,
nonmilitary capital stock is held by the federal
government and the per capita federal capital
stock has been roughly constant throughout the
post-World War II period.
There was a slowing in the growth of state
and local government highways, roads and
educational buildings relative to population
growth in the 1970s and early 1980s. The
demographic and energy-price changes that
gave rise to reductions in the growth of demand
for these goods, however, began to reverse in
the early 1980s. Thus, if there was a deficit in­
dicated by the trend in public capital formation,
it seems to have begun to disappear almost a
decade ago.
The purported link between public capital and
private sector productivity has been widely criti­
cized for distorting the role of public capital,
yielding implausible estimates of the private sec­
tor productivity gains that could arise from pub­
lic capital formation, and reversing the connec­
tion between the two. The fundamental problem
with earlier estimates is that they result from
spurious or unrelated movements in the quanti­
ty of public capital and business sector output
and productivity. While both private sector
productivity and the public capital stock per
hour have risen over time, their movements
have not been closely related. Indeed, in the
1980s the two measures generally moved in­
versely with one another. Of special note is the
rebound in private sector productivity growth
until 1988, which was accompanied by an ac­
celerated decline in the stock of public capital
per hour. The bottom line here is that no one
has produced evidence that an increase in the
nation’s public capital stock will boost private
sector output or productivity, within the year
or even some future period. Quite simply, when
the hypothesis has been explicitly tested this
way, the evidence strongly rejects it.

Aaron, Henry J. “ Comments on 'Historical Perspectives on In­
frastructure Investment: How Did We Get Where We Are?’
by George E. Peterson” presented at the American Enter­
prise Institute for Public Policy Research Conference on
“ Infrastructure Needs and Policy Options for the 1990s,”
Washington, D.C. (February 4, 1991).
_______ . “ Discussion,” in Alicia H. Munnell, ed., Is There a
Shortfall in Public Capital Investment? Federal Reserve
Bank of Boston, Conference Series No. 34 (1990), pp.


Aschauer, David Alan. “ Public Capital and Economic
Growth,” in The Jerome Levy Economics Institute of Bard
College, Public Infrastructure Investment: A Bridge to Pro­
ductivity Growth?, Public Policy Brief No. 4, 1993, pp. 9-30.
_______ . “ Why Is Infrastructure Important?” in Alicia H.
Munnell, ed., Is There a Shortfall in Public Capital Invest­
ment? Federal Reserve Bank of Boston, Conference Series
No. 34 (1990), pp. 21-50.
_______ . “ Public Investment and Productivity Growth in the
Group of Seven,” Federal Reserve Bank of Chicago, Eco­
nomic Perspectives (September/October 1989a), pp. 17-25.
_______ . “ Does Public Capital Crowd Out Private Capital?”
Journal of Monetary Economics (September 1989b), pp.
_______ . “ Is Public Expenditure Productive?” Journal of
Monetary Economics (March 1989c), pp. 177-200.
Bartlett, Bruce. “ If It Ain’t Broke, Don’t Fix It,” Wall Street
Journal, December 2, 1992.
Belsie, Laurent. “ France Relaunches Its Mid-80s Privatization
Push,” The Christian Science Monitor, August 3, 1993.
Berndt, Ernst R., and Bengt Hansson. “ Measuring the Contri­
bution of Public Infrastructure Capital in Sweden,” National
Bureau of Economic Research, Working Paper No. 3842
(September 1991).
Butler, Alison. “ Is the United States Losing Its Dominance in
High-Technology Industries?” this Review (November/De­
cember 1992), pp. 19-34.
Cato Institute. “ Is Technology Policy on the Right Track?”
Cato Policy Report (May/June 1993), pp. 6-9.
Clinton, Bill, and Al Gore. Putting People First: How We Can
All Change America. New York: Times Books, 1992.
Clinton, William J. A Vision of Change for America (Washing­
ton, D.C.: U.S. Government Printing Office, 1993).
Council of Economic Advisers, Economic Report of the Presi­
dent, 1993 (Washington, D.C.: U.S. Government Printing
Office, 1993).
Duffy-Deno, Kevin T., and Randall W. Eberts. “ Public
Infrastructure and Regional Economic Development: A
Simultaneous Equations Approach,” Journal of Urban
Economics (1991), pp. 329-43.
Eberts, Randall W. “ Cross-Sectional Analysis of Public Infras­
tructure and Regional Productivity Growth,” Federal
Reserve Bank of Cleveland, Working Paper No. 9004, May
_______ . “ Public Infrastructure and Regional Economic De­
velopment,” Federal Reserve Bank of Cleveland, Economic
Review (Quarter 1, 1990a), pp. 15-27.
_______ . “ Estimating the Contribution of Urban Public Infras­
tructure to Regional Growth,” Federal Reserve Bank of
Cleveland, Working Paper No. 8610, December 1986.
Eisner, Robert. “ Infrastructure and Regional Economic Perfor­
mance: Comment,” Federal Reserve Bank of Boston, New
England Economic Review (September/October 1991), pp.
Erenburg, Sharon J. “ The Relationship Between Public and
Private Investment,” The Jerome Levy Economics Institute
Working Paper No. 85, February 1993.
European Community. “ Conclusions of the Presidency,” Bulle­
tin of The European Community (Issue 6, 1993), pp. 8-23.
Finn, Mary. “ Is All Government Capital Productive?” Federal
Reserve Bank of Richmond, Economic Quarterly (fall 1993),
pp. 53-80.
Fogel, Robert W. Railroads and American Economic Growth:
Essays in Econometric History. Baltimore: The Johns Hop­
kins University Press, 1964.


Ford, Robert, and Pierre Poret. “ Infrastructure and PrivateSector Productivity,” OECD Economic Studies, No. 17 (au­
tumn 1991), pp. 63-89.
Garcia-Mila, Teresa, and Therese McGuire. “ The Contribution
of Publicly Provided Inputs to States’ Economies,” Regional
Science and Urban Economics (June 1992), pp. 229-41.
Gilder, George. “America’s Best Infrastructure Program,” Wall
Street Journal, March 2, 1993.
Gramlich, Edward M. "State and Local Budgets The Day
After It Rained: Why is the Surplus So High?” Brookings
Papers on Economic Activity (1:1978), pp. 191-216.
Griliches, Zvi. “ Productivity Puzzles and R&D: Another
Nonexplanation,” Journal of Economic Perspectives (fall
1988), pp. 9-21.
Holtz-Eakin, Douglas. “ New Federal Spending for Infrastruc­
ture: Should We Let This Genie Out of the Bottle?” in The
Jerome Levy Economics Institute at Bard College, Public
Infrastructure Investment: A Bridge to Productivity Growth?,
Public Policy Brief No. 4, 1993, pp. 31-46.
_______ . “ Public Sector Capital and the Productivity Puzzle,”
National Bureau of Economic Research, Working Paper
No. 4122, 1992.
_______ . “ Private Output, Government Capital, and the
Infrastructure ’Crisis’,” Columbia University, Discussion
Paper No. 394, May 1988.
Hulten, Charles R., and Robert M. Schwab. “ Is There Too Lit­
tle Public Capital? Infrastructure and Economic Growth,”
paper presented at the American Enterprise Institute Con­
ference on Infrastructure Needs and Policy Options for the
1990s, Washington, D.C. (February 4, 1991).
International Monetary Fund. World Economic Outlook, May
Jondrow, James, and Robert A. Levy. “ The Displacement of
Local Spending for Pollution Control By Federal Construc­
tion Grants,” American Economic Review (May 1984), pp.
Jorgenson, Dale W. “ Fragile Statistical Foundations: The
Macroeconomics of Public Infrastructure Investment,” com­
ment on Hulten and Schwab (1991), presented at the
American Enterprise Institute Conference on Infrastructure
Needs and Policy Options for the 1990s, Washington, D.C.
(February 4, 1991).
Lynde, Catherine, and J. Richmond. “ Public Capital and Total
Factor Productivity,” International Economic Review (May
1993), pp. 401-14.

McMillan, W. Douglas, and David J. Smyth. “ Multivariate
Time Series Analysis of the Aggregate Production Func­
tion,” Louisiana State University Working Paper, October
Moore, Stephen. “ Crisis? What Crisis? George Bush’s Never
Ending Domestic Budget Build-Up,” Cato Institute Policy
Analysis No. 173, June 19, 1992.
Munnell, Alicia H. “ Policy Watch: Infrastructure Investment
and Economic Growth,” Journal of Economic Perspectives
(fall 1992), pp. 189-98.
_______ . “ How Does Public Infrastructure Affect Regional
Economic Performance?” in Is There A Shortfall in Public
Capital Investment? Federal Reserve Bank of Boston, Con­
ference Series No. 34 (1990a), pp. 69-103.
_______ . “ Why Has Productivity Growth Declined? Produc­
tivity and Public Investment,” Federal Reserve Bank of
Boston, New England Economic Review (January/February
1990b), pp. 3-22.

Musgrave, Richard A. “ Discussion,” in Alicia H. Munnell, ed.,
Is There a Shortfall in Public Capital Investment? Federal
Reserve Bank of Boston, Conference Series No. 34 (1990),
pp. 64-8.
_______ and Peggy B. Musgrave. Public Finance in Theory
and Practice, 5th edition (New York: McGrawHill Book
Company, 1989).
Nadiri, M. Ishaq, and Theofanis P Mamuneas. “ The Effects
of Public Infrastructure and R&D Capital on the Cost Struc­
ture and Performance of U.S. Manufacturing Industries,”
National Bureau of Economic Research, Working Paper
No. 3887, October 1991.
Organization for Economic Co-operation and Development.
National Accounts, Detailed Tables, Volume II.
Peterson, George E. “ Is Public Infrastructure Under­
supplied?” in Alicia Munnell, ed., Is There a Shortfall in
Public Capital Investment? Federal Reserve Bank of
Boston, Conference Series No. 34 (1990), pp. 113-30.
Ratner, Jonathan B. “ Government Capital and The Production
Function for U.S. Private Output,” Economic Letters (1983),
pp. 213-17.
Reich, Robert B. “ The REAL Economy,” The Atlantic Monthly
(February 1991), pp. 35-52.
Reuters Limited. “ Council Agrees Two-Speed Phone Liberali­
zation Plan,” The Reuter European Community Report (June
16, 1993).
Riding, Alan. “ France is Selling 21 Big Companies,” New
York Times, May 27, 1993.
Rodgers, T. J. “ What Silicon Valley Needs From Clinton,” The
Wall Street Journal, April 12, 1993.
Rubin, Laura. “ Productivity and the Public Capital Stock:
Another Look,” Board of Governors of the Federal Reserve
System, Division of Research and Statistics, Working Paper
Series No. 118, May 1991.
Seely, Bruce. “A Republic Bound Together,” Wilson Quarterly
(winter 1993), pp. 19-39.
Tatom, John A. “ Paved With Good Intentions: The Mythical
National Infrastructure Crisis,” Cato Institute Policy Analysis
No. 196 (August 12, 1993).
_______ . “ Public Capital and Private Sector Performance,”
this Review (May/June 1991a), pp. 3-15.
_______ . “ Should Government Spending on Capital Goods
Be Raised?” this Review (March/April 1991b), pp. 3-15.
Tiebout, Charles M. “A Pure Theory of Local Expenditures,”
Journal of Political Economy (1956), pp. 416-24.
U.S. Congressional Budget Office. Trends in Public Infrastruc­
ture Outlays and The President’s Proposals for Infrastructure
Spending in 1993, Washington, D.C.: Government Printing
Office, May 1992.
_______ . How Federal Spending for Infrastructure and Other
Public Investments Affect the Economy, Washington, D.C.:
Government Printing Office, July 1991.
_______ . New Directions for the Nation’s Public Works,
Washington, D.C.: Government Printing Office, September
_______ . Federal Policies for Infrastructure Management,
Washington, D.C.: Government Printing Office, June 1986.
U.S. Office of Management and Budget. Budget Baselines,
Historical Data, and Alternatives for the Future, Washington,
D.C.: Government Printing Office, January 1993.
Wall Street Journal. Editorial, “ Europe Gets the Message,”
May 17, 1993.
Washington Post. “ Inventing Clintonomics: The Advisers
Seeking to Shape Policies Mix Pragmatism, Activism,”
November 8, 1992.




Sangkyun Park
Sangkyun Park is a senior economist at the Federal Reserve
Bank of St. Louis. Jonathan Ahibrecht provided research

The Determinants o f Consumer
Installment Credit

factors in making their borrowing decisions.
Thus, to interpret the movement of consumer
credit accurately, one needs to identify the eco­
nomic factors that influence consumer borrow­
ing and understand the ways those variables
affect consumers' decisions. This article studies
consumers' borrowing behavior by investigating
both long-term trends and short-term fluctua­
tions o f consumer credit in relation to economic
and institutional factors, including the Tax Re­
form Act of 1986, which phased out tax deduc­
tions for interest expense on consumer debt.
The focus is on consumer installment credit,
which includes major categories of consumer
loans such as automobile and credit card loans.
The behavior of consumer credit has attracted
considerable attention during the last 10 years.
Many analysts argue that consumers accumulat­
ed excessive debt in the 1980s and became
reluctant to use credit in the early 1990s. In
fact, after growing rapidly in the mid-1980s,
consumer installment credit declined in many
quarters during 1991 and 1992. The decline of
consumer installment credit in the early 1990s
is particularly interesting because it occurred
despite low interest rates. The decline during
'Data on consumer installment credit are collected and pub­
lished by the Federal Reserve Board of Governors in their
G19 release.
2There is also noninstallment consumer credit, which con­
sists mostly of short-term credit such as charge card
balances that need to be paid in full within the billing
cycle. This credit category totaled $52 billion at the end of

the early 1990s after a period of rapid growth
may not be fully explained by changed con­
sumption expenditures. Thus, it appears that
consumers have changed the pattern of financ­
ing their purchases.
The change in consumer installment credit is
the difference between the extension o f new
credit and the repayment o f the principal of ex­
isting debt. This article examines the variables
that may affect the proportion of consumption
that is financed by debt and the rate at which
consumers repay existing debt principal.

Consumer installment credit covers most
short- and intermediate-term credit extended to
individuals for which repayment is scheduled in
two or more installments, excluding loans se­
cured by real estate.1 Consumer installment
credit, which totaled about $760 billion at the
end of 1992, consists o f three main categories:
automobile credit, revolving credit and other
credit.2 Revolving credit is mainly credit card
1992. Because it reflects more of delayed settlements than
credit extension and accounts for a relatively small portion
of consumer credit, noninstallment credit is not discussed



loans, and the category other credit includes
loans to finance purchases of mobile homes,
home appliances and furniture, and personal
loans. Major lenders are commercial banks,
finance companies, credit unions, savings insti­
tutions, retailers and gasoline companies.
The economic importance of consumer install­
ment credit may be illustrated by examining it
in relation to other components of the house­
hold balance sheet. Table 1 shows the balance
sheet of the household sector for selected years
between 1960 and 1992. Tw o main components
of household liabilities are home mortgages and
consumer installment credit. Consumer install­
ment credit emerged as a main financing tool
for households after W orld W ar II and has
been a major component o f the household
balance sheet since the early 1950s. Between
1960 and 1992, consumer installment credit
held fairly stable at about 20 percent of total
Although its long-term trend can be described
as a steady increase in line with other compo­
nents of the household balance sheet, consumer
installment credit grew at uneven rates over
short time spans. Particularly notable are rapid
growth in the mid-1980s and stagnation in the
early 1990s. In most quarters of the years be­
tween 1984 and 1986, the annualized growth
rate o f consumer installment credit was sub­
stantially more than 10 percent. Between 1991
and 1992, however, the outstanding amount
of consumer installment credit declined in
many quarters. In particular, automobile credit
decreased in all but one quarter of the two

directly comparable because the former is a
stock, a value at a point in time, whereas the
latter is a flow, a rate per unit of time. For pur­
poses of comparability, the change in consumer
installment credit, which is a flow, is compared
with consumption expenditures. Figure 1 shows
the ratio of the change in each category of con­
sumer installment credit to the consumption
expenditures on the relevant category of goods,
which is referred to as the credit ratio.3 For
total credit, consumption expenditures on dura­
ble goods are used as the denominator because
consumers obtain credit mostly to finance the
purchase o f big-ticket items such as automo­
biles, furniture and home appliances. The
denominator for the automobile credit ratio is
consumption expenditures on automobiles. For
the revolving credit ratio, consumption expendi­
tures on all items but automobiles are used as
the denominator because consumers use credit
cards for a wide variety of purposes but gener­
ally not for the purchase o f automobiles. For
the other credit ratio, expenditures on durable
goods other than automobiles serve as the
A high or low credit ratio may be interpreted
as fast or slow credit growth relative to con­
sumption. Thus, fluctuations of the credit ratio
reflect changes in the financing pattern of con­
sumers, that is, changes in the proportion of
debt-financed consumption, in the rate of repay­
ment of existing debt, or in both. In other
words, substantial changes in the credit ratio
suggest that factors other than consumption
have affected consumer borrowing.4

To investigate the possibility that the financ­
ing pattern of consumers has changed over
time, we need to examine the behavior of con­
sumer installment credit in relation to consump­
tion. The outstanding amounts of consumer
credit and consumption expenditures are not

In Figure 1, the total credit ratio shows no
apparent long-term trend but exhibits wide
short-term fluctuations. Excluding the early
1990s, the automobile credit ratio fluctuates
around 0.13 and has no apparent long-term
trend. For revolving credit, the credit ratio
shows an upward trend, reflecting the increased
use of credit cards during the last two decades.

3Data are quarterly and seasonally adjusted. The Federal
Reserve Board’s data on consumer installment credit show
several breaks that may arise from modified classifications.
For example, securitized consumer loans were added to
the data in January 1989. To alleviate such problems, the
quarterly changes are linearly interpolated when obvious
breaks are found. The interpolated data points are 1971:1,
1977:1 and 1989:1.

for revolving credit and -0.482 for other credit. Time
trends in credit ratios, which were upward for revolving
credit and downward for other credit, can explain the sig­
nificant magnitudes of the coefficients for revolving and
other credit. On the other hand, the correlation coefficients
between credit ratios and changes in consumer installment
credit were all greater than 0.8 for the four categories of

4The correlation coefficients between credit ratios and rele­
vant consumption expenditures confirm that credit ratios
were unrelated to the cyclical behavior of consumption be­
tween 1970 and 1992. The correlation coefficients were
-0.133 for total credit, -0.098 for automobile credit, 0.533



Table 1
Balance Sheet of the Household Sector
(in billions of 1982-84 dollars)1
Tangible assets
Owner-occupied real estate
Nonprofit tangible assets
Consumer durables
Financial assets
Home mortgages
Installment consumer credit
Other liabilities












'Includes personal trusts and nonprofit organizations. Dollars are deflated by consumer price index. Numbers in parentheses
are percent of total assets for asset items and percent of total liabilities for liability items.
SOURCE: Board of Governors.

In contrast, the other credit ratio appears to
have declined over time.
Short-term fluctuations in the credit ratios are
much more notable. Between 1970 and 1992,
the total credit ratio ranged from -0.07 to 0.26.
The movement of the total credit ratio generally
confirms that consumers borrowed aggressively
in the mid-1980s but became reluctant to bor­
row in early years of the 1990s. The changed
borrowing behavior is particularly evident for
automobile credit; the automobile credit ratio
plunged in 1991 after peaking in the mid1980s.5 A reason for the wider fluctuation of
automobile credit may be that consumers con­
sider the economic environment more seriously
when they obtain larger loans.
5Eugeni (1993) suggests that an increase in auto leases
partly explains the slow growth of consumer credit in recent periods. The credit ratio, however, is not seriously af­
fected by auto leasing. The Bureau of Economic Analysis,
which publishes the data on consumption expenditures,
classifies rental and leasing expenses under expenditures
on services as opposed to goods. The credit extension
involving leasing is classified under business credit.

The main economic decisions of consumers
are to allocate available resources to various
types o f consumption and to construct a desira­
ble personal financial structure. The resources
available to consumers include existing wealth,
current income and future income. Consumers
allocate these resources between current con­
sumption and future consumption. By distribut­
ing resources prudently over time, consumers
can avoid excessive consumption and prevent
future financial hardships. Changes in the eco­
nomic environment also require a restructuring
in consumer balance sheets. A well-managed
Accordingly, an increase in leasing reduces consumption
expenditures on durable goods as well as consumer credit,



Figure 1a
Total Credit Ratio1

0.250 . 20 0.150 . 10 -


-0.05 H

- ° - 1 0 _ !— i— i— i— i— i— i— i— i— i— i— i— i— i— i— i— i— i— i— i— i— i— r









1Change in total consumer installment credit/consumption
expenditures on durable goods

Figure 1b
Automobile Credit Ratio1

'Change in automobile credit/consumption
expenditures on automobiles






Figure 1c
Revolving Credit Ratio1

1Change in revolving credit/consumption
expenditures other than automobiles

Figure 1d
Other Credit Ratio1

Change in other credit/consumption expenditures
on durable goods other than automobiles



household balance sheet can increase the net
worth and liquidity of the household.
The change in outstanding consumer credit in
a given period is the difference between acquisi­
tion of new credit and repayment of existing
credit. Acquisition of new credit, which is a
decision to use future income for current pur­
chases, reflects various factors such as the level
and type of consumption, characteristics of con­
sumers, the relative cost of resources, as well as
macroeconomic conditions. The relative cost of
resources along with macroeconomic conditions
may also affect repayment of existing credit,
which is an act of transferring current income
to wealth. The variables affecting the acquisition
and repayment of loans should explain changes
in consumer credit.
Acquisition of new credit would tend to in­
crease with consumption, especially with expen­
ditures on durable goods. As shown in Figure 1,
however, consumption alone is not enough to
explain the growth of consumer credit. This
section focuses on other factors that may in­
fluence the financing pattern of consumers and
thereby affect credit growth. An examination of
those factors helps clarify the relationship be­
tween credit growth and the economic environ­
ment and sheds light on the unusually fast
growth of consumer credit in the mid-1980s
and the particularly slow credit growth in the
early 1990s.
Other factors considered here are the growth
of home equity lines o f credit, demographic
characteristics, income distribution, interest
expense on consumer installment credit, con­
sumer confidence, the debt burden of house­
holds, and measures of banks' willingness to
lend. Tax deductibility may play an important
role as well. It is considered as a part of the
discussion of home equity lines of credit and
interest expense.
Consumers can select among various credit
instruments to satisfy a given borrowing need.
Because home equity lines of credit can serve
6Home equity lines of credit are generally classified under
home mortgages.
7See Canner and Luckett (1989).
in te re st expense on consumer loans was 100 percent tax
deductible before 1987. The deductibility decreased to 65
percent in 1987, 40 percent in 1988, 20 percent in 1989, 10
percent in 1990 and 0 percent thereafter.


as close substitutes for consumer installment
credit, any discussion of consumer credit must
take into account the growth o f home equity
lines of credit. Demographic characteristics and
income distribution may influence the long-term
trend of consumer borrowing by affecting the
income profile o f typical consumers. Interest
rates will influence the use of consumer install­
ment credit. Consumer confidence may indicate
the consumers' anticipation o f future income,
which is an important consideration in making
borrowing decisions. The size o f debt burden
may also affect the borrowing decision. The
supply of credit shall also be considered. If
lenders are reluctant to lend, consumers cannot
borrow as much as they want.

H om e Equity Lines o f Credit
Home equity lines o f credit deserve particular
attention because they can serve as close substi­
tutes for consumer installment credit and are
offered at a lower rate, especially on an after­
tax basis. Home equity lines of credit, which
became widely available in the mid-1980s,
emerged as an important financing tool for
households. Once they open a line of credit,
households can conveniently obtain extra credit
and flexibly repay the outstanding amount. The
flexibility of home equity lines of credit allows
households to easily substitute this credit for
conventional consumer loans.6 Home equity
lines of credit, which are secured, are offered
at comparatively low interest rates, generally at
1.5 percentage points above the prime rate.7
Furthermore, home equity lines of credit are
treated as home mortgages for tax purposes
and, hence, the interest expense is fully tax
deductible in most cases. Consequently, home
equity lines of credit have been more attractive
than conventional consumer credit since the
Tax Reform Act of 1986 phased out tax deduc­
tions for interest expense on conventional con­
sumer credit.8
Home equity lines of credit at commercial
banks and S&Ls almost tripled between 1987


and 1991, from $32 billion to $86 billion.9 The
surge of home equity lines of credit, however,
may have occurred at the expense of conven­
tional consumer loans. The 1988 Survey o f Con­
sumer Attitudes shows that the major reasons
for drawing on home equity lines of credit
include the repayment of other debts and the
purchase of automobiles.1 In addition, the surge
of home equity lines of credit coincided with
the phase-out of tax deductions for interest ex­
pense on consumer installment credit.
After the initial surge, home equity lines of
credit stagnated in 1992. The stagnation might
be explained by active refinancing o f home
mortgages in recent years, another way of
extracting home equity. Canner and Luckett
report that nearly 60 percent o f those who
refinanced their residential mortgages increased
their mortgage debt.1 It is obvious that some
consumers have substituted home equity for
consumer installment credit by using home
equity lines of credit in the late 1980s and early
1990s, and also mortgage refinancing in the
early 1990s. This finding suggests a significant
effect of the phase-out of tax deductions for
interest expense on conventional consumer
debt. The phase-out o f tax deductions certainly
appears to have contributed to the slowing in
the growth of consumer credit in the late 1980s
and the early 1990s by accelerating the substitu­
tion of consumer installment credit with loans
secured by residential properties.

income. Because younger individuals in general
have accumulated little wealth and have low
current incomes relative to their future in­
comes, they are more likely to finance current
consumption with future income.
Table 2 shows the percentage distribution of
U.S. population by age between 1960 and 1990.
The population may roughly be classified into
the following three groups: (1) those who do
not make independent financial decisions—
younger than 20 years of age; (2) those who
make independent financial decisions and rely
heavily on future income—between 20 and 34
years o f age; and (3) those who make indepen­
dent financial decisions and primarily rely on
current income and existing wealth—35 years of
age and older.1 During the 30-year period, both
the percentages o f the 20-34 group and the 35
and older group generally increased, but the in­
crease was larger for the 20-34 group. Thus,
the net long-run effect is likely to be increased
consumer borrowing. The increase in the 20-34
group was particularly marked from 1970 to
1980, which seems to be consistent with heavy
borrowing in the mid-1980s. Furthermore, the
stabilizing of the 20-34 group, along with con­
tinued growth in the 35 and older group, sug­
gest that age distribution may have contributed
to a slowing in consumer credit during the late
1980s and early 1990s.

Incom e Distribution
Dem ographic Characterisitics
Borrowing decisions differ across consumers.
Therefore, the aggregate outcome depends on
the demographic composition of consumers. The
age of consumers may be important. According
to the permanent income hypothesis, consumers
maximize lifetime utility by using credit to cre­
ate a pattern of consumption over their life­
times that is smoother than the pattern of

9ln 1988 about 85 percent of home equity lines of credit
was held by commercial banks and savings institutions
(Canner and Luckett, 1989). The Call Report data (Consoli­
dated Reports of Condition and Income) of the Federal
Reserve Board have included home equity lines of credit
since 1987:4 for commercial banks and since 1988:4 for
S&Ls. For the period between 1987:4 and 1988:3, home eq­
uity lines of credit at S&Ls were estimated. The credit at
S&Ls was assumed to have grown at the same rate as that
at commercial banks.

The distribution of income also influences the
aggregate borrowing behavior of consumers.
Middle-income individuals, who do not have
large current income but may expect stable
future income, may on average actively borrow
to finance current consumption. On the other
hand, high-income individuals generally have
less need to borrow, and low-income individuals
without stable employment may be afraid to
borrow, unable to borrow, or both. Hendricks

"S e e Canner and Luckett (1990). They study mortgage
refinancing based on consumer surveys conducted in 1988
and 1989.
12Hendricks and others (1973) report that the ratio of install­
ment debt to income decreases sharply between the 30-34
age group and the 35-40 age group. The study looks at the
relationship between consumer characteristics and
borrowing behavior using survey data.

10See Canner and Luckett (1989).



Table 2
Percentage of U.S. Population by Age






Under 20
65 and over






Under 20
35 and over






SOURCE: U.S. Bureau of Census.

and others report the highest ratios of install­
ment debt to income for middle-income fami­
lies.1 Kennickell and Shack-Marquez also show
that the proportion of families carrying credit
cards and other consumer debt is the largest
for middle-income families.1 Consequently, a
shift in the distribution o f income toward
middle-income families might be associated with
more consumer borrowing for a given amount
of consumption.
Table 3 shows that the proportion of middleincome households (annual income between
$25,000 and $49,999 in 1989 dollars) gradually
decreased from 38.8 percent to 33.2 percent be­
tween 1970 and 1989. Median income increased
slightly during the period, but the slight in­
crease does not appear to be enough to explain
the changed income distribution. With other
things constant, the decreased proportion of
middle-income households should have reduced
consumer borrowing. Figure 1 does not show
either an upward or downward long-term trend
in consumers’ borrowing behavior. It is possible
that the long-term effects of age distribution
and income distribution have roughly offset
each other.
13See Hendricks and others (1973).
14See Kennickell and Shack-Marquez (1992).


Interest Rates
The price of credit is expected to have an ef­
fect on consumers’ borrowing decisions. Higher
interest rates on consumer credit mean larger
sacrifice o f future income for a given level of
current consumption financed by future in­
come. Thus, higher interest rates on consumer
credit will discourage current consumption in
general and have an even larger effect on the
use of consumer credit for current purchases.
Consumers with heavy borrowing needs are
more likely to defer purchases. Therefore, the
proportion of debt-financed consumption should
be lower. A high cost of carrying debt will also
induce households to repay existing debt faster.
Hence, in addition to slowing consumption,
increases in interest rates reduce the proportion
of consumption financed with debt and increase
the repayment rate, causing growth of con­
sumer credit to slow relative to consumption.
To illustrate the effect of interest rates on
consumer installment credit, this study com­
pares interest rates on 48-month new car loans
and the automobile credit ratio, instead o f the
total credit ratio, because it is difficult to obtain
an interest rate applicable to consumer install­


Table 3
Percent Distribution of Household Income
(in thousands of 1989 dollars)





Less than 10.0
75.0 or more






Less than 25
50 or more






Annual Income

Median Income






SOURCE: U.S. Bureau of Census.

ment credit in general.1 Figure 2 shows the
real after tax interest rate on automobile credit
along with the automobile credit ratio.1 The
real after-tax rate is more relevant than the
nominal interest rate. Because future income
tends to rise with the rate of inflation, the in­
terest rate net of inflation more accurately
reflects the price of present consumption in
terms of future income. Despite lower nominal
interest rates, the real after-tax interest rate on
automobile credit stayed high in the early 1990s
because of low rates of inflation and the phase­
out of tax deductions for interest expense on
consumer loans.
In Figure 2, the automobile credit ratio gener­
ally exhibits a negative relationship with the
interest rate except for years between 1983 and
1986, when both the credit ratio and the in­
15Data on the interest rate on automobile credit are available
from the first quarter of 1972. No appropriate measure of
the interest rate is available for other credit, which consists
of various types of loans. The interest rate on credit card
loans is available, but the demand for credit card loans is
known to have been unusually insensitive to the interest
rate. Thus, revolving credit is not a good candidate for this
16ftea/ after-tax interest rate = nominal interest rate - expect­
ed rate of inflation - tax deduction. The four-quarter aver­
age (the current and next three quarters) of the annualized
rate of change in the consumer price index is used as the
measure of expected inflation. The assumption here is that
consumers on average forecast the rate of inflation ac­
curately. Tax deduction is the nominal interest rate mul­

terest rate w ere high. One possibility is that
consumers, who had experienced high inflation
in the early 1980s, mistakenly expected high
inflation and, hence, underestimated the real
interest rate in those years. The figure shows
very high real after-tax interest rates in the ear­
ly 1990s, when the credit ratio was very low.
Thus, when the effects o f tax deductibility and
inflation are incorporated, the movement of
interest rates is consistent with the slow credit
growth o f the early 1990s.

Relative Interest Rates
When consumers have more than one financ­
ing alternative, they will choose the least costly
method. A financing alternative available to
tiplied by the proportion of tax-deductible interest expense
multiplied by the marginal federal income tax rate for fourperson, median-income families. The data source of the
marginal federal income tax rate is the Department of the
Treasury (1991); the tax rate for 1992 is assumed to be the
same as in 1991. State income taxes, which vary, are not
considered. The incorporation of state income taxes would
make the tax deduction more significant and, hence, raise
the real after-tax interest rate of recent periods.



Figure 2
Real After-Tax Interest Rate on Automobile Credit
Credit Ratio

Interest Rate



- -



Interest rate
Credit ratio














households is to draw down their wealth. Be­
cause bank deposits offer financial flexibility,
many households with large bank accounts may
still want to finance their automobile purchases
with loans even though the interest rate on au­
tomobile loans is higher than the return on
deposits. When there is a large gap between the
interest rate on household liabilities and the
return on household financial assets, however,
households may use their assets to finance con­
sumption instead o f incurring more debt. Fur­
thermore, the high cost of carrying liabilities
relative to the return on assets prompts the
repayment of existing debt. Thus, the relative
interest rates on assets and liabilities can have
an effect on the growth of consumer install­
ment credit.
Assuming that the returns on major house­
hold assets, such as certificates of deposits and
money market shares, are closely tied to the
17Spread = [interest rate on automobile loans - (interest rate
on automobile loans x marginal tax rate x tax deductibili­
ty)] - [three-month Treasury bill rate - (three-month Treasu­
ry bill rate x marginal tax rate)]. In this calculation, all
interest income of households is assumed to have been
subjected to income tax throughout the period.



















Treasury bill rate, we can estimate the spread
between the interest rate on consumer credit
and the return on household financial assets
using the Treasury bill rate as a proxy for the
return on household assets. Figure 3 shows the
relationship between the spread of after-tax
interest rates on automobile loans over threemonth Treasury bills on an after-tax basis and
the automobile credit ratio.1 The phase-out of
tax deductions widens this spread as much as it
raises after-tax interest rates on automobile
loans because the Tax Reform Act of 1986 had
little effect on the after-tax return on household
assets. The spread and credit ratio tended to
move in opposite directions for the most part.
The spread can partly explain the rapid credit
growth in the mid-1980s, unlike the real after­
tax interest rate, and well explains the slow­
down of credit growth in the early 1990s. This
analysis suggests that the interest rate spread


Figure 3
Spread of Interest Rates on Automobile Credit
Over Three-Month Treasuries

significantly affects consumer borrowing be­
havior and also confirms the importance of the
Tax Reform Act of 1986.

Confidence in the E conom y
The purchase of goods on credit is an act of
financing current consumption with anticipated
future income. Because future income is uncer­
tain, consumers’ borrowing decisions depend on
their confidence in the future. In particular,
confidence about future income may significant­
ly affect the financing decisions of the con­
sumers who rely heavily on future income.
When confidence is low, those consumers are
discouraged from purchasing goods in the cur­
rent period.1 If those heavy credit users defer
consumption, consumer borrowing in aggregate
will be smaller per unit of consumption. In
addition, consumers in general may wish to
18According to the panel study of Hendricks and others
(1973), families that are more optimistic about financial
progress borrow more. In addition, the index of past and
future financial progress is more highly correlated with bor­
rowing than it is with consumption expenditures. These
results suggest that households borrow more per unit of
consumption when they are optimistic about the future.

Credit Ratio

consume less and reduce the level of debt to
prepare for an uncertain future. Therefore, the
repayment rate o f existing debt tends to be
Measures of consumers' confidence are de­
signed to capture consumers’ subjective feelings
about economic conditions that might influence
their spending decisions. Those feelings can
have an impact on consumer borrowing, regard­
less of their accuracy.1 What is more relevant
for consumers making their decisions might be
the perception about future income rather than
the actual future income. Figure 4 compares the
Conference Board’s index of consumer confi­
dence and the total credit ratio. The two varia­
bles show a strong tendency to move together.
The only exception is the period between 1987
and 1989, when the credit ratio declined despite
19Weinberg (1993) discusses the validity of consumer confi­
dence indices and concludes that their usefulness as a
forecasting tool is limited.



Figure 4
Consumer Confidence Index
Credit Ratio
- 0.25

i i i i i i i i i i i i i i i I i i i i i r







very high levels of consumer confidence. This
measure of consumer confidence is consistent
with high credit ratios between 1984 and 1986
and low ratios between 1990 and 1992. Overall,
it appears that consumers’ perception about the
economy significantly influences their borrow­
ing behavior.

D ebt Burden o f Households
Many analysts have cited the large debt bur­
den of households as a factor contributing to
the slowdown o f consumer installment credit in
the early 1990s. A heavy debt burden means
that consumers have already used a large por­
tion of future income and, hence, have less
future income available for consumption. Then
they are likely to consume less in the current
period and repay debt faster in an effort to
smooth out consumption.
Figure 5 shows that the stock of consumer
installment credit as a percentage of disposable
personal income increased rapidly through most
20Debt service payments have been estimated by the Board
of Governors based on the methodology suggested by
Paquette (1986).








of the 1980s. That the debt/income ratio peaked
toward the end of the 1980s might appear to be
consistent with the slowdown o f credit growth
in the early 1990s. A careful comparison of the
debt/income ratio with the total credit ratio,
however, does not convincingly support the eco­
nomic relationship between the two variables.
Since 1970, the debt/income ratio generally
lagged behind the credit ratio, indicating that
changes in the debt/income ratio may have been
a result rather than a cause of movements of
the credit ratio. Thus, Figure 5 shows more
of an accounting relationship than economic
causality; the debt burden increased as a result
of heavy borrowing in previous periods.
An alternative measure o f the debt burden is
debt service payments, which are principal and
interest payments on household debt. A possible
advantage of this measure over the debt/income
ratio is the incorporation o f the effects of
interest rate changes. Figure 5 also shows the
estimated debt service payments o f households
as a share of disposable personal income.2 The


Figure 5a
Ratio of Consumer Installment Credit to
Annual Disposable Personal Income
Credit Ratio

Debt Ratio

Figure 5b
Ratio of Debt Service Payments to
Annual Disposable Personal Income
Credit Ratio

Debt Burden

0 .1 8 -

0 .1 7 -


0 .1 5 -



I I I I I I i I I I I I I I i I I I I l~T













Figure 6
Index of Banks' Willingness to Lend

estimate includes mortgage payments, as well as
payments on consumer credit. The debt service
burden shows a similar pattern of movements
to the debt/income ratio, especially during the
1980s and 1990s. Consequently, it is difficult to
draw any conclusions about the effect of debt
burdens on financing patterns because the two
measures are so closely intertwined.

Willingness o f Lenders to Lend
The factors considered thus far have dealt
mainly with the demand for consumer install­
ment credit. Supply conditions may also affect
the quantity of consumer installment credit.
Tightening of lending standards by financial in­
stitutions may force many consumers with
21The index is derived from the Senior Loan Officer Opinion
Survey of Bank Lending Practices. In the survey, the par­
ticipating banks (60 major banks) indicate the change in
their willingness to lend during the last three months by
selecting one of the following five willingness categories:
much more willing, somewhat more willing, basically un­
changed, somewhat less willing and much less willing. In
constructing the index, a number is assigned to each
category (2, 1,0, -1 and -2 ). The index is the weighted
average of the assigned numbers multiplied by 100.


Credit Ratio

heavy credit needs to defer consumption and
many others to find other financing means.
Hence, the reluctance of lenders to extend
credit may result in slow growth of consump­
tion and even slower growth o f consumer
Figure 6 plots an index based on a survey of
banks’ willingness to make consumer installment
loans along with the total credit ratio.2 Positive
index values indicate that banks on average
were more willing to lend during the period,
whereas negative values mean that banks were
less willing to lend. During the period exa­
mined, the credit ratio generally followed the
index with a lag of a few quarters. In terms of
the direction o f changes, the relationship be­
tween the two variables appears to have been


Table 4
Is the Factor Consistent with the Financing Pattern of Consumers?
Home equity lines of credit
Real after-tax interest rate
Interest rate spread
Consumer confidence index
Debt burden
Willingness to lend


Early 1990s




systematic throughout the period examined.
High values for the index were closely in line
with the credit ratio during the mid-1980s. In
the early 1990s, however, the relationship be­
came questionable because of a large gap be­
tween the two variables, though the direction of
changes remained consistent.

This article has examined consumers’ borrow­
ing behavior between 1970 and 1992, with par­
ticular emphasis on consumer installment credit.
Consumption expenditures do not fully explain
consumer borrowing because consumers vary
their financing pattern (the proportion of debtfinanced consumption and the repayment rate
o f existing debt) in response to economic and
institutional changes.
Although showing no apparent long-term
trend, the ratio of the change in consumer in­
stallment credit to consumption expenditures
fluctuated widely since 1970, indicating that the
financing pattern of consumers is volatile in the
short run. In particular, consumer installment
credit declined in the early 1990s after increas­
ing rapidly during the second half o f the 1980s.
Short-term borrowing behavior can be affected
by many economic and institutional factors,
such as the emergence of a new borrowing in­
strument, the cost of consumer credit, the cost
of consumer credit relative to the return on
household financial assets, confidence in the
economy, the debt burden of households and
the supply condition of consumer credit. This
article has examined the relationship between
these factors and the ratio of the change in
consumer installment credit to consumption

Table 4 summarizes the qualitative results.
The movements of most of the economic varia­
bles are found to have been fairly consistent
with the behavior of consumer installment
credit during the period examined. Overall, the
growth of consumer installment credit relative
to consumption expenditures is particularly well
explained by the difference between the cost of
consumer credit and the return on household
financial assets (interest rate spread) and the
consumer confidence index. In explaining the
heavy borrowing of the mid-1980s, the high
willingness of banks to lend appears significant.
The small interest rate spread, high levels of the
consumer confidence index and the high
proportion of young adults also help explain the
heavy borrowing. The slow growth of consumer
installment credit in the early 1990s is fairly
well explained by high real after-tax interest
rates, large interest rate spreads, low levels of
the consumer confidence index and the emer­
gence of home equity lines of credit. The phase­
out of the tax deductions for interest expense
on consumer credit appears to have played a
significant role in slowing the growth of con­
sumer installment credit after 1986. The change
in tax law induced households to substitute
household financial assets and home equity lines
of credit for consumer installment credit.

Board of Governors of the Federal Reserve System. “ Senior
Loan Officer Opinion Survey on Bank Lending Practices,”
various issues.
Canner, Glenn B., James T. Fergus, and Charles A. Luckett.
“ Home Equity Lines of Credit,” Federal Reserve Bulletin
(June 1988), pp. 361-73.


_______ , and Charles A. Luckett. “ Home Equity Lending,”
Federal Reserve Bulletin (May 1989), pp. 333-44.

Survey of Consumer Finances,” Federal Reserve Bulletin
(January 1992), pp. 1-18.

_______ , a n d ________“ Mortgage Refinancing,” Federal
Reserve Bulletin (August 1990), pp. 604-12.

Paquette, Lynn. “ Estimating Household Debt Service Pay­
ments,” Federal Reserve Bank of New York, Quarterly
Review (summer 1986), pp. 12-23.

Eugeni, Francesca. “ Consumer Debt and Home Equity
Borrowing,” Federal Reserve Bank of Chicago, Economic
Perspectives (March/April 1993), pp. 2-14.

U.S. Bureau of the Census. Statistical Abstract of the United
States, various issues.

Hendricks, Gary, Kenwood C. Youmans, and Janet Keller.
Consumer Durables and Installment Credit: A Study of
American Households. Survey Research Center, University
of Michigan, 1973.

U.S. Department of the Treasury. “Average and Marginal In­
come Tax, Social Security, and Medicare Tax Rates for
Four-Person Families at the Same Relative Positions in the
Income Distribution, 1955-1991,” September 1991.

Kennickell, Arthur, and Janice Shack-Marquez. “ Changes in
Family Finances from 1983 to 1989: Evidence from the

Weinberg, Norman. “ The Confidence Gap,” Forecast
(September/October 1993), pp. 47-50.



Joseph A. Ritter
Joseph A. Ritter is an economist at the Federal Reserve Bank
of St. Louis. Leslie Banks and Heidi L. Beyer provided
research assistance.

Measuring Labor Market
Dynamics: Gross Flows o f
Workers and Jobs

truction of specific jobs or the movement of
workers into and out of employment—are the
immediate outcomes of labor market processes.
When a firm closes a plant, it destroys jobs.
When it opens a plant, it creates jobs. When an
adult leaves a job to return to school full time
or take care of a child, there is a flow from the
pool of those employed to the pool of those
not in the labor force. If the job itself is not
destroyed, another person may move from un­
employment to employment to fill it. If a
construction worker's job ends with the first
snowfall, that is a job destroyed. Simultaneous­
ly, the worker may move from employed to un­
employed or leave the labor force. On the other
hand, he or she may move immediately into
another job, perhaps one created in anticipation
o f the Christmas boom in retail sales, one that
will be destroyed in January. Taking a wider
view, an observer of the U.S. economy might
notice that since the trough of the most recent
recession, prominent employers have laid off
thousands o f workers—jobs destroyed—but
more diffuse (and thus less visible) job creation
has nevertheless raised overall employment by
more than 3 million.
Standard measures o f labor market develop­
ments condense all of these events into a single
number, the net change in employment. Useful
as they are, these statistics hide an interesting

and potentially informative (though difficult
to measure) dimension of labor market de­
velopments: the gross numbers of jobs created
and destroyed and the gross movements of in­
dividuals into and out of employment. An em­
ployment increase of 10,000 by one of the usual
measures may mean 10,000 hires and no job
separations, or it may mean 500,000 hires and
490,000 separations. Clearly, the nature of eco­
nomic forces underlying these two scenarios
may be radically different. The first portrays an
economy with stable, perhaps rigid, labor mar­
kets, while the second conveys a picture with
much more activity. The U.S. economy turns
out to be much more like the second scenario,
with surprisingly high levels of job destruction
and creation, particularly during recession and
recovery periods, respectively.
This article introduces the ideas behind meas­
urement of gross labor market flows, presents
several such measures, including a new one,
and outlines some of the ways these data may
influence economists’ views of macroeconomic
events. The article first examines three sources
of information on gross labor market flows.
These are (1) establishment-level data assembled
by Steven Davis and John Haltiwanger from the
Census Bureau’s Survey of Manufactures, (2)
industry employment data from the Bureau of
Labor Statistics’ (BLS) Current Employment
Statistics (CES) program (often called the estab­



lishment survey, though information on in­
dividual establishments is not available from this
source) and (3) household data extracted from
the Current Population Survey (CPS). To pre­
vent confusion with the establishment-level data
from the Survey of Manufactures, the second
data source (CES) will subsequently be termed
industry data.

decreasing employment, respectively, divided by
total employment in sample establishments:

JC, =

lk r ii

After describing the gross flow data, the arti­
cle turns to a discussion of ambiguities that can
arise because of the interval between surveys
or the choice of measurement unit (household,
establishment or industry). The last two sections
in the article note differences and similarities
among the different gross flow measures and
some implications of looking at labor market
data in this way, especially the hypothesis that
business cycles are driven by sectoral shifts.

Davis and Haltiwanger (1990, 1992) have as­
sembled and analyzed gross flow data from the
Annual Survey of Manufactures undertaken by
the Census Bureau. In 1977 the Survey of
Manufactures covered approximately 19 percent
of manufacturing establishments (including all
establishments above a certain size) and 76 per­
cent of manufacturing employment.1
Davis and Haltiwanger’s series for gross job
creation and destruction rates are defined as
the sum of the absolute values of employment
changes in establishments with increasing and
1Davis and Haltiwanger (1990), p. 128. An establishment is
defined as a single physical location. Thus, one firm may
comprise several establishments.
2To compensate for the stratified sampling design, establish­
ments are weighted by the inverse of their sampling proba­
bilities. For a description of how births and deaths of
establishments are handled, see Davis and Haltiwanger




The Survey of Manufactures and industry
data look at gross flows from the standpoint of
employers, that is, from the demand side of the
market. Measured gross job creation is the sum
of increases in employment at those firms/
industries that experience increases. Measured
job destruction instead sums decreases. The
household data measure gross flows from the
supply side o f the labor market as the sum of
individuals’ movements into employment (gross
job finding) and the sum o f their movements
out of employment (gross job separation).


i= 1

JD , =

where E f is total employment in sample estab­
lishments, E it is employment in establishment i,
N t is the number o f establishments in the sam­
ple, 6jt,+) = 1 ifAE.( > 0 and 0 otherwise, and
d/-' = J if AE it < 0 a n d 0 otherwise.2
The series for job creation and destruction as
calculated by Davis and Haltiwanger are shown
in Figure l . 3 Davis and Haltiwanger draw atten­
tion to several features of these time series.
First, the magnitude of job creation and destruc­
tion is dramatic. Job creation and destruction
average 5.4 percent and 5.6 percent, respective­
ly, at a q u a rte rly rate over the 1973-86 period.
Second, there is a clear negative correlation be­
tween creation and destruction during reces­
sions. Third, job destruction accounts for much
more of the movement in employment during
recessions than does job creation. Fourth, at no
time is either job creation or job destruction
near zero; simultaneous creation and destruc­
tion is the rule without exception.

Limitations o f the Manufacturing
Establishment Data
Though establishment-level data have impor­
tant advantages for measuring gross flows, this
data source also suffers from serious limitations.
The most obvious is that it is restricted to
manufacturing, which accounted for only about
17 percent of employment in 1992 (down from
26 percent at the start of the Davis and Hal­
tiwanger sample in 1972). Second, these data
are available only with a substantial lag, and the
raw data are not publicly available.
In principle an establishment could incorrectly
report employment levels in a quarter, thus
generating spurious job creation or destruction
3These series are updated versions of the POS and NEG
series used in Davis and Haltiwanger (1990). The data
were kindly provided by John Haltiwanger.


Figure 1
Rates of Gross Job Creation and Destruction in
Manufacturing (Davis and Haltiwanger data)
Seasonally adjusted


0 .08-

0.0 6 -

0.0 4 -

0.02 i----- 1 ---- 1 ---- 1 ---- 1 ---- 1 ---- 1 ---- 1 ---- 1 ---- 1 ---- 1 ---- 1 ---- 1 ---- r
1972 73 74 75 76 77 78 79 80 81 82 83 84 85 1986
Shaded areas represent recessions.
(or both if the error were subsequently correct­
ed). This kind of classification error plagues the
household data but seems unlikely to be a seri­
ous problem here, though there is no evidence
available on the question.

A second approach to measuring gross job
creation and destruction, developed for this arti­
cle, is similar to Davis and Haltiwanger's, but
uses a breakdown of employment by industry
based on the monthly BLS Current Employment
Statistics (CES) survey. While there are disad­
vantages to basing gross flow measures on
industry-level data (particularly the netting of
job creation and destruction within industries),
this approach offers several significant advan­
tages: (1) industry coverage can be quite com­
prehensive; (2) the data are publicly available;
and (3) the data are available monthly without a
major publication lag.

The raw data are employment levels in sever­
al hundred industries in the private nonfarm
sector of the economy. The CES sample current­
ly covers more than 370,000 establishments, in­
cluding all firms with more than 250 employees
and a subset of smaller firms.4 These data are
benchmarked annually using yet more compre­
hensive information. The CES sample excludes
agricultural workers, unpaid family workers,
domestic workers in private homes, and selfemployed persons (all of whom are included in
the household data described in the next sec­
tion). To focus on job creation and destruction
driven primarily by market forces, the data
used for this paper also exclude government
workers, though the CES sample includes them.
People who hold jobs at more than one estab­
lishment will be counted more than once.
Though the data are collected from individual
establishments, only industry totals are publicly
In a month t when there is no change in
the industrial classification (most months), the

4A detailed description of the CES program can be found in
the BLS Handbook of Methods. Each issue of Employment
and Earnings contains an abbreviated description in the
“ Explanatory Notes” section.



gross job creation and destruction rates are
defined analogously to Davis and Haltiwanger's

JC =~T E

i =1

JD, = ~ t L ^ (-W,


where £, is total employment in these industries
and Ejt is employment in industry i. The con­
struction of job creation and destruction series
using CES data is complicated by the evolving
classification of industries. At various times the
standard industrial classification (SIC) used by
BLS to allocate employment among industries is
revised. In general, the revision results in a
finer breakdown of industries already included,
but sometimes it adds coverage of entirely new
industries.5 The job creation and destruction
series are constructed so that the breadth of in­
dustrial coverage does not change from the first
period to the last. A finer breakdown within a
larger industry is exploited, however. An adjust­
ment at the "birth" of a new three- or four-digit
industry accounts for the fact that the start of
data on the industry does not indicate job crea­
tion, but reclassification. Since new three- and
four-digit industries are generally created to
subdivide growing industries, this procedure
tends to limit the extent of netting of job crea­
tion and destruction within industries. The
procedure followed in periods when a finer
breakdown of an industry appears in the data is
described in the appendix.
Figure 2 shows rates of job creation and des­
truction using a base of two-digit industries for
which data are available since 1947. Almost all
of these are manufacturing industries, so this
series is dominated by manufacturing, which
has been a declining share o f total employment
for several decades. Since the breadth of indus­
trial coverage increases substantially in 1958
and 1972, Figure 3 shows results of the same
5For example, starting with 1958 data, the industrial
machinery and equipment (SIC 35) category is broken
down into engines and turbines (SIC 351), construction and
related machinery (SIC 353) and so on. In 1972, industry
353 was itself subdivided into construction machinery (SIC
3531) and mining machinery (SIC 3532). Also in 1972, the
remainder of industry 35 was further subdivided by addi­
tion of farm and garden machinery (SIC 352), general in­
dustrial machinery (SIC 356), and miscellaneous industrial
and commercial machinery (SIC 359).


calculations on industries for which data are
available in 1972. Neither Figure 2 nor Figure 3
is affected in a substantial way by excluding
government from the base of industries. The
data plotted in Figures 2 and 3 are seasonally
adjusted using the X -ll procedure and further
smoothed using a five-month centered moving
The industries in the 1972 base are a compre­
hensive cross-section of the nonfarm business
sector. In January 1972, employment was 59.2
million for all private nonfarm payrolls, only
23.5 million (39.7 percent of the total) for the
1947 base of industries, but 57.8 million (97.6
percent) for the 1972 base. By June 1993, total
employment was 91.3 million for all private
nonfarm payrolls, 23.3 million (25.5 percent) in
the 1947 base and 87.3 million (95.6 percent)
for the 1972 base.
One notable aspect of Figures 2 and 3 is that
job creation rates are substantially higher dur­
ing the 1980s using the 1972 base than using
the 1947 base, whereas job destruction rates do
not differ much between the two bases. This is
largely because many of the industries that are
excluded from the 1947 base are those which
experienced rapid growth during the 1980s
relative to other industries. Most segments of
construction, transportation, communications,
utilities, trade, insurance, real estate and serv­
ices (including medical) that are included in the
1972 base are not in the 1947 base.
These gross flow measures based on industry
data also show a pronounced cyclical pattern.
Job destruction still dominates cyclical move­
ments in total employment, though creation ap­
pears more cyclical in the industry measures
than in the Davis and Haltiwanger measure.
The most recent recession was marked by un­
usually small changes in job creation and des­
truction rates (see Figure 3). The job destruction
rate rose and fell, but by far less than in recent
recessions. This surprising fact is discussed
more extensively later in this article. The job


Figure 2
Rates of Gross Job Creation and Destruction, 1947 Base
Seasonally adjusted, centered 5-month moving average

0.025 -



0 .0 1 5 -



T T y T T ^ T T y T T y T T | I I | I I | I I "| I I j I I | I

1947 50 53 56 59 62 65 68 71
Shaded areas represent recessions.

74 77

I I | Ty

80 83 86 89 1992

Figure 3
Rates of Gross Job Creation and Destruction, 1972 Base
Seasonally adjusted, centered 5-month moving average

0.0180 .0 1 6 0 .0 1 4 0.0120.010-

0 .0 0 8 0 .0 0 6 0 .0 0 4 0. 0 0 2 -


























Shaded areas represent recessions.



Table 1
Labor Force Transitions in the CPS,
May 1993 (percent of adult population,
seasonally adjusted*)______________
Current Month
Last E
Month U




* Excluding population inflows and outflows.

creation rate showed no sharp increase around
the time of the trough as it did in previous
recessions. Shortly before the trough of the
1990-91 recession the job creation rate reached
its lowest level since the 1950s. The earlier low
points w ere the result of sharp downward
swings, however, while the recent low resulted
from a small downward swing that followed a
decade-long downward trend in the rate. In ad­
dition, job creation and destruction remained
close together following the March 1991 trough,
illustrating the slow growth of employment af­
ter the recession.

Limitations o f Industry-Based Data
The most compelling problem with this ap­
proach to gross flows is the large measurement
unit (an industry). To the extent that some
firms within an industry increase employment
during the same month that other firms
decrease employment, we get net rather than
gross flows. Obviously, this is far more likely to
be a serious problem when the measurement
unit is an entire industry rather than a single
establishment or household. As industries grow,
this problem becomes more severe. This effect
is largely offset, however, as industry detail in­
creases over time. A more extensive discussion
o f the netting issue is deferred until a later
In principle, these data, like the Davis and
Haltiwanger data, are subject to classification er­
6Uchitelie (1993) and Kreisler (1993) describe a recent inci­
dent that illustrates the vulnerability of data from establish­
ments to reporting errors.
7U.S. Department of Labor, Bureau of Labor Statistics, BLS
Handbook of Methods, Bulletin 2414, September 1992, pp.
5-6. The actual proportion of households will be somewhat
smaller than 75 percent for reasons discussed later.


rors. An establishment could incorrectly report
employment levels in one month, thus generat­
ing spurious job creation or destruction (or both
if the error were subsequently corrected). If
these errors are not correlated within an indus­
try, they may cancel out, but there is no evi­
dence available on this question.6

Each month the Current Population Survey
(CPS) collects employment data from a sample
of about 60,000 households, obtaining informa­
tion on about 113,000 persons 16 years of age
or older (about 0.6 percent of this population).
The survey attempts to establish whether each
member of the household was employed (E), un­
employed (U), or not in the labor force (N) dur­
ing the previous week. Though there are some
refinements to deal with special situations,
broadly speaking an individual who worked
during the survey week is counted as employed,
and one who did not work but was actively
looking for work is counted as unemployed.
Otherwise the individual is not in the labor
Each household is in the sample for a total of
eight months in two separated segments of four
consecutive months. The households are divided
into overlapping rotation groups so that about
75 percent of the households are the same in
adjacent months.7 These continuing households
make it possible to track changes in the labor
market status of many individuals. The infor­
mation from each household is weighted to
produce estimates of economy-wide flows.8
The April 1993 to May 1993 flows among E, U
and N and the relative sizes of the E, U and N
pools are shown in Table 1 and Figure 4. The
relative magnitudes of these flows are fairly
typical. As Table 1 indicates, most of the adult
population either stays employed (59.1 percent)
or out of the labor force (31.5 percent) from
month to month. The E to N (3.2 million) and N
to E (3.0 million) flows shown in Figure 4 are
the largest in absolute magnitude, but the U to
E (2.0 million) and U to N (1.5 million) flows are
8The number of individuals unemployed five weeks or less
and initial unemployment claims have also been used as
crude measures of the gross flow of workers into unem­
ployment. The former would measure movements from
both employment and out of the labor force. The latter is
tied to movements from employment to unemployment.


Figure 4
Gross Flows in the Current Population Survey, May 1993

Millions, seasonally adjusted

N=Not in the Labor Force

Status of Working Age Population, May 1993
Millions, seasonally adjusted



much larger in relation to the size o f the group
from which they are drawn. Table 1 illustrates
this: More than two-thirds as many people left
unemployment (1.0 + 0.8 = 1.8 percent) as re­
mained unemployed (2.6 percent). On the other
hand, less than 5 percent as many people left
employment as stayed employed (2.5 percent
compared with 59.1 percent).
Figure 5 shows gross job finding (the sum of
U to E and N to E movements) and gross job
separation (the sum of E to U and E to N), as a
proportion o f total employment, from mid-1968
to mid-1993.9 The terms job finding and job
separation will be used throughout in connec­
tion with the household data to emphasize that
these data are based on worker movements
rather than the creation or destruction of
specific jobs. Though job creation and finding
are closely bound together, as are job destruc­
tion and separation, the measured gross flows
are based on fundamentally different approaches.
The data in Figure 5 are seasonally adjusted
using the X -ll procedure and are further
smoothed using a five-month centered moving
These data show a striking cyclical pattern
similar to the demand-side measures. The net
drop in employment during recessions (the
usual way of viewing employment) is clearly
dominated by job separations, just as job des­
truction dominates in the establishment and in­
dustry gross flow measures. In four o f the five
recessions shown, job finding actually increases
during the recession. A second prominent fea­
ture of Figure 5 is the downward trend in gross
job finding and separation rates that starts
around 1984. This may be accounted for by
changes in the demographic structure of the
working-age population but there are difficulties
with this interpretation. Further discussion is
deferred until the end of the article.

Limitations o f CPS Gross Flows
Several serious problems with the CPS gross
flow data have limited their usefulness. The
least serious is sampling error. Even though the
CPS sample is quite large, the number of transi­
tions among states is relatively small; most of
those reported as E/U/N this month will be iden­
9Various problems with the CPS gross flow data greatly dis­
tort the relationship between job finding and job separation
rates in Figure 5. These problems are discussed extensive­
ly in the next subsection.


tically reported next month (see Table 1). This
means that the standard error around this esti­
mate of the true number of people changing
status will be large in proportion to the num­
ber. The sampling error, while comparatively
large, is zero on average, so it does not bias the
estimated flows.
The second problem is missing observations.
The sampling unit for the CPS is actually a resi­
dence rather than a household; the interviewers
return to the same a d d re ss for four consecutive
months. If the household moves, it drops from
the sample and is replaced by the household liv­
ing at that address, if any. If an adult moves
into or out of the household, the individual ap­
pears in or disappears from the sample. About
7.5 percent of individuals in particular resi­
dences in the previous month cannot be found
in data for the current month. In addition,
about 7.5 percent of individuals in particular
residences in the current month’s data w ere not
recorded in the previous month's data.1 In­
dividuals who move are probably more likely to
change labor force status than those who do
not. This would bias the gross flows downward.
Abowd and Zellner estimated gross flows cor­
rected for nonrandom missing data and found
that corrected flows into employment w ere 22
percent higher and corrected flows out of em­
ployment were 16 percent higher than unadjust­
ed flows.1
The third problem, classification error, has
generated the most attention. If an employed in­
dividual is classified correctly in month 1, incor­
rectly as unemployed in month 2, and correctly
in month 3, with no change in true status, two
spurious transitions (E to U and U to E) have
been recorded. These response errors arise
partly because o f the design o f the survey. One
individual from each household answers ques­
tions about every adult in the household, but
this is not necessarily the same individual each
month. Different respondents may answer ques­
tions about the labor force status of household
members in different ways. In addition, there is
some ambiguity about where the lines are
drawn between employed, unemployed and not
in the labor force. The line between unem­
ployed and not in the labor force is particularly
fuzzy (though not relevant for Figure 5).
10Abowd and Zellner (1985), p. 254.
1'Abowd and Zellner, Table 3, p. 264.


Figure 5
Rates of Gross Job Finding and Separation from the
Current Population Survey
Seasonally adjusted, centered 5-month moving average





1967 69 71 73 75 77 79
Shaded areas represent recessions.






91 1993

Because actual transitions are relatively rare,
a small probability of classification error can
generate errors in the gross flow data that are
quite large in proportion to the true flows. In
January 1993, for example, the data indicate
that 61.5 percent of the adult population was
employed. If 2.3 percent of employed workers
incorrectly report their status as not employed
(the estimate reported by Poterba and Summers
in table 3), then 1.41 percent of the sample
report a spurious transition out of employ­
ment.1 Only about 3 percent of the sample ac­
tually reports such a transition in January 1993,
so 46 percent (1.41/3) of the gross flow out of
employment would be spurious under this

bias in gross flow data using reinterview data
collected by BLS as part of a quality control
program. A small fraction of the original sample
is surveyed a second time by experienced per­
sonnel, most of whom are asked to try to
reconcile differences between the first and se­
cond interviews. Reinterview data are assumed
to be correct and are used to estimate the prob­
abilities of classification error for different
demographic groups. These estimates are then
used to correct the gross flow data. Poterba and
Summers (1986) adjust both job finding and job
separation downward by more than 60 percent
for the 1977-82 period. Abowd and Zellner
(1985) adjust the same flows downward by
more than 25 percent.

There have been several attempts, including
those by Abowd and Zellner (1985) and Poterba
and Summers (1986), to correct this upward

This approach is not wholly satisfactory,
however, for at least two reasons. First, two
studies taking similar approaches to the problem

12This example assumes that the worker’s status was
correctly recorded in the previous month. Using that infor­
mation, if it were available, would lower the error rate.
However, it also ignores the offsetting possibility that
people who were incorrectly recorded as employed last
month are now correctly recorded, thus generating a
further spurious flow out of employment.



Figure 6
Change in Household Employment:
Actual vs. Gross Flows
Seasonally adjusted, centered 5-month moving average



1 i i i i i i i i i i i i i i i i i i i i i i i i r
1967 69





come up with adjustments that differ widely,
and neither is clearly superior. Second, the reinterview program does not sample randomly
(from the entire CPS sample), but rather con­
centrates attention on interviews that are prone
to error (for example, because the original inter­
viewer is new to the job). This implies that the
reinterview data are likely to exaggerate the ex­
tent of classification error.
An idea of the overall seriousness of the
problems with CPS gross flow data can be
gleaned from Figure 6, which compares the
change in household employment estimated in
the usual way with the difference between
flows into and out of employment. The former
is based on the change in the number of people
in the CPS who report that they are employed,
whereas the latter is based only on the responses
of people who were surveyed in consecutive
months and reported a change in employment
status. In principle, the two should match quite
closely, but the gross flows substantially under­
state employment growth over the entire sam­
ple period. This situation may improve
significantly when a revised CPS is implemented
in January 1994.








91 1993

Because the subject o f this article is g ro s s em­
ployment flows, an important concern is the ex­
tent to which any particular measure of gross
flows really measures gross rather than net
flows. This issue arises on both the time and
cross-section dimensions of the data.
Netting occurs intertemporally if, for example,
an individual reports working in two consecutive
surveys but was unemployed between the two
reports. Similarly, a firm may have laid o ff and
rehired workers from one quarterly report to the
next. Obviously, this intertemporal netting will
be more important the longer the interval be­
tween observations on a measurement unit. The
household and industry measures are based on
monthly data, whereas the Davis and Haltiwanger
data are based on quarterly information.
Netting also occurs in the cross-section dimen­
sion when the measurement unit is larger than
a single worker. A firm may hire some workers
and fire others within the observation interval.
An industry may be a mix of firms that are


increasing and firms that are decreasing employ­
ment (as well as firms that do both within the
Finally, since the degree o f industry detail
differs between sectors, with manufacturing
employment by far the most finely subdivided,
the degree of intraindustry netting almost cer­
tainly varies systematically across sectors.
Since there is an additional layer o f netting in­
volved in using industry gross flow data, it is
important to understand the relationship be­
tween establishment- and industry-level varia­
tion in employment. Figure 7 compares the
Davis and Haltiwanger series on job creation
and destruction in manufacturing (labeled e sta b ­
lis h m e n t data) with series based on industry em­
ployment data for manufacturing industries and
using employment changes between the months
of the Survey of Manufactures (February, May,
August and November). These series differ
mainly because they use different units o f meas­
urement (establishment vs. industry), although
they are also based on different survey metho­
dology. Two features of the comparison stand
out. First, the profiles of the series are quite
similar; the larger peaks and troughs coincide
and have roughly the same size in both series,
though the similarity is less apparent for job
creation. The same is true for seasonally unad­
justed data (not shown). Second, the Davis and
Haltiwanger series are substantially higher than
the industry series. A large share of this gap is
gross job creation or destruction that nets out
when the unit of measurement is the industry.
A closer look at the gap reveals that it has no
pronounced trend and is not noticeably cyclical,
suggesting that most of the job creation and
destruction that disappears in this way using
the industry series may not be of great interest
from a macroeconomic perspective.

Though the magnitudes of the three gross
flow measures differ for many reasons, some
informative and some spurious, all three meas­
ures are strikingly high. Davis and Haltiwanger’s
manufacturing establishment data show, for ex­
13Adding together monthly job finding and separation rates to
get quarterly rates (as was done here) is somewhat mis­
leading since it is likely that many of the same people are
moving repeatedly into and out of employment.

ample, 5 and 7 percent rates of creation and
destruction for the first quarter of 1986. The
industry-based (1972 base) data indicate job cre­
ation and destruction rates o f about 2 percent
(with or without government) for the same
quarter. The household data indicate that job
finding and separation rates were about 14 per­
cent each for the same quarter.1 In addition,
there are dramatic seasonal swings in these ser­
ies (see the next section), so during the year the
rates can be much higher than the average.
Even the smallest of these magnitudes implies a
labor market in which a great deal of activity
takes place even when overall employment is
not changing. There is evidence that European
countries experience gross flows that are the
same order of magnitude.1
The household data (Figure 5) show a sharp
downward trend in both job finding and separa­
tion rates, starting around 1984. Total job find­
ing and separation levels (not shown) have a
clear upward trend to this point and no appar­
ent trend afterwards. There is a plausible demo­
graphic explanation for the downward trend in
the household data: Workers in their 30s, 40s
and 50s have lower rates of job separation and
job finding. The baby-boom generation started
to enter these years of stable labor force partic­
ipation and job attachment in the early 1980s,
and this increase in the proportion of workers
with lower finding and separation rates would
therefore depress the overall levels of gross job
finding and separation.
The puzzle remains, however. This demo­
graphic hypothesis should also apply to the in­
dustry data because private payroll employment
(from the CES) and household employment
(from the CPS) show similar trends over time.
The 1972 industry base includes almost all pri­
vate payroll employment. However, the down­
ward trend in job creation (industry data, 1972
base) is weaker than that in job finding (CPS
data) and there is no downward trend in job
destruction comparable to that in job separa­
tions. The main coverage differences between
the CPS and CES employment series are (1) the
CES excludes agricultural workers, self-employed
workers and several smaller categories and (2)
the CES data will record individuals with more
than one job in the nonfarm payroll sector
14Burda and Wyplosz (1990).



Figure 7
Rates of Job Creation in Manufacturing
Seasonally adjusted

Rates of Job Destruction in Manufacturing


Seasonally adjusted


more than once. Neither of these seems likely to
account for such a large difference in trends
during the 1980s. The industry job creation and
destruction data shown in Figures 2 and 3 do
not include government workers, but the trends
do not change when government workers are

Seasonal M ovem ents
Not surprisingly, there are extremely pro­
nounced seasonal patterns in the gross flow
measures. The 1992 seasonal patterns estimated
by the X -ll seasonal adjustment procedure for
the household and industry series (1972 base)
are shown in Figure 8. These are seasonal fac­
tors for levels of creation and destruction (the
data are not divided by total employment). They
are the ratio of the unadjusted series to the
seasonally adjusted series.
The industry data for job destruction show a
dramatic seasonal peak in January, when the ra­
tio of unadjusted to adjusted data is nearly 4.
The ratio of the highest to lowest seasonal fac­
tors for job destruction is nearly IS, meaning
that about 15 times as many jobs are destroyed
in January on average as in April. The ratio is
about 7.5 for job creation, with the high and
low months being June and January.
Seasonal movements in job finding and separa­
tion are also quite significant, though the scaling
of Figure 8 hides this. Seasonal factors in 1992
range from 0.8 to 1.2 for job finding levels and
from 0.8 to 1.3 for job separations.1 By con­
trast, seasonal factors for total civilian employ­
ment are only about one-tenth of this size (0.98
to 1.02).
Figure 8 highlights the fact that the seasonal
fluctuations in the CPS gross flows data have
much smaller amplitude than those in the indus­
try data. The different underlying data sources
are one reason for this. A worker who moves
immediately from one seasonal job to another is
15Seasonal factors (1986) for the Davis and Haltiwanger
manufacturing series range from 0.9 to 1.3 for destruction
and from 0.98 to 1.04 for creation. Part of the reason
seasonal factors are smaller for the Davis and Haltiwanger
series is that they are quarterly, so some of the seasonal
fluctuations have already been smoothed out.

not unemployed between the jobs, but this
movement can correspond to seasonal job des­
truction in the first industry and seasonal job
creation in the second. Similarly, if a seasonal
job is a second job for an individual, there is no
change in his or her labor market status as
recorded by the CPS when the job begins or
ends. No gross flow is generated in the CPS
data in either case, but a job is recorded as
both created and destroyed in data based on
employers’ payroll records. These are relatively
infrequent occurrences, but because labor mar­
ket transitions are also relatively rare, their
relative importance is much greater in measur­
ing gross flows than in measuring total em­

Cyclical M ovem ents
One common feature of all three approaches
to measuring gross flows is that employment
declines during recessions are dominated by
rises in job destruction or separation. Job crea­
tion or finding rates usually begin to decline
well before the business cycle peak. Similarly,
job destruction or separation rates tend to begin
rising before the official onset of a recession.
The timing o f the business cycle peaks in job
destruction and troughs in job creation is in­
teresting in two respects. First, the two usually
almost coincide in both the industry and estab­
lishment series.1 This does not occur in the
CPS data. Second, the peak of job destruc­
tion/separation tends to occur toward the
trough of the recession and never occurs at
the peak.
The household and industry data both indi­
cate that the 1990-91 recession was character­
ized by much smaller movements in gross
destruction/separation and creation/finding rates
than earlier recessions. This suggests that highly
visible downsizing efforts by firms are some­
what misleading for the economy as a whole;
in e( then shifts the distribution of gjt to the left, throwing
some firms/industries from the job creation column to the job
destruction column. Job creation falls and job destruction

16This regularity is even more pronounced in unsmoothed
data. A very simple model of job creation would predict
that creation and destruction should be (distorted) mirror
images of each other. Suppose that the change in employ­
ment for a particular firm or industry is given by git = e( +
u(, where e( is an aggregate shock and u, is a firm/industry
specific shock that does not depend on time. A decrease



Figure 8
1992 Seasonal Factors (NSA/SA) for Job Finding
and Creation (1972 Base)

1992 Seasonal Factors (NSA/SA) for Job Separation
and Destruction (1972 Base)




despite its visibility, job destruction was at sur­
prisingly lo w levels during and after the 1990-91
recession. Tw o observations may help to recon­
cile perceptions with the job destruction statis­
tics. First, BLS data indicate that an unusually
large proportion of job losses during and after
the 1990-91 recession were permanent rather
than temporary layoffs (as reported by w or­
kers), increasing the perceived seriousness of
the job destruction. Second, though job destruc­
tion in manufacturing did not reach particularly
high levels during the recession, manufacturing
has continued to shed jobs (that is, job destruc­
tion exceeded job creation) in almost every
month since the end of the recession. Manufac­
turing layoffs tend to be quite visible.

Most macroeconomic analyses involving gross
labor market flows have tried to assess what is
often called the sectoral shift hypothesis. This
hypothesis focuses on changes in the distribu­
tion of demand among sectors of the economy
rather than on aggregate shocks. The macroeco­
nomics literature typically assumes that business
cycles are driven by aggregate shocks. Various
sources of aggregate shocks have been hypothe­
sized by macroeconomists—private expectations,
monetary policy, oil price increases and technol­
ogy shocks, to name a few. They share the
common feature that all firms and individuals in
the model are affected in relatively similar ways.
In a seminal paper, Lilien (1982) argued that
shocks to the d is trib u tio n of demand among
different sectors might account for a large por­
tion of the variation in the le ve l of economic ac­
tivity. Adverse shocks to demand in specific
industries could cause dislocation of workers
and other resources that would not flow
smoothly into more productive pursuits. The
adjustment period would be characterized by a
decline in economic activity generally, and an
increase in the unemployment rate in particu­
lar, so there would be a positive relationship be­
tween cross-sectional variation in industry
employment growth and the unemployment
rate. Lilien estimated the relationship between

the unemployment rate and a measure of the
cross-sectional dispersion of employment growth
in 11 broad industry groups. He found that
more than half of the variation in the overall
unemployment rate could be accounted for by
variations in the cross-sectional dispersion of
employment changes.1
Lilien’s results have not been regarded as con­
clusive. Abraham and Katz (1986) pointed out
that an increase in Lilien’s measure of the crosssectional dispersion o f employment growth
could be induced by aggregate shocks, given
plausible assumptions about industries’ trend
rates o f growth and cyclical sensitivities. They
argued that if the positive correlation between
Lilien’s dispersion measure and the unemploy­
ment rate were accounted for by sectoral shifts,
there would also be an increase in job vacancies
when the dispersion measure increased. Holding
the overall level of aggregate demand fixed,
some industries would be trying to hire as
others’ laid-off workers, causing a mismatch of
workers and jobs and an increasing vacancy
rate. If the business cycle is driven by aggregate
demand, however, the relationship between dis­
persion and vacancy rates would be negative.
They found a strong negative relationship, im­
plying that aggregate demand fluctuations are
the dominant source of variation in the unem­
ployment rate.1
The sum o f job creation and destruction ser­
ies such as those by Davis and Haltiwanger or
those produced using industry data can be used
as a cross-sectional dispersion measure. For the
Davis and Haltiwanger data:


SU M , = JC t + JD , = - ^ £ \ A E u\.

S U M , increases when there is more variation in

employment change in individual industries.
Both SUM , and an analogous measure based on
industry data move countercyclically since JD,
tends to rise more than JC' falls during
To get more insight into what might drive
changes in dispersion, equation (1) can be re­
written in terms of growth rates:

17Lilien (1982), p. 792.
18Abraham and Katz used a normalized help-wanted index to
proxy for a direct measure of vacancies in the United
States. They found similar results in British data using a
direct measure of vacancies.



but its effects clearly differed from one firm to
another. Some firms disappeared altogether.
where g u is the growth rate of employment in
establishment i. The growth rate of establish­
ment i can be written as follows:

S„ = & + g,s + S,>

where g( is the average growth rate for all
manufacturing establishments, g* is the average
growth rate for industry or sector s minus g,
and g lt is the residual, or idiosyncratic, growth
of this establishment.
Davis and Haltiwanger find that nearly all of
the variation in S U M l over time can be at­
tributed to the g it component. For example,
when the sectors indexed by s correspond to
two-digit industries (a relatively broad industrial
classification), g:t accounts for 87.6 percent of
the variation in S U M r Replacing g.( with gi( in (2)
gives a dispersion measure S U M t that has aver­
age aggregate and industry growth rates re­
moved. Davis and Haltiwanger find that S U M t is
also countercyclical. They conclude, "W e inter­
pret these variance ratio results as a decisive
rejection of the hypothesis that the normal
pattern of sectoral responses to aggregate fluc­
tuations can account for the significant time
variation in [S U M t]... The time variation in
IS U M J results overwhelmingly from time varia­
tion in the contribution of idiosyncratic effects.”1
This empirical observation does not necessari­
ly mean that shifts in industry demand, broadly
defined, are not the source of much of the vari­
ation in S U M t or S U M t, however. The variance
decomposition technique labels only variation
common to all establishments as industry varia­
tion. In other words, only if employment grows
at exactly the same rate in all establishments
will the variance decomposition attribute all var­
iation to industry shocks. Any assumption used
to distinguish industry shocks from establish­
ment shocks, however, will be to some extent
arbitrary, and this particular assumption may
not be the best way to think about industry
shocks. In the 1970s and 1980s, for example,
the U.S. steel industry shrank dramatically,
largely as a result of international competition.
This increasing international competition could
reasonably be interpreted as an industry shock,
19Davis and Haltiwanger (1990), p. 138.


Rather than assuming, as Davis and Hal­
tiwanger do, that all firms respond identically to
an industry shock, we could assume that the in­
dustry shock hits the weakest firms hardest.
This would imply that firms’ responses to indus­
try shocks are extremely heterogeneous—in
other words, that firms have idiosyncratic
responses to industry shocks rather than truly
idiosyncratic shocks. It is easy to demonstrate
that this assumption can dramatically change
the results of a variance decomposition exercise.
The new identifying assumption may be equally
arbitrary but illustrates the sensitivity of vari­
ance decomposition exercises to the assumptions
used to identify industry shocks. Another way
to state this conclusion is that the idiosyncratic
shock g jt as constructed by Davis and Hal­
tiwanger is not necessarily independent of the
fortunes o f the industry.
Though the best way to isolate responses to
industry or aggregate shocks is a topic that
deserves further study, it is certainly clear from
Davis and Haltiwanger’s work that firms’
responses to cyclical shocks vary dramatically
even within industries. Understanding the size
and sources of heterogeneity in firms’ employ­
ment responses may be critical to understand­
ing the role of business cycles in the economy.
Macroeconomic models that assume firms’
responses to shocks are homogeneous within
industries will not capture any of the possible
ramifications of this heterogeneity.
Dispersion (S U M t) is not the only aspect of
these data that should be of interest to macroeconomists and, in fact, may be considerably
less informative than its two halves, job creation
and destruction. An important puzzle is the
asymmetry between job creation and destruc­
tion in recessions, with changes in job destruc­
tion swamping those in job creation. Blanchard
and Diamond (1990) argue that standard text­
book models of entry and exit would predict
the opposite. Existing firms exit (destroy jobs)
only when they cannot cover variable costs, so
exit is relatively insensitive to economic condi­
tions. Potential entrants (job creators) must ex­
pect to cover total costs, including any fixed
costs of entry. This implies that job creation will
vary more than job destruction. Blanchard and
Diamond speculate that differences in the costs


o f hiring and firing workers may lead to bunch­
ing of job destruction during recessions. They
point out, however, that aggregate behavior is
often not analogous to microeconomic behavior
in these types of models, so they find the expla­
nation only partly persuasive.
Blanchard and Diamond also observe that,
since cyclical changes in employment are domi­
nated by job destruction, Schumpeterian the­
ories of business cycles seem to be ruled out.
These theories argue that booms are brought
about by waves of product innovations (com­
puters, for example) that produce new jobs,
whereas recessions occur when these waves re­
cede.2 This kind of theory implies that employ­
ment changes will be dominated by changes in
job creation.
Job creation and destruction data that are
based on a comprehensive cross-section of the
labor market, such as the industry-based series
constructed for this paper, could prove useful
in several areas. Although establishment-level
data are clearly closer to ideal than industry
data for the study of job creation and destruc­
tion, it is important to know what happens be­
yond manufacturing (where establishment data
are available), as the experience of the most re­
cent cycle indicates.
Comprehensive data on gross flows could also
provide insight into the employment outcomes
o f a free-trade agreement. Opponents of such
agreements argue that jobs will be lost. Propo­
nents argue that while jobs will be lost, there
will be a n e t gain in employment. Little is
known, however, about the patterns of job
gains and losses surrounding such agreements.
If there is a net employment gain, as most
economists would predict, are the correspond­
ing consequences for job creation and destruc­
tion of a comparable or greater order of
magnitude than the net gain? In other words,
how significant is the inevitable worker disloca­
tion relative to the net job gain? Although intra­
industry netting makes it impossible to disag­
gregate job creation and destruction very far
using industry-based data, it may be possible to
discern which broad sectors experience the
largest effects.

Gross flow data, for all their faults, provide a
perspective on the U.S. labor market that can­
not be obtained from any other source. This
paper studies three approaches to measuring
gross flows of workers and jobs, including a
new, broadly based measure based on detailed
industry employment data. Each of the meas­
ures is flawed in a different way, but an impor­
tant message comes through nevertheless: Both
seasonal and business cycle downturns are
dominated by increases in job destruction, not
by declines in job creation. This may have in­
teresting and important implications for macro­
economics, but analysis of gross job creation
and destruction is a relatively undeveloped area
of macroeconomics.
The data also point to a striking fact about
the most recent business cycle: Job destruction
during the downturn appears to have stayed at
very low levels compared with previous reces­
sions. Moreover, in contrast to previous recov­
eries, there was no surge in job creation
following the trough.

Abowd, John M., and Arnold Zellner. “ Estimating Gross Labor
Force Flows,” Journal of Business and Economic Statistics
(July 1985), pp. 254-83.
Abraham, Katharine G., and Lawrence F. Katz. “ Cyclical Un­
employment: Sectoral Shifts or Aggregate Disturbances?"
Journal of Political Economy (June 1986), pp. 507-22.
Blanchard, Olivier Jean, and Peter Diamond. “ The Cyclical
Behavior of the Gross Flows of U.S. Workers,” Brookings
Papers on Economic Activity (1990, No. 2), pp. 85-155.
Burda, Michael, and Charles Wyplosz. “ Gross Labor Market
Flows in Europe: Some Stylized Facts,” Centre for Econom­
ic Policy Research Discussion Paper No. 439 (August
Davis, Steven J., and John Haltiwanger. “ Gross Job Creation,
Gross Job Destruction, and Employment Reallocation,”
Quarterly Journal of Economics (August 1992), pp. 819-63.

20One recent example of this type of theory is Shleifer



_______ , a n d _______ . “ Gross Job Creation and Destruc­
tion: Microeconomic Evidence and Macroeconomic Implica­
tions,” NBER Macroeconomics Annual (1990) pp. 128-86.

Poterba, James M., and Lawrence H. Summers. “ Reporting
Errors and Labor Market Dynamics,” Econometrica (Novem­
ber 1986), pp. 1319-38.

Kreisler, Stephen. “ BLS Establishment Estimates Revised to
Incorporate March 1992 Benchmarks and Historical Correc­
tions,” Employment and Earnings (U.S. Department of
Labor, Bureau of Labor Statistics, June 1993), pp. 6-12.

Shleifer, Andrei. “ Implementation Cycles,” Journal of Political
Economy (December 1986), pp. 1163-90.

Lilien, David M. “ Sectoral Shifts and Cyclical Unemploy­
ment,” Journal of Political Economy (August 1982), pp.

U.S. Department of Labor, Bureau of Labor Statistics, BLS
Handbook of Methods, Bulletin 2414. U.S. Government
Printing Office, September 1992.

Uchitelle, Louis. “Job Loss in Recession: Scratch Those
Figures,” The New York Times, May 7, 1993.

Constructing Job Creation and Destruction Using
Industry Employment Data
This appendix describes fully the procedure
used to generate the job creation and destruc­
tion series described in the Industry Data sec­
tion. The raw data are employment levels for
606 industries and are not seasonally adjusted.
They are a mixture of two-, three- and fourdigit SIC industries with varying start dates. The
following steps detail the procedure.
1. For each two- or three-digit industry for
which three- or four-digit subindustries are de­
fined, a new re s id u a l in d u s try is defined by sub­
tracting employment in all of the subindustries
from the total. The original two- or three-digit
industry is then dropped from the data, leaving
a set of non-overlapping industries that still in­
clude all employment in the original set o f in­
dustries. In cases where four-digit industries
start before three-digit "parent” industries, the
two-digit residual is created by subtracting only
employment in the four-digit industry from em­
ployment in the two-digit industry until the
three-digit industry starts. After this point, both
the four-digit industry and three-digit residual
are subtracted from employment in the twodigit industry to get the two-digit residual. In
some cases, the subindustries partition the en­
tire industry, leaving no employment in the
residual industry. If data for a two-digit indus­
try start after the start date, the corresponding
residual industry is dropped (though some
three- or four-digit subindustries may be includ­
ed if their data go back to the start date).
2. A start date is chosen, say 1947 (all industries
start in January). All three-digit industries that
start after 1947 and are part of a two-digit in­
1Davis and Haltiwanger handle new establishments differ­
ently because the birth of an establishment really does cor­
respond to job creation.


dustry not in the 1947 data are dropped. If a
four-digit industry starts after 1947 and is part
of a three-digit industry that starts after 1947,
and the three-digit industry is part of a twodigit industry that starts after 1947, the fourdigit industry is dropped. Three- or four-digit
industries that start after 1947 but are part of a
two-digit industry that starts in or before 1947
are retained, however. These will be referred to
below as new industries and are all treated as
spin-offs of the appropriate residual industry.
In some cases four-digit subindustries start be­
fore their parent three-digit industries. In this
case, the three-digit residual industry is treated
as a new industry that spins o ff from the ap­
propriate two-digit residual and is considered
zero until its start date.
3. In months when no new industries start
(most months), job creation and destruction are
calculated by totaling employment in industries
where employment change is positive and nega­
tive, respectively.
4. In the starting month for a new industry, the
data show employment going from zero to some
positive number in the new industry and show
a drop of the same amount in the residual in­
dustry (plus the growth of the rest of the
residual industry). These changes are induced
by reclassification of jobs, not job creation or
destruction.1 Therefore, employment in the new
industry is added to employment in the residual
industry and the new industry is ignored in
creating job creation and destruction for that
month. With this proviso, the job creation and


destruction totals can be calculated by summing
employment in industries in which employment
change is positive and negative, respectively.

E x a m p le 2


5. The final series for gross job creation, gross
job destruction and total employment in the in­
dustries under consideration are separately
seasonally adjusted using X -ll.
The three following examples may help to
clarify the procedure.


Residual industries
SIC275R = SIC275 - SIC2752 - SIC2759

E x a m p le 1





SIC27R = SIC27 - SIC275R
- SIC2752 - SIC2759
- (other subindustries of SIC27)
would be created. SIC27 and SIC275 would be
dropped, and SIC275R = SIC275 until 1972. If
the start date is 1947, SIC275R is treated as a
new industry in 1958. SIC2752 and SIC2759 are
new industries in 1972.

The residual industry would be
E x a m p le 3





SIC17R = SIC17 - SIC171 - SIC172 - SIC173
- SIC174 - SIC175 - SIC176.

If the start date for Example 1 were 1947, all of
these industries would be dropped. If the start
date were 1972, they would all be included. If
the start date were 1958, SIC17R, SIC174 and
SIC175 would be dropped, but SIC171, SIC172,
SIC173 and SIC176 would be included.

If the start date is 1947 in this example, only
data from SIC413 would be used. If the start
date is 1958 or later, SIC41R and all of the
three-digit industries would be included.



“ The Government’s Role in Deposit Insurance.”
A collection of essays by David C. Wheelock,
Kevin Dowd, J. Huston McCulloch, Philip H.
Dybvig, Anjan Thakor and Mark D. Flood.
Steven Russell, ed.
R. Alton Gilbert, “ Implications of Annual Exami­
nations for the Bank Insurance Fund.”
Keith M. Carlson, “ On the Macroeconomics of
Private Debt.”

“ Dimensions in Monetary Policy; Essays in
Honor of Anatol B. Balbach.” Proceedings of the
Seventeenth Annual Economic Policy Confer­
ence. Michael T. Belongia, ed.
Robert H. Rasche, “ Monetary Aggregates, Mone­
tary Policy and Economic Activity.”
W. Lee Hoskins, “ Views on Monetary Policy.”
Harold Demsetz, “ Financial Regulation and the
Competitiveness of the Large U.S. Corporation.”
Carl F. Christ, “ Assisting Applied Econometric
Allan H. Meltzer, “ Real Exchange Rates: Some
Evidence from the Postwar Years.”
Michael D. Bordo, “ The Gold Standard, Bretton
Woods and Other Monetary Regimes: A Histori­
cal Appraisal.”

Michael J. Dueker, “ Can Nominal GDP Targeting
Rules Stabilize the Economy?”
Joseph A. Ritter, “ The FOMC in 1992: A Mone­
tary Conundrum.”

Terence C. Mills and Geoffrey E. Wood, “ Does
the Exchange Rate Regime Affect the
Patricia S. Pollard, “ Central Bank Independence
and Economic Performance.”
Michael J. Dueker, “ Hypothesis Testing with
Near-Unit Roots: The Case of Long-Run
Purchasing-Power Parity.”
Richard G. Anderson, “ The Effect of Mortgage
Refinancing on Money Demand and the Mone­
tary Aggregates.”

David C. Wheelock, “ Is the Banking Industry in
Decline? Recent Trends and Future Prospects
from a Historical Perspective.”
Michael J. Dueker, “ Indicators of Monetary Policy:
The View from Implicit Feedback Rules.”
Mathias Zuriinden, “ The Vulnerability of Pegged
Exchange Rates: The British Pound in the


(Commentaries by David A. Dickey, David Laidler, Manfred J.M. Neumann, Charles I. Plosser,
Georg Rich, Julio J. Rotemberg and Pedro

John A. Tatom, “ Is an Infrastructure Crisis
Lowering the Nation’s Productivity?”


Sangkyun Park, “ The Determinants of Consumer
Installment Credit.”

Gerald P. Dwyer, Jr., “ Rules and Discretion in
Monetary Policy.”

Joseph A. Ritter, “ Measuring Labor Market Dy­
namics: Gross Flows of Workers and Jobs.”


F ederal R es erve Bank o f St. Lou is
Post Office Box 442
St. Louis, Missouri 63166

The R e v ie w is published six
times p e r y ea r by the Research
and P u b lic In fo rm a tio n
D epartm ent o f the Federal
Reserve Rank o f St. Louis.
Sirtgle-cnpy subscriptions are
available to the p u b lic f r e e o f
charge. M ail requests f o r
subscriptions, ba ck issues, o r
address changes to: Research
and P u b lic In fo rm a tio n
Departm ent, Federal R eserve
Rank o f St. Louis, P.O. R ox 442,
St. Louis, M issou ri 63166.
The views expressed are those
o f the individual authors and do
n ot necessarily re fle c t o ffic ia l
positions o f the Federal R eserve
Rank o f St. Louis o r the Federal
Reserve System. A rticles herein
may be rep rin ted p rovid ed the
sou rce is credited. Please p ro v id e
the R ank’s Research and P u b lic
In fo rm a tio n Departm ent with a
co p y o f rep rin ted material.